Now, there are almost no more interpreted scripts.
On 17 Oct 2026 08:40:24 GMT, St|-phane CARPENTIER wrote:
Now, there are almost no more interpreted scripts.
Shell scripts are still interpreted. Not only that, but the shell
actually seeks through the source file to do a transfer of control.
How do I know this? Try making changes to the script while it is
running. ;)
On 2026-10-17 22:47, Lawrence DrCOOliveiro wrote:
On 17 Oct 2026 08:40:24 GMT, St|-phane CARPENTIER wrote:
Now, there are almost no more interpreted scripts.
Shell scripts are still interpreted. Not only that, but the shell
actually seeks through the source file to do a transfer of control.
How do I know this? Try making changes to the script while it is
running. ;)
I've just done this ("a script change while it's running") [*]
(with ksh and bash) and there was no change in behavior of the
running script.
You may want to provide an example what you've actually done
to come to that conclusion (or reformulate your conclusion).
BTW; I've in other (more complex?) contexts observed that some
intrusions (tackling the shell program while it runs) do affect
the running script, but back then I haven't examined the reason
or circumstances of that effect, or whether it's some effect of
an interaction with the OS. - So the principle question, what
exactly happens in what circumstances, is still interesting.
Janis
[*] What I've done was to run the following script "dynchg.sh"
while read ; do
echo $REPLY
done
and change it (from another terminal) while it runs by
sed -i '2p' dynchg.sh
I could even remove that file from the directory and it still
processes data as expected. And there's also no difference if
that program is explicitly run with the interpreter or running
it using a shebang interpreter as executable script.
Janis Papanagnou <janis_papanagnou+ng@hotmail.com> writes:
On 2026-10-17 22:47, Lawrence DrCOOliveiro wrote:
On 17 Oct 2026 08:40:24 GMT, St|-phane CARPENTIER wrote:
Now, there are almost no more interpreted scripts.
Shell scripts are still interpreted. Not only that, but the shell
actually seeks through the source file to do a transfer of control.
How do I know this? Try making changes to the script while it is
running. ;)
I've just done this ("a script change while it's running") [*]
(with ksh and bash) and there was no change in behavior of the
running script.
You may want to provide an example what you've actually done
to come to that conclusion (or reformulate your conclusion).
BTW; I've in other (more complex?) contexts observed that some
intrusions (tackling the shell program while it runs) do affect
the running script, but back then I haven't examined the reason
or circumstances of that effect, or whether it's some effect of
an interaction with the OS. - So the principle question, what
exactly happens in what circumstances, is still interesting.
Janis
[*] What I've done was to run the following script "dynchg.sh"
while read ; do
echo $REPLY
done
and change it (from another terminal) while it runs by
sed -i '2p' dynchg.sh
I could even remove that file from the directory and it still
processes data as expected. And there's also no difference if
that program is explicitly run with the interpreter or running
it using a shebang interpreter as executable script.
If your remove the script while it's still running, chances are the interpreter still has it open, so the data is not removed until the file
is closed---that is, you don't more free space until every reader and
writer have closed the file. That's the typical UNIX semantics of open files. (If I understood the problem here correctly.)
I have learned not to edit a script while it was running, weird things happened when I did. My theory was that bash kept an index pointing to
the current line. Dunno if as byte or as line.
On 2026-10-17 22:47, Lawrence DrCOOliveiro wrote:
I've just done this ("a script change while it's running") [*]
[*] What I've done was to run the following script "dynchg.sh"
while read ; do
echo $REPLY
done
and change it (from another terminal) while it runs by
sed -i '2p' dynchg.sh
GNU sed doesn't modify the original shell script, which hasbeen opened by the shell.
I could even remove that file from the directory and it still
processes data as expected.
Janis Papanagnou <janis_papanagnou+ng@hotmail.com>:
On 2026-10-17 22:47, Lawrence DrCOOliveiro wrote:
I've just done this ("a script change while it's running") [*]
[*] What I've done was to run the following script "dynchg.sh"
-a-a while read ; do
-a-a-a-a-a-a echo $REPLY
-a-a done
and change it (from another terminal) while it runs by
-a-a sed -i '2p' dynchg.sh
As the "-i" option is not defined by the POSIX standard but by GNU sed,
I assume that your example uses GNU sed.
[...]
On 2026-10-21 00:54, Helmut Waitzmann wrote:
Janis Papanagnou <janis_papanagnou+ng@hotmail.com>:
On 2026-10-17 22:47, Lawrence DrCOOliveiro wrote:
I've just done this ("a script change while it's running") [*]
[*] What I've done was to run the following script "dynchg.sh"
-a-a while read ; do
-a-a-a-a-a-a echo $REPLY
-a-a done
and change it (from another terminal) while it runs by
-a-a sed -i '2p' dynchg.sh
As the "-i" option is not defined by the POSIX standard but by GNU
sed, I assume that your example uses GNU sed.
Yes. (I've just used '-i' for simplicity.)
A better example would have been to issue (from another shell window)
the command
-a-a echo "echo END" >> dynchg.sh
and we then see that after terminating the above loop the "END" will
be printed.
(I'm testing with ksh but suppose that in bash it would be similar.)
-a KSH and BASH are "mostly compatible". Minor diffs but
-a nothing hard to get around. Anything you can do in KSH
-a can be done in BASH.
[...]
-a If KSH works for you then stick with it.
Note however that MOST distros default to BASH now -
so if you plan
-a to 'distribute' your work it'd be better to move your
-a stuff over to BASH.
Most distros, users would have to
-a intentionally install KSH and scripts would need a
-a shebang at the top to specify which shell to use.
-a (good practice, always add the shebang)
KSH and BASH are "mostly compatible". Minor diffs but
nothing hard to get around. Anything you can do in KSH
can be done in BASH.
On 2026-10-24 05:59, c186282 wrote:
KSH and BASH are "mostly compatible". Minor diffs but
nothing hard to get around. Anything you can do in KSH
can be done in BASH.
This is not correct. While Bash has over time adopted quite
some features from Ksh - but sadly needed often far too many
years for that - there's still a lot substantial functions
in Ksh that Bash has not implemented (or substituted by some
equivalent feature).
On 2026-10-24 05:59, c186282 wrote:
Note however that MOST distros default to BASH now -
Yes, Linux often has Bash as its default.
In article <11bhv61$1bjd2$2@dont-email.me>,
Janis Papanagnou <janis_papanagnou+ng@hotmail.com> wrote:
On 2026-10-24 05:59, c186282 wrote:
KSH and BASH are "mostly compatible". Minor diffs but
nothing hard to get around. Anything you can do in KSH
can be done in BASH.
This is not correct. While Bash has over time adopted quite
some features from Ksh - but sadly needed often far too many
years for that - there's still a lot substantial functions
in Ksh that Bash has not implemented (or substituted by some
equivalent feature).
I think you're taking Speedy way too seriously here. All of your points
are valid, of course, but it should not be necessary to respond to the
troll.
BTW, you assert that there are (many) features of Ksh that are not (yet) present in bash. I'm not doubting you, but I am curious about examples.
It seems to me that if you assert something like that, you should give examples.
Le 24-10-2026, Janis Papanagnou <janis_papanagnou+ng@hotmail.com> a |-crit-a:
On 2026-10-24 05:59, c186282 wrote:
Note however that MOST distros default to BASH now -
Yes, Linux often has Bash as its default.
I just keep this because I guess, if I don't consider it to be wrong
because I know what it exactly means, I believe it's often a source of confusion. Because it can mean two different things and three different things must be considered when speaking about the default shell.
Mostly, BASH is the default for the three things, one of them can be
changed effortlessly, one of them would be very difficult to change in
the long term and the last one depend on the interaction with others.
The first thing is the way by which you interact directly with your
SHELL, a kind of REPL. It's just a field in the /etc/passwd file, it's
easy to change and if you prefer any SHELL, just install it, change the
field and it rocks. It's just a matter of taste, nothing can go wrong by changing it.
The second thing is the shell used by the distro to run its script. This
one is clearly often BASH, changing it can be challenging. And things
can go wrong in a lot of ways, from the shebang poorly defined to some specificities expected by the distro. But really, I'm not sure it's a
real issue because I'm not sure there are reason to change it. The
fact that systemd doesn't need any shell script is a reason why it's nit
that important. One can want to change the default provided by the
distro, but changing the script shell provided by the distro mean each
time an update is done, the changes will be removed. And they will need special care. When there are ways to change the default without changing
the scripts provided by the distro.
The second thing is the shell used by the distro to run its script.
This one is clearly often BASH, changing it can be challenging. And
things can go wrong in a lot of ways, from the shebang poorly
defined to some specificities expected by the distro.
On 2026-10-24 05:59, c186282 wrote:
-a-a KSH and BASH are "mostly compatible". Minor diffs but
-a-a nothing hard to get around. Anything you can do in KSH
-a-a can be done in BASH.
This is not correct. While Bash has over time adopted quite
some features from Ksh - but sadly needed often far too many
years for that - there's still a lot substantial functions
in Ksh that Bash has not implemented (or substituted by some
equivalent feature).
[...]
-a-a If KSH works for you then stick with it.
I use Ksh meanwhile for ~35 years, and since I use a lot of
Ksh features that Bash just doesn't support there's certainly
no reason to switch. (Also its speed is a criterion, BTW. And
back then there were no Bash versions on commercial Unixes;
but commercial Linuxes haven't the prevalence any more today,
and play no role now in my private systems context.)
-a-a Note however that MOST distros default to BASH now -
Yes, Linux often has Bash as its default.
-a-a so if you plan
-a-a to 'distribute' your work it'd be better to move your
-a-a stuff over to BASH.
Certainly not.
And this is short-sighted. If _portability_ is an issue you'd
better stick with the POSIX subset of both, Ksh and Bash.
(Sadly, as we observe, especially "Linux kiddies" often don't
know what that POSIX subset actually is; in their scripts they
are writing #!/bin/sh and using some bash-specific syntax not
realizing that there's a difference, and that Bash is not the
[portable] *standard*.)
-a-a Most distros, users would have to
-a-a intentionally install KSH and scripts would need a
-a-a shebang at the top to specify which shell to use.
I don't recall to have ever "intentionally installed" Ksh on
Linux. Earlier Linuxes had just a (sub-optimal!) PD-ksh that
you could use out of the box. Later they provided also the
much better "original" [AT&T] Ksh, and meanwhile you find on
(some?) Linuxes ksh93u+m, Martijn Dekker's improved version.
Of course you can call shell scripts with the executable-flag
set and with a she-bang line on top. Personally I run most
shell scripts with the interpreter explicitly provided, as in
-a ksh my-script flags arguments
allowing easy debugging by adding '-x' without explicit 'set'
command in the script. Only if some script gets the status of
a "standard-command" I regularly use I set the shebang. YMMV.
(So you don't "need" a shebang line, but it's useful.)
-a-a (good practice, always add the shebang)
Yes, if you are running your scripts executable script (or if
you want to give your editor a hint for syntax highlighting,
what shell is actually used, or just for specification purpose;
but note, as said above, that we see many people have no clear
idea about what a "/bin/sh" is and where /bin/bash or /bin/ksh
differ from the standard.
On 2026-10-24, c186282 wrote:
KSH and BASH are "mostly compatible". Minor diffs but
nothing hard to get around. Anything you can do in KSH
can be done in BASH.
Now that sounds like something Greg Sullivan of Microsoft would say.
On 2026-10-24 14:10, Kenny McCormack wrote:
In article <11bhv61$1bjd2$2@dont-email.me>,
Janis Papanagnou-a <janis_papanagnou+ng@hotmail.com> wrote:
On 2026-10-24 05:59, c186282 wrote:
-a-a KSH and BASH are "mostly compatible". Minor diffs but
-a-a nothing hard to get around. Anything you can do in KSH
-a-a can be done in BASH.
This is not correct. While Bash has over time adopted quite
some features from Ksh - but sadly needed often far too many
years for that - there's still a lot substantial functions
in Ksh that Bash has not implemented (or substituted by some
equivalent feature).
I think you're taking Speedy way too seriously here.-a All of your points
are valid, of course, but it should not be necessary to respond to the
troll.
Oh, I didn't knew that he's considered a troll. His post looked just uninformed, or opinionated, not trollish.
BTW, you assert that there are (many) features of Ksh that are not (yet)
present in bash.-a I'm not doubting you, but I am curious about examples.
It seems to me that if you assert something like that, you should give
examples.
Mind that his post says "Minor diffs"; and I was not (not primarily) addressing the absolute number of features but those features that
are hard(er) or in practice impossible [based on existing functions]
to substitute. Off the top of my head I had been thinking of discipline functions, or the "type system" - these are very closely embedded in
the shell.
Note that while these features are probably not widely used there's
also many often used details I used, functions and options, that Bash
doesn't have as I noticed over the time. I've occasionally mentioned
those in some posts that where about such functionality, but I'm too
lazy to create a list or something, which would require to re-test
every features whether one or another has meanwhile been implemented;
that would be too much effort.[*])
Janis
[*] After finishing writing the post I remembered that we could ask
a search machine AI, but strangely it had no or found any substantial information, not even the obvious examples; so we have to rely on the
experts or own experience here.
In article <11bhv61$1bjd2$2@dont-email.me>,
Janis Papanagnou <janis_papanagnou+ng@hotmail.com> wrote:
On 2026-10-24 05:59, c186282 wrote:
KSH and BASH are "mostly compatible". Minor diffs but
nothing hard to get around. Anything you can do in KSH
can be done in BASH.
This is not correct. While Bash has over time adopted quite
some features from Ksh - but sadly needed often far too many
years for that - there's still a lot substantial functions
in Ksh that Bash has not implemented (or substituted by some
equivalent feature).
I think you're taking Speedy way too seriously here. All of your points
are valid, of course, but it should not be necessary to respond to the
troll.
BTW, you assert that there are (many) features of Ksh that are not (yet) present in bash. I'm not doubting you, but I am curious about examples.
It seems to me that if you assert something like that, you should give examples.
On 10/24/26 05:53, Janis Papanagnou wrote:
On 2026-10-24 05:59, c186282 wrote:
-a-a KSH and BASH are "mostly compatible". Minor diffs but
-a-a nothing hard to get around. Anything you can do in KSH
-a-a can be done in BASH.
This is not correct. While Bash has over time adopted quite
some features from Ksh - but sadly needed often far too many
years for that - there's still a lot substantial functions
in Ksh that Bash has not implemented (or substituted by some
equivalent feature).
[...]
-a-a If KSH works for you then stick with it.
I use Ksh meanwhile for ~35 years, and since I use a lot of
Ksh features that Bash just doesn't support there's certainly
no reason to switch. (Also its speed is a criterion, BTW. And
back then there were no Bash versions on commercial Unixes;
but commercial Linuxes haven't the prevalence any more today,
and play no role now in my private systems context.)
-a-a Note however that MOST distros default to BASH now -
Yes, Linux often has Bash as its default.
-a-a so if you plan
-a-a to 'distribute' your work it'd be better to move your
-a-a stuff over to BASH.
Certainly not.
And this is short-sighted. If _portability_ is an issue you'd
better stick with the POSIX subset of both, Ksh and Bash.
(Sadly, as we observe, especially "Linux kiddies" often don't
know what that POSIX subset actually is; in their scripts they
are writing #!/bin/sh and using some bash-specific syntax not
realizing that there's a difference, and that Bash is not the
[portable] *standard*.)
-a Delay does not necessarily equate with 'ignorance'.
-a Different developers ... everyone had their own idea
-a of a 'perfect shell' - but then had to adapt and adapt
-a again as the Real Needs became more obvious. Now, most
-a WILL do the same things - just slightly different syntax.
-a I rather liked the 'c-shell' long back because it, well,
-a 'looked like C'. It was one of the things that attracted
-a me to Unix/Linux in the first place. Now I'm happy (ok,
-a that's not really the word) with BASH. Not much it
-a can't do ... but the fudged-fudged-fudged syntax to add
-a features over time CAN be a pain in the ass.
-a If I want 'scripting' now, I do it with Python. BASH
-a or whatever, maybe it just EVOKES the Python scripts,
-a a one or two-liner. Wrote many rather long BASH scripts
-a in the past, and while they worked they were, well, just
-a TOO messy/cryptic.
-a Last big BASH script I did was for backing up a bunch
-a of on-network drives. Mount-checking (important), calls
-a to rsync and openSSH and ftp, local AND 'cloud', all
-a with some confirmation checks along the way. SEMI-repeating
-a units but each was a bit 'special'.
-a (ALWAYS pre-encrypt before sending to 'cloud' - I do
-a NOT trust the providers one bit. Mine had an extension
-a that 'suggested' some kind of ZIP, just to help them
-a waste time)
-a RE-did it in Python. MUCH shorter and more comprehensible
-a and a nice easy CSV config file you could tweak in a text
-a editor or spreadsheet. Python made it all BETTER.
-a Had a 'C' version of that backup pgm earlier. Worked
-a fine, but there was a lot of string checking/editing
-a involved and 'C' is CLUNKY with string shit.
Python
-a stole the FORTRAN string-slicing stuff, good deal !
-a Won't ever use BASH or any other 'shell script' for
-a that kind of task again. Those might be "trad -IX"
-a but IMHO they're just OVER. "Solutions" MOVED ON.
-a-a Most distros, users would have to
-a-a intentionally install KSH and scripts would need a
-a-a shebang at the top to specify which shell to use.
I don't recall to have ever "intentionally installed" Ksh on
Linux. Earlier Linuxes had just a (sub-optimal!) PD-ksh that
you could use out of the box. Later they provided also the
much better "original" [AT&T] Ksh, and meanwhile you find on
(some?) Linuxes ksh93u+m, Martijn Dekker's improved version.
-a Um ... these days, almost all distros, you do have
-a to "intentionally install" KSH or any rivals. BASH
-a rules by default.
Of course you can call shell scripts with the executable-flag
set and with a she-bang line on top. Personally I run most
shell scripts with the interpreter explicitly provided, as in
-a-a ksh my-script flags arguments
allowing easy debugging by adding '-x' without explicit 'set'
command in the script. Only if some script gets the status of
a "standard-command" I regularly use I set the shebang. YMMV.
-a Direct spec of the shell IS useful, however the
-a 'shebang' trick works well too.
(So you don't "need" a shebang line, but it's useful.)
-a-a (good practice, always add the shebang)
Yes, if you are running your scripts executable script (or if
you want to give your editor a hint for syntax highlighting,
what shell is actually used, or just for specification purpose;
but note, as said above, that we see many people have no clear
idea about what a "/bin/sh" is and where /bin/bash or /bin/ksh
differ from the standard.
-a I have, I think, FOUR shells on this machine. The
-a shebang IS required/desired.
-a Still kinda like 'tcsh', it's there in case I get
-a a brain-worm and wanna fool around with olde-dayz
-a stuff. Zsh ain't so bad either.
-a But, as said, for 'scripting' now, Python is by
-a far my #1 choice unless the script is like a
-a simple little ten-liner. Tend to use those on
-a my PIs.
-a Fooled around a bit with Lua and Ruby for scripting,
-a but, really, Python and the MASSIVE number of libs
-a have them beat. Python will do ANYTHING these days,
-a just not as fast as 'C'. However CPU speed is less
-a of a thing these days - I/O ops drag things down
-a more than the CPU - so Python and 'C' can usually
-a compete pretty well.
-a There are Python compilers too. The main issue there
-a is all the sub-sub-versions of Python, so you always
-a have to include the entire binaries into each compilation
-a to ensure it'll work. But then space is also a minor
-a issue now as well unless you're doing micro-controllers.
-a Did my share of those - DO teach you how to really SQUEEZE
-a shit in - all computer kiddies should learn MCs first.
-a . . .
-a ("Harry talks in his sleep, didn't you notice ?"
-a-a-a Slight telling delay, then Hermonie says 'No' :-)
On 10/24/26 12:53, Janis Papanagnou wrote:
On 2026-10-24 14:10, Kenny McCormack wrote:
In article <11bhv61$1bjd2$2@dont-email.me>,
Janis Papanagnou-a <janis_papanagnou+ng@hotmail.com> wrote:
On 2026-10-24 05:59, c186282 wrote:
-a-a KSH and BASH are "mostly compatible". Minor diffs but
-a-a nothing hard to get around. Anything you can do in KSH
-a-a can be done in BASH.
This is not correct. While Bash has over time adopted quite
some features from Ksh - but sadly needed often far too many
years for that - there's still a lot substantial functions
in Ksh that Bash has not implemented (or substituted by some
equivalent feature).
I think you're taking Speedy way too seriously here.-a All of your points >>> are valid, of course, but it should not be necessary to respond to the
troll.
Oh, I didn't knew that he's considered a troll. His post looked just
uninformed, or opinionated, not trollish.
-a Thanks, I think-a :-)
-a I never 'troll' - just relate MY experiences and what
-a I've learned from them. I had my own path, own needs,
-a my own 'environments', starting in the punch-card days.
-a But what served ME best may not be what served YOU, what
-a I learned may not be what YOU learned.
-a A great purpose of groups like these is to bring in
-a a LOT of experiences and perspectives. You MAY find
-a something compatible/useful in all the diverse advice.
-a I don't pretend to know-it-all, nothing close (though
-a a few, well, SAY they do :-)
-a As said elsewhere, "NEED" eventually supersedes all else.
-a Earlier on, all the shells maybe started with a narrow
-a 'vision' but HAD to include capabilities the other people
-a put in. Sometimes the additions were neat/clear, sometimes
-a not so much, but they're THERE.
-a A sort of 'functional convergence' due to NEEDS.
-a Anyway, I reject 'religions' built around shells or
-a langs and such. For each need, pick the best tools
-a and approaches. "Best" - sometimes that's what most
-a harmonizes with YOUR brand of thinking.
[...]--- Synchronet 3.22a-Linux NewsLink 1.2
On 10/24/26 07:06, Nuno Silva wrote:
On 2026-10-24, c186282 wrote:
-a-a KSH and BASH are "mostly compatible". Minor diffs but
-a-a nothing hard to get around. Anything you can do in KSH
-a-a can be done in BASH.
Now that sounds like something Greg Sullivan of Microsoft would say.
-a You "can do" most anything in COBOL or FORTRAN or 'C'
-a in ANY other lang - just as anything in one shell script
-a you can do in any other. The issue is 'how easily' - and
-a that depends on your particular application.
[...]--- Synchronet 3.22a-Linux NewsLink 1.2
On 2026-10-26 04:49, c186282 wrote:
On 10/24/26 05:53, Janis Papanagnou wrote:
On 2026-10-24 05:59, c186282 wrote:
-a-a KSH and BASH are "mostly compatible". Minor diffs but
-a-a nothing hard to get around. Anything you can do in KSH
-a-a can be done in BASH.
This is not correct. While Bash has over time adopted quite
some features from Ksh - but sadly needed often far too many
years for that - there's still a lot substantial functions
in Ksh that Bash has not implemented (or substituted by some
equivalent feature).
[...]
-a-a If KSH works for you then stick with it.
I use Ksh meanwhile for ~35 years, and since I use a lot of
Ksh features that Bash just doesn't support there's certainly
no reason to switch. (Also its speed is a criterion, BTW. And
back then there were no Bash versions on commercial Unixes;
but commercial Linuxes haven't the prevalence any more today,
and play no role now in my private systems context.)
-a-a Note however that MOST distros default to BASH now -
Yes, Linux often has Bash as its default.
-a-a so if you plan
-a-a to 'distribute' your work it'd be better to move your
-a-a stuff over to BASH.
Certainly not.
And this is short-sighted. If _portability_ is an issue you'd
better stick with the POSIX subset of both, Ksh and Bash.
(Sadly, as we observe, especially "Linux kiddies" often don't
know what that POSIX subset actually is; in their scripts they
are writing #!/bin/sh and using some bash-specific syntax not
realizing that there's a difference, and that Bash is not the
[portable] *standard*.)
-a-a Delay does not necessarily equate with 'ignorance'.
Not sure why you bring up that statement; nobody said anything
like that.
My statement about the delayed adoption of Ksh features by Bash
was just pointing to the fact that Bash users had to wait before
they could use comfortable features that Ksh had already provided
for (often many) years.
(It's also a bit unfortunate that Bash implemented some features,
and some of them IMO unnecessarily, in a slightly different way,
so that you cannot easily switch between the two shells.)
-a-a Different developers ... everyone had their own idea
-a-a of a 'perfect shell' - but then had to adapt and adapt
-a-a again as the Real Needs became more obvious. Now, most
-a-a WILL do the same things - just slightly different syntax.
Where not outright wrong, this is very fuzzy. (I don't think it
makes sense to comment on that.)
-a-a I rather liked the 'c-shell' long back because it, well,
-a-a 'looked like C'. It was one of the things that attracted
-a-a me to Unix/Linux in the first place. Now I'm happy (ok,
-a-a that's not really the word) with BASH. Not much it
-a-a can't do ... but the fudged-fudged-fudged syntax to add
-a-a features over time CAN be a pain in the ass.
If you think it's a good idea to use C-shell for scripting, and
since for you it seems to be just a syntax-difference, well, do
what you like in your private working context.
-a-a If I want 'scripting' now, I do it with Python. BASH
-a-a or whatever, maybe it just EVOKES the Python scripts,
-a-a a one or two-liner. Wrote many rather long BASH scripts
-a-a in the past, and while they worked they were, well, just
-a-a TOO messy/cryptic.
Yes, I share your valuation of shells being somewhat cryptic and
also "messy".
-a-a Last big BASH script I did was for backing up a bunch
-a-a of on-network drives. Mount-checking (important), calls
-a-a to rsync and openSSH and ftp, local AND 'cloud', all
-a-a with some confirmation checks along the way. SEMI-repeating
-a-a units but each was a bit 'special'.
Well, that's what their task mostly is, to "glue" together tools,
even if that glue is quite powerful (yet still somewhat "messy").
-a-a (ALWAYS pre-encrypt before sending to 'cloud' - I do
-a-a NOT trust the providers one bit. Mine had an extension
-a-a that 'suggested' some kind of ZIP, just to help them
-a-a waste time)
I suppose you mean here that you encrypt your backups before you
store them in the cloud. That's sensible. (I don't store my files
or backups in the cloud in the first place.)
-a-a RE-did it in Python. MUCH shorter and more comprehensible
-a-a and a nice easy CSV config file you could tweak in a text
-a-a editor or spreadsheet. Python made it all BETTER.
I'm not too surprised that you say that.
-a-a Had a 'C' version of that backup pgm earlier. Worked
-a-a fine, but there was a lot of string checking/editing
-a-a involved and 'C' is CLUNKY with string shit.
Shell has higher abstractions compared to "C".
-a-a Python
-a-a stole the FORTRAN string-slicing stuff, good deal !
Not sure what you are actually referring here to. (Shell has
such things as well. Unless you mean something more fancy when
you talk about Python or FORTRAN? My memories on FORTRAN are
very faint after 40 years when I had last used it.)
-a-a Won't ever use BASH or any other 'shell script' for
-a-a that kind of task again. Those might be "trad -IX"
-a-a but IMHO they're just OVER. "Solutions" MOVED ON.
With the POSIX-Shells you have at least a standard language on
Unixes.
-a-a Most distros, users would have to
-a-a intentionally install KSH and scripts would need a
-a-a shebang at the top to specify which shell to use.
I don't recall to have ever "intentionally installed" Ksh on
Linux. Earlier Linuxes had just a (sub-optimal!) PD-ksh that
you could use out of the box. Later they provided also the
much better "original" [AT&T] Ksh, and meanwhile you find on
(some?) Linuxes ksh93u+m, Martijn Dekker's improved version.
-a-a Um ... these days, almost all distros, you do have
-a-a to "intentionally install" KSH or any rivals. BASH
-a-a rules by default.
Mine has even ksh93u+m. Some years ago I've got original ksh. Yet
more years ago I've seen PD-ksh. All in the standard distribution.
As said, I've never had to "intentionally install" a Ksh.
There's one exception; back in the days when there was only PD-Ksh
provided in the distributions and that was the only existing Ksh,
but AT&T released their ksh93 for free, I got that from the AT&T
to replace the PD-thing. That *was* an intentional install. But
all those who didn't want The Original they could just use the one
that came with the distros.
Of course you can call shell scripts with the executable-flag
set and with a she-bang line on top. Personally I run most
shell scripts with the interpreter explicitly provided, as in
-a-a ksh my-script flags arguments
allowing easy debugging by adding '-x' without explicit 'set'
command in the script. Only if some script gets the status of
a "standard-command" I regularly use I set the shebang. YMMV.
-a-a Direct spec of the shell IS useful, however the
-a-a 'shebang' trick works well too.
(So you don't "need" a shebang line, but it's useful.)
-a-a (good practice, always add the shebang)
Yes, if you are running your scripts executable script (or if
you want to give your editor a hint for syntax highlighting,
what shell is actually used, or just for specification purpose;
but note, as said above, that we see many people have no clear
idea about what a "/bin/sh" is and where /bin/bash or /bin/ksh
differ from the standard.
-a-a I have, I think, FOUR shells on this machine. The
-a-a shebang IS required/desired.
-a-a Still kinda like 'tcsh', it's there in case I get
-a-a a brain-worm and wanna fool around with olde-dayz
-a-a stuff. Zsh ain't so bad either.
I don't consider it a good idea to use 'csh' or 'tcsh' for shell *programming*. (Browse the web for the well known paper with the
thoroughly explained rationales.)
I know some folks use and prefer 'tcsh' as interactive shells.
(Although the Bash, Zsh, or Ksh, are meanwhile quite comfortable
to use also interactively.)
-a-a But, as said, for 'scripting' now, Python is by
-a-a far my #1 choice unless the script is like a
-a-a simple little ten-liner. Tend to use those on
-a-a my PIs.
-a-a Fooled around a bit with Lua and Ruby for scripting,
-a-a but, really, Python and the MASSIVE number of libs
-a-a have them beat. Python will do ANYTHING these days,
-a-a just not as fast as 'C'. However CPU speed is less
-a-a of a thing these days - I/O ops drag things down
-a-a more than the CPU - so Python and 'C' can usually
-a-a compete pretty well.
-a-a There are Python compilers too. The main issue there
-a-a is all the sub-sub-versions of Python, so you always
-a-a have to include the entire binaries into each compilation
-a-a to ensure it'll work. But then space is also a minor
-a-a issue now as well unless you're doing micro-controllers.
-a-a Did my share of those - DO teach you how to really SQUEEZE
-a-a shit in - all computer kiddies should learn MCs first.
(I'm not the appropriate person to discuss Python, Lua, Ruby, and
similar.)
On 2026-10-26 05:28, c186282 wrote:
On 10/24/26 12:53, Janis Papanagnou wrote:
On 2026-10-24 14:10, Kenny McCormack wrote:
In article <11bhv61$1bjd2$2@dont-email.me>,
Janis Papanagnou-a <janis_papanagnou+ng@hotmail.com> wrote:
On 2026-10-24 05:59, c186282 wrote:
-a-a KSH and BASH are "mostly compatible". Minor diffs but
-a-a nothing hard to get around. Anything you can do in KSH
-a-a can be done in BASH.
This is not correct. While Bash has over time adopted quite
some features from Ksh - but sadly needed often far too many
years for that - there's still a lot substantial functions
in Ksh that Bash has not implemented (or substituted by some
equivalent feature).
I think you're taking Speedy way too seriously here.-a All of your
points
are valid, of course, but it should not be necessary to respond to the >>>> troll.
Oh, I didn't knew that he's considered a troll. His post looked just
uninformed, or opinionated, not trollish.
-a-a Thanks, I think-a :-)
-a-a I never 'troll' - just relate MY experiences and what
-a-a I've learned from them. I had my own path, own needs,
-a-a my own 'environments', starting in the punch-card days.
(I've still a block of punch-cards in the cellar, so I've got an
impression where you're coming from.)
-a-a But what served ME best may not be what served YOU, what
-a-a I learned may not be what YOU learned.
Oh, in my previous reply I said something like that; that it's
fine if it serves you in your private contexts.
A discrepancy might appear when you are working in a larger area
with rules that are not tailored to your personal likings but to
experiences from a larger (or another) group of persons.
-a-a A great purpose of groups like these is to bring in
-a-a a LOT of experiences and perspectives. You MAY find
-a-a something compatible/useful in all the diverse advice.
-a-a I don't pretend to know-it-all, nothing close (though
-a-a a few, well, SAY they do :-)
Oh, I'm completely fine with that. - If statements are formulated
as opinions, what else can we do than agree to those or disagree.
A problem, though, is, when opinions are stated as facts, and if
statements are even wrong, then they should be corrected. (Or
clarified if it was just a communication or expression problem.)
-a-a As said elsewhere, "NEED" eventually supersedes all else.
-a-a Earlier on, all the shells maybe started with a narrow
-a-a 'vision' but HAD to include capabilities the other people
-a-a put in. Sometimes the additions were neat/clear, sometimes
-a-a not so much, but they're THERE.
-a-a A sort of 'functional convergence' due to NEEDS.
-a-a Anyway, I reject 'religions' built around shells or
-a-a langs and such. For each need, pick the best tools
-a-a and approaches. "Best" - sometimes that's what most
-a-a harmonizes with YOUR brand of thinking.
(Unfortunately language topics are often presented with religious
fervor.)
On 2026-10-26 04:59, c186282 wrote:
On 10/24/26 07:06, Nuno Silva wrote:
On 2026-10-24, c186282 wrote:
-a-a KSH and BASH are "mostly compatible". Minor diffs but
-a-a nothing hard to get around. Anything you can do in KSH
-a-a can be done in BASH.
Now that sounds like something Greg Sullivan of Microsoft would say.
-a-a You "can do" most anything in COBOL or FORTRAN or 'C'
-a-a in ANY other lang - just as anything in one shell script
-a-a you can do in any other. The issue is 'how easily' - and
-a-a that depends on your particular application.
(Be careful; if your argument goes in the direction of "language A
is Turing-complete so you can use it simulating language B, which
makes it effectively the same", then that statement would be of no
practical value!)
I want to give you an example code from a little shell tool I wrote:
-a function LINES.set { .kosta.rows=${.sh.value} ; .kosta.update ;}
(What this does is that whenever the environment variable 'LINES'
changes - e.g. if you pick your mouse and resize the X-window - then
that function gets triggered and the actions defined executed.)
I'd be interested how you'd implement or emulate that, say, in Bash.
I want to give you an example code from a little shell tool I wrote:
function LINES.set { .kosta.rows=${.sh.value} ; .kosta.update ;}
(What this does is that whenever the environment variable 'LINES'
changes - e.g. if you pick your mouse and resize the X-window - then
that function gets triggered and the actions defined executed.)
I'd be interested how you'd implement or emulate that, say, in Bash.
Janis Papanagnou <janis_papanagnou+ng@hotmail.com> writes:
[...]
I want to give you an example code from a little shell tool I wrote:
function LINES.set { .kosta.rows=${.sh.value} ; .kosta.update ;}
(What this does is that whenever the environment variable 'LINES'
changes - e.g. if you pick your mouse and resize the X-window - then
that function gets triggered and the actions defined executed.)
I'd be interested how you'd implement or emulate that, say, in Bash.
I do something similar in bash, but using a different mechanism (so this isn't a response to your challenge).
I trap on SIGWINCH.
On 10/26/26 01:41, Janis Papanagnou wrote:
On 2026-10-26 04:59, c186282 wrote:
On 10/24/26 07:06, Nuno Silva wrote:
On 2026-10-24, c186282 wrote:
-a-a KSH and BASH are "mostly compatible". Minor diffs but
-a-a nothing hard to get around. Anything you can do in KSH
-a-a can be done in BASH.
Now that sounds like something Greg Sullivan of Microsoft would say.
-a-a You "can do" most anything in COBOL or FORTRAN or 'C'
-a-a in ANY other lang - just as anything in one shell script
-a-a you can do in any other. The issue is 'how easily' - and
-a-a that depends on your particular application.
(Be careful; if your argument goes in the direction of "language A
is Turing-complete so you can use it simulating language B, which
makes it effectively the same", then that statement would be of no
practical value!)
-a I know I will offend The Religious.
-a Fuck 'em.
-a Task, BEST solution for YOU. That is The
-a Standard so far as I'm concerned.
I want to give you an example code from a little shell tool I wrote:
-a-a function LINES.set { .kosta.rows=${.sh.value} ; .kosta.update ;}
(What this does is that whenever the environment variable 'LINES'
changes - e.g. if you pick your mouse and resize the X-window - then
that function gets triggered and the actions defined executed.)
I'd be interested how you'd implement or emulate that, say, in Bash.
-a I'd write it in PYTHON now-a :-)
-a Some OS.System() and maybe Subprocess calls.
-a Leverage the existing simple/low-level utils
-a in Linux to your advantage. Run, parse,
-a mutate, USE.
-a Done a lot of apps that simply grab output
-a from those simple Linux utils - then parse
-a and trim and mix. Why re-invent the wheel ?
-a You CAN - but WHY ???
On 10/26/26 01:04, Janis Papanagnou wrote:[...]
On 2026-10-26 04:49, c186282 wrote:
On 10/24/26 05:53, Janis Papanagnou wrote:
On 2026-10-24 05:59, c186282 wrote:
Not sure why you bring up that statement; nobody said anything
like that.
-a My following exposition explains why.
My statement about the delayed adoption of Ksh features by Bash
was just pointing to the fact that Bash users had to wait before
they could use comfortable features that Ksh had already provided
for (often many) years.
-a Well, good for them.
-a I'll stick to my philosophic "Functional Convergence"
-a theory here. If shell X does something neat-o then it
-a HAS to be incorporated into all others - some way, some
-a how, neat or not.
(It's also a bit unfortunate that Bash implemented some features,
and some of them IMO unnecessarily, in a slightly different way,
so that you cannot easily switch between the two shells.)
-a-a Different developers ... everyone had their own idea
-a-a of a 'perfect shell' - but then had to adapt and adapt
-a-a again as the Real Needs became more obvious. Now, most
-a-a WILL do the same things - just slightly different syntax.
Where not outright wrong, this is very fuzzy. (I don't think it
makes sense to comment on that.)
-a I think it makes good sense. Everybody just HAS to
-a somehow incorporate everyone else's good ideas
-a eventually. Otherwise --> TRASH BIN of history.
[snip Csh talk]
-a-a If I want 'scripting' now, I do it with Python. BASH
-a-a or whatever, maybe it just EVOKES the Python scripts,
-a-a a one or two-liner. Wrote many rather long BASH scripts
-a-a in the past, and while they worked they were, well, just
-a-a TOO messy/cryptic.
Yes, I share your valuation of shells being somewhat cryptic and
also "messy".
-a VERY messy and OFTEN 'cryptic' (unless very short).
-a Just no NEED for such shit anymore.
-a-a Last big BASH script I did was for backing up a bunch
-a-a of on-network drives. Mount-checking (important), calls
-a-a to rsync and openSSH and ftp, local AND 'cloud', all
-a-a with some confirmation checks along the way. SEMI-repeating
-a-a units but each was a bit 'special'.
Well, that's what their task mostly is, to "glue" together tools,
even if that glue is quite powerful (yet still somewhat "messy").
-a But NOT 'comprehensible' a year or two LATER.
[snip Python talk]
-a-a (ALWAYS pre-encrypt before sending to 'cloud' - I do
-a-a NOT trust the providers one bit. Mine had an extension
-a-a that 'suggested' some kind of ZIP, just to help them
-a-a waste time)
I suppose you mean here that you encrypt your backups before you
store them in the cloud. That's sensible. (I don't store my files
or backups in the cloud in the first place.)
-a I very much mean to encrypt the backups BEFORE you
-a send them to 'cloud'.
Don't put un-encrypted shit
-a on their sites for a millisecond. Do NOT trust them !
-a AES-xxx is good, though sometimes used Camilla.
-a With several langs/FTP you can put a file on 'cloud'
-a and then set the datetime/ownership to whatever the
-a original had. That's how I did it for a decade+ - works
-a very well. Encrypt, upload, then tweak the specifics.
-a Pirates get NOTHING.
-a-a RE-did it in Python. MUCH shorter and more comprehensible
-a-a and a nice easy CSV config file you could tweak in a text
-a-a editor or spreadsheet. Python made it all BETTER.
I'm not too surprised that you say that.
-a Do you disagree ???
-a-a Had a 'C' version of that backup pgm earlier. Worked
-a-a fine, but there was a lot of string checking/editing
-a-a involved and 'C' is CLUNKY with string shit.
Shell has higher abstractions compared to "C".
-a Slightly - but also CLUNKY and hard to understand
-a a year or two later.
Doing serious string stuff in
-a BASH and several others ... nasty !
-a Look, if you're 20-something it MAY be possible to
-a hold all the sense/logic of older shit in your head.
-a Believe me - that CHANGES. Too much shit on a
-a small drive, so to speak.
-a-a Python
-a-a stole the FORTRAN string-slicing stuff, good deal !
Not sure what you are actually referring here to. (Shell has
such things as well. Unless you mean something more fancy when
you talk about Python or FORTRAN? My memories on FORTRAN are
very faint after 40 years when I had last used it.)
-a Um ... check. Almost all the Python string shit
-a was 'borrowed' from FORTRAN. This is a GOOD thing.
-a SO often you have to slice and dice and squish
-a string stuff together in the real world. It'd
-a better be EASY and OBVIOUS or you're doomed.
[snip]
-a-a Um ... these days, almost all distros, you do have
-a-a to "intentionally install" KSH or any rivals. BASH
-a-a rules by default.
Mine has even ksh93u+m. Some years ago I've got original ksh. Yet
more years ago I've seen PD-ksh. All in the standard distribution.
As said, I've never had to "intentionally install" a Ksh.
-a Umm ... what IS this 'ksh'-defaulting distro ???
[snip]
[...]
-a Never got TOO far into ZSH ... but it DOES have a lot
-a of potential. Don't count it out of the picture.
[big snip of digressions]Janis
-a My first computer was a PDP-11. Great unit.
-a But still remember having to submit my card
-a stack (DON'T drop it !) to the IT overlords.
-a Helped a bit if you offered them 'gifts'.
On 2026-10-26 08:38, c186282 wrote:
[...]
-a-a My first computer was a PDP-11. Great unit.
-a-a But still remember having to submit my card
-a-a stack (DON'T drop it !) to the IT overlords.
Corrupted data (if you've dropped it)...
-a-a Helped a bit if you offered them 'gifts'.
...and corrupted people?
Janis
PS: The nastiest thing I've encountered with the
card deck was that a friend exchanged all my cards
which carried the word 'BEGIN' with cards carrying
the word 'SADAT' - just as a hilarious play with
words WRT two international politicians back then.
Not so funny for me when I noticed all the syntax
errors and had to wait another hour for the next
batch run.
On 10/27/26 00:01, Janis Papanagnou wrote:
[...]
-a The SysLord had a serial terminal, but all
-a the rest of us had to do cards.
PS: The nastiest thing I've encountered with the
card deck was that a friend exchanged all my cards
which carried the word 'BEGIN' with cards carrying
the word 'SADAT' - just as a hilarious play with
words WRT two international politicians back then.
Not so funny for me when I noticed all the syntax
errors and had to wait another hour for the next
batch run.
-a Admit it, was a CLEVER joke !
-a SADAT ... do, but barely, remember it now.
[...]
On 2026-10-27 17:24, c186282 wrote:
On 10/27/26 00:01, Janis Papanagnou wrote:
[...]
-a-a The SysLord had a serial terminal, but all
-a-a the rest of us had to do cards.
I started with punch cards and switched early to a
terminal.
(Frankly, the cards had an advantage; the terminal
made us stay constantly in front of it; write code,
make some changes/fixes, compile, rinse repeat.
With the cards we spent time beforehand designing
the code, seeing and fixing all errors in one turn.
It turned out to be, WRT net-time, more efficient.
But, OTOH, fixing a punching mistake was more time
consuming; needed a lot discipline and attention.)
PS: The nastiest thing I've encountered with the
card deck was that a friend exchanged all my cards
which carried the word 'BEGIN' with cards carrying
the word 'SADAT' - just as a hilarious play with
words WRT two international politicians back then.
Not so funny for me when I noticed all the syntax
errors and had to wait another hour for the next
batch run.
-a-a Admit it, was a CLEVER joke !
(I can't tell you how pissed I was at that moment!)
But, with some distance, it was a marvellous prank.
(That's why I still remember it after 40+ years.)
He was all like that. Sadly he died extremely young.
Checked a couple hours ago ... barely ANY reference-a-a SADAT ... do, but barely, remember it now.
(I do recall his name only because of that joke.)
On 2026-10-27 17:24, c186282 wrote:
On 10/27/26 00:01, Janis Papanagnou wrote:
[...]
-a-a The SysLord had a serial terminal, but all
-a-a the rest of us had to do cards.
I started with punch cards and switched early to a
terminal.
(Frankly, the cards had an advantage; the terminal
made us stay constantly in front of it; write code,
make some changes/fixes, compile, rinse repeat.
With the cards we spent time beforehand designing
the code, seeing and fixing all errors in one turn.
It turned out to be, WRT net-time, more efficient.
But, OTOH, fixing a punching mistake was more time
consuming; needed a lot discipline and attention.)
PS: The nastiest thing I've encountered with the
card deck was that a friend exchanged all my cards
which carried the word 'BEGIN' with cards carrying
the word 'SADAT' - just as a hilarious play with
words WRT two international politicians back then.
Not so funny for me when I noticed all the syntax
errors and had to wait another hour for the next
batch run.
-a-a Admit it, was a CLEVER joke !
(I can't tell you how pissed I was at that moment!)
But, with some distance, it was a marvellous prank.
(That's why I still remember it after 40+ years.)
He was all like that. Sadly he died extremely young.
-a-a SADAT ... do, but barely, remember it now.
(I do recall his name only because of that joke.)
[...]
I had been told once that IBM developers did not have a compiler or such
at their disposal but only a text editor, so they, too, had to think
twice before committing their code.
PS: The nastiest thing I've encountered with the
card deck was that a friend exchanged all my cards
which carried the word 'BEGIN' with cards carrying
the word 'SADAT' - just as a hilarious play with
words WRT two international politicians back then.
Not so funny for me when I noticed all the syntax
errors and had to wait another hour for the next
batch run.
-a-a Admit it, was a CLEVER joke !
(I can't tell you how pissed I was at that moment!)
But, with some distance, it was a marvellous prank.
(That's why I still remember it after 40+ years.)
He was all like that. Sadly he died extremely young.
-a-a SADAT ... do, but barely, remember it now.
(I do recall his name only because of that joke.)
This is SOOO funny! At that time I myself had put a note on the wall
stating that with the developments then the keyword BEGIN should be
replaced by SADAT.
I'm 70 now, retired 4 years ago.
Josef
Ah ... yes ... The Olde Times ;-)
I started studying CS in 1977 (after two years trying my luck at EE).
Indeed, we too had to use punched cards in the beginning.
I miss the old Gen-3 programming IDEs. Not
too many of them left. Very helpful and
easy to navigate. Quick debugging and fixes.
The successors - FAR too complicated and/or
overly-abstract. Can't even figure out what
they're saying - much less the problems with
your code.
On Wed, 28 Oct 2026 17:01:44 +0100, Josef M||llers wrote:
Ah ... yes ... The Olde Times ;-)
I started studying CS in 1977 (after two years trying my luck at EE).
Indeed, we too had to use punched cards in the beginning.
The IBM 2260 terminal came out in 1964 but I didn't see one until I interviewed with IBM/Owego in 1968. If there were any in the college they were buried in a cave.
I didn't get the job. IBM found my programming aptitude questionable.
That was much earlier in my life, perhaps Junior High when I was
also in an elective class to earn my Amateur Radio Novice Licence. Upon earning that and quickly moving to Technician, I became a member of the
local ARRL Club. There I met all kinds of electricians, radio and
antenna experts, and one Advanced class leader with his own antenna farm
on his farm. He was doing moon bounce and slow scan TV. He showed me
his latest project -- a home built Z80 Computer that booted by flipping switches to input the very tiny OS or a simple program. Now that really helped me form my programming and diagnostic capabilities.
The closest I ever came to using an IDE was the symbolic debugger that
came with Lattice C for MS-DOS. Simple, effective... and gone.
I too grew up accessing computers (the big iron) first with cards
and later with terminals as they became available around campus. I
agree that cards or anything low tech makes the developer first
think about what they want to build and then do so.
On Wed, 28 Oct 2026 23:28:12 -0000 (UTC), David LaRue wrote:
I too grew up accessing computers (the big iron) first with cards
and later with terminals as they became available around campus. I
agree that cards or anything low tech makes the developer first
think about what they want to build and then do so.
Imagine if your calculator was like that: you had to think carefully
what you wanted to do, and write it down on a coding form or
something, before simply switching the thing on and punching buttons
...
Well ... THAT'S HOW IT WAS not ALL so long ago.
Get it right the FIRST time, or SUFFER.
That plus minimal memory/storage ensured good quality, tight, code.
On Wed, 28 Oct 2026 19:33:23 GMT, Charlie Gibbs wrote:
The closest I ever came to using an IDE was the symbolic debugger that
came with Lattice C for MS-DOS. Simple, effective... and gone.
I missed that one although I did use MS C after they bought Lattice and developed it further. I will say for its era Microsoft C had excellent
hard copy documentation.
On Wed, 28 Oct 2026 22:36:37 -0400
c186282 <c186282@nnada.net> wrote:
Well ... THAT'S HOW IT WAS not ALL so long ago.
Get it right the FIRST time, or SUFFER.
That plus minimal memory/storage ensured good quality, tight, code.
:s/ensured/encouraged/
I too grew up accessing computers (the big iron) first with cards and later with terminals as they became available around campus.
On 2026-10-28, David LaRue <huey.dll@tampabay.rr.com> wrote:
I too grew up accessing computers (the big iron) first with cards and later >> with terminals as they became available around campus.
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
On 2026-10-28, David LaRue <huey.dll@tampabay.rr.com> wrote:
I too grew up accessing computers (the big iron) first with cards and later >> with terminals as they became available around campus.
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
Depends on the country and the uni. Even depends on the department.
I think I saw it in 1983, but there were departments that had terminals earlier.
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
On 2026-10-29, Carlos E.R. <robin_listas@es.invalid> wrote:
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
Depends on the country and the uni. Even depends on the department.
I'm asking for personal recollections. When did it happen for you?
I've been wondering for some time, and this seems like the perfect opportunity to ask, given that people have reminisced about their
punch card days.
And yes, I expect computer science got terminals earlier than
engineering.
I think I saw it in 1983, but there were departments that had terminals
earlier.
That's one data point, thanks.
On 2026-10-29, Carlos E.R. <robin_listas@es.invalid> wrote:
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
Depends on the country and the uni. Even depends on the department.
I'm asking for personal recollections. When did it happen for you?
I've been wondering for some time, and this seems like the perfect opportunity to ask, given that people have reminisced about their
punch card days.
And yes, I expect computer science got terminals earlier than
engineering.
I think I saw it in 1983, but there were departments that had terminals
earlier.
That's one data point, thanks.
I attended university from 1968 to 1971. In the fall of 1968 they
converted from an IBM 7044 to a 360/67. The 360 had a few 2260s, but I wouldn't call them common and access was fairly restricted.
rbowman wrote this screed in ALL-CAPS (fixed):
On Wed, 28 Oct 2026 19:33:23 GMT, Charlie Gibbs wrote:
The closest I ever came to using an IDE was the symbolic debugger that
came with Lattice C for MS-DOS. Simple, effective... and gone.
I missed that one although I did use MS C after they bought Lattice and
developed it further. I will say for its era Microsoft C had excellent
hard copy documentation.
I agree with that last sentence. In the 80's the manuals for their
C and FORTRAN were very good, and obviously inspired by the layout
of DEC's manuals.
On 2026-10-28, David LaRue <huey.dll@tampabay.rr.com> wrote:
I too grew up accessing computers (the big iron) first with cards and later >> with terminals as they became available around campus.
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
On 2026-10-29, John Ames <commodorejohn@gmail.com> wrote:
On Wed, 28 Oct 2026 22:36:37 -0400
c186282 <c186282@nnada.net> wrote:
Well ... THAT'S HOW IT WAS not ALL so long ago.
Get it right the FIRST time, or SUFFER.
That plus minimal memory/storage ensured good quality, tight, code.
:s/ensured/encouraged/
It also encouraged efficient, logical, consistent system design.
There just weren't the resources to implement the kinds of cruft
that seems to be mandatory these days.
On Thu, 29 Oct 2026 17:37:23 -0000 (UTC), Christian Weisgerber wrote:
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
How about rCLthe 1970srCY?
her
programming course was in some language that I don't think survived
either.
That vaguely gybes with my memory. I THINK that serial terminals for
micros existed prior to 1980, but have no idea about mainframes etc.
VT100 was a 1978 released serial terminal
On 30/10/2026 04:33, rbowman wrote:
her programming course was in some language that I don't think survived
either.
Probably ALGOL.
That was all the rage back then
On 2026-10-29, Carlos E.R. <robin_listas@es.invalid> wrote:
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
On 10/29/26 12:27, Chris Ahlstrom wrote:
rbowman wrote this screed in ALL-CAPS (fixed):
On Wed, 28 Oct 2026 19:33:23 GMT, Charlie Gibbs wrote:
The closest I ever came to using an IDE was the symbolic debugger that >>>> came with Lattice C for MS-DOS. Simple, effective... and gone.
I missed that one although I did use MS C after they bought Lattice and
developed it further. I will say for its era Microsoft C had excellent
hard copy documentation.
I agree with that last sentence. In the 80's the manuals for their
C and FORTRAN were very good, and obviously inspired by the layout
of DEC's manuals.
You are correct - manuals/docs were EXTREMELY good
for such stuff "back in the day". Found a manual
for "Aztec C" (great !) from the old days ... WOW ! ...
literally bit-level shit in there in case you needed
to hyper-optimize something. Clearly the people who
wrote the doc also wrote the compiler.
More VOLUME of docs these days ... but "As Good" ???
DO have the Aztec compiler ... MIGHT still work in
DOSBox or a DOS or CP/M VM ........
On Fri, 30 Oct 2026 00:09:50 GMT, Charlie Gibbs wrote:
I attended university from 1968 to 1971. In the fall of 1968 they
converted from an IBM 7044 to a 360/67. The 360 had a few 2260s, but I
wouldn't call them common and access was fairly restricted.
The 360/67 was the start of something new for IBM -- timesharing. The
360/30 maxxed out at 64K and became a satellite processor for some of the later 360 machines. I wasn't too interested but a friend was and spoke of the future with awe. He eventually dropped out and moved to Boston where there was a lot more happening.
My girlfriend started at SUNY Albany before transferring out of that zoo.
I can't remember what they had, CDC maybe, but definitely not IBM and her programming course was in some language that I don't think survived
either.
On an early morning (about 2 am) job run, I literally brought the
mainframe to its knees while simultaneously burning up the entire
annual department time-sharing budget, via an extensive numerical
simulation involving 500x500 matrix calculations.
From that experience I learned a valuable computer science lesson
that is not found in the standard textbooks: the most powerful
digital computers are DAMNED SLOW AND IMPOTENT.
From that experience I learned a valuable computer science lesson
that is not found in the standard textbooks: the most powerful
digital computers are DAMNED SLOW AND IMPOTENT.
I wouldn't call that a computer science lesson; many CS weenies
assume that computers are infinitely powerful, and start whining when confronted with systems in the real world. I had a confrontation
with a guy like that when trying to determine why a COBOL program
was running like molasses through a pinhole in January. One reason
was that this genius had declared all subscripts as COMP-3 (packed
decimal) rather than binary. Just changing those declarations alone
knocked 30% off the execution time. I wish I could have had more
time to work on that program; I bet I could have really sped it up.
I calculated that the nested PERFORM VARYING loop that cleared a 3-dimensional array would have taken 25 minutes of the estimated
50-hour run time, due to a combination of those decimal subscripts
and subroutine calling overhead. (The CPU did about 130 KIPS.)
I still have the hard-copy manuals; they were well done.
On Fri, 30 Oct 2026 19:24:59 GMT, Charlie Gibbs wrote:
I still have the hard-copy manuals; they were well done.
ThatrCOs rare nowadays.
Some think thatrCOs a mis-steak ...
On 2026-10-30, c186282 <c186282@nnada.net> wrote:
On 10/29/26 12:27, Chris Ahlstrom wrote:
rbowman wrote this screed in ALL-CAPS (fixed):
On Wed, 28 Oct 2026 19:33:23 GMT, Charlie Gibbs wrote:
The closest I ever came to using an IDE was the symbolic debugger that >>>>> came with Lattice C for MS-DOS. Simple, effective... and gone.
I missed that one although I did use MS C after they bought Lattice and >>>> developed it further. I will say for its era Microsoft C had excellent >>>> hard copy documentation.
I agree with that last sentence. In the 80's the manuals for their
C and FORTRAN were very good, and obviously inspired by the layout
of DEC's manuals.
You are correct - manuals/docs were EXTREMELY good
for such stuff "back in the day". Found a manual
for "Aztec C" (great !) from the old days ... WOW ! ...
literally bit-level shit in there in case you needed
to hyper-optimize something. Clearly the people who
wrote the doc also wrote the compiler.
More VOLUME of docs these days ... but "As Good" ???
Hardly. Most manuals back in the day had a section titled
"Theory of Operation". I really miss that.
On 2026-10-29, Carlos E.R. <robin_listas@es.invalid> wrote:
For those of us too young: When exactly did terminals become common
on university campuses? Please be more specific than "the 1980s".
Depends on the country and the uni. Even depends on the department.
I'm asking for personal recollections. When did it happen for you?
I've been wondering for some time, and this seems like the perfect opportunity to ask, given that people have reminisced about their
punch card days.
And yes, I expect computer science got terminals earlier than
engineering.
I think I saw it in 1983, but there were departments that had terminals
earlier.
That's one data point, thanks.
On 2026-10-30 20:24, Charlie Gibbs wrote:
On 2026-10-30, c186282 <c186282@nnada.net> wrote:
On 10/29/26 12:27, Chris Ahlstrom wrote:
rbowman wrote this screed in ALL-CAPS (fixed):
On Wed, 28 Oct 2026 19:33:23 GMT, Charlie Gibbs wrote:
The closest I ever came to using an IDE was the symbolic debugger that >>>>>> came with Lattice C for MS-DOS. Simple, effective... and gone.
I missed that one although I did use MS C after they bought Lattice and >>>>> developed it further. I will say for its era Microsoft C had excellent >>>>> hard copy documentation.
I agree with that last sentence. In the 80's the manuals for their
C and FORTRAN were very good, and obviously inspired by the layout
of DEC's manuals.
You are correct - manuals/docs were EXTREMELY good
for such stuff "back in the day". Found a manual
for "Aztec C" (great !) from the old days ... WOW ! ...
literally bit-level shit in there in case you needed
to hyper-optimize something. Clearly the people who
wrote the doc also wrote the compiler.
More VOLUME of docs these days ... but "As Good" ???
Hardly. Most manuals back in the day had a section titled
"Theory of Operation". I really miss that.
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
On 2026-10-29, Carlos E.R. <robin_listas@es.invalid> wrote:
I think I saw it in 1983, but there were departments that had terminals
earlier.
That's one data point, thanks.
On 2026-10-31, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
Isn't that more about the viewer using to read the manual pages, than a
flaw of the online manual in itself?
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
There were only PCs when I was there, but my school also had accessYou can run this under SDL Hercules. I have a D6 system. Unfortunately,
to time sharing with the mainframes of the Michigan Terminal System,
or MTS:
<https://en.wikipedia.org/wiki/Michigan_Terminal_System>
My current desktop is probably a lot more powerful than thoseAbsolutely. A smartphone is.
mainframes but most tasks of interest are still largely untouchable.
[...] So, when I arrived to Madrid for my second year (I think that was in the 83), a friend told me that the computer building that was shared by
all in the campus had switched from cards to terminals, and that I had
just missed it.
And he described me the nightmare that first/second year students had to suffer to get their stack of cards punched and executed. Could take days.
[...]
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
I was forced by the system to take first year of engineering at the university in my province. All engineering had the same first year. Once
we (if!) we passed, we could actually choose what college to continue
in.
I have one recollection of maybe 1983 or 84, visiting the microwave department, which was designing the aerial complex of a new frigate.
They had a model to scale in an anechoic chamber; they had to scale the frequencies up to match the ship model. And they had an Apple-II (or was
it I?) doing some long calculations. The display was off, the teacher
was in the back, and he asked us near the machine to turn the display on
to see the simulation. None of us knew how to do that. He had to come to
the front, grumbling about how come we did not know where the switch
was.
Well, none of us had the money to buy an Apple II! And the IBM PC was
still a year or two away.
Le 31-10-2026, Nuno Silva <nunojsilva@invalid.invalid> a |-crit-a:
On 2026-10-31, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
Isn't that more about the viewer using to read the manual pages, than a
flaw of the online manual in itself?
Maybe. When I wanted to learned about Linux, I hated man pages. I found
them awful and useless. Now, I'm using them only when I know what I'm
looking for. I mean, when I don't know something, I know that the man
page will send me a useless shit load of details without anything
useful. Or, if anything is useful it's buried and hidden among useless
stuff. If I'm looking for a fast way to use a command, I'm using tldr.
If I want to understand a command, I avoid man pages and go look for
some information on Internet. It's only when I want to know what an
option is really doing or if an option exists that I'm going to look for
it in the man pages.
On 2026-10-31 11:40, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
I agree with you on the _horrendous large_ 'man' "pages".
They shouldn't replace exhaustive and thorough documentation, IMO.
But worse I think are the ("maneuverable") 'info' pages.
Or the habit of some "Unixes" to have no man-page information but
only a stub with the hint to look into the (IMO horrendous) 'info'
pages. (I'm aware that some certainly prefer them; how else would
their proliferation be otherwise explainable.)
Concerning "clickable" information generally, mind the time when
man-pages were introduced!
Janis, preferring to avoid mouse-devices in the first place
On 2026-10-31, Carlos E. R. wrote:
On 2026-10-30 20:24, Charlie Gibbs wrote:
On 2026-10-30, c186282 <c186282@nnada.net> wrote:
On 10/29/26 12:27, Chris Ahlstrom wrote:
rbowman wrote this screed in ALL-CAPS (fixed):
On Wed, 28 Oct 2026 19:33:23 GMT, Charlie Gibbs wrote:
The closest I ever came to using an IDE was the symbolic debugger that >>>>>>> came with Lattice C for MS-DOS. Simple, effective... and gone.
I missed that one although I did use MS C after they bought Lattice and >>>>>> developed it further. I will say for its era Microsoft C had excellent >>>>>> hard copy documentation.
I agree with that last sentence. In the 80's the manuals for their
C and FORTRAN were very good, and obviously inspired by the layout
of DEC's manuals.
You are correct - manuals/docs were EXTREMELY good
for such stuff "back in the day". Found a manual
for "Aztec C" (great !) from the old days ... WOW ! ...
literally bit-level shit in there in case you needed
to hyper-optimize something. Clearly the people who
wrote the doc also wrote the compiler.
More VOLUME of docs these days ... but "As Good" ???
Hardly. Most manuals back in the day had a section titled
"Theory of Operation". I really miss that.
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
Isn't that more about the viewer using to read the manual pages, than a
flaw of the online manual in itself?
On 2026-10-31 11:49, Carlos E. R. wrote:
[...] So, when I arrived to Madrid for my second year (I think that
was in the 83), a friend told me that the computer building that was
shared by all in the campus had switched from cards to terminals, and
that I had just missed it.
And he described me the nightmare that first/second year students had
to suffer to get their stack of cards punched and executed. Could take
days.
Strong words. And absurd time scales. They certainly don't match my observations. All of us did our punch cards based coding often just
during the pauses between the lectures; the basic code maybe in an
hour, the fixes - usually just one or two iterations needed - done
during the "cum tempore" 10 minutes. But most programming in those
early beginner courses also wasn't too voluminous, to be fair.
What certainly was painful was the (for obvious reasons) inherent non-existence of a <Del> or <Backspace> key on the punch devices.
(They had a useful "Copy" key (<Dup>?), though.)[*]
I don't remember if my friend punched his own code, or had it punched.
Would depend on how many punch machines they had for lowly students.
I remember his saying that he had to submit his stack of cards to a
clerk, then given a paper slip to collect the results another day.
On 2026-10-31 16:57, Janis Papanagnou wrote:
On 2026-10-31 11:40, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
I agree with you on the _horrendous large_ 'man' "pages".
They shouldn't replace exhaustive and thorough documentation, IMO.
But worse I think are the ("maneuverable") 'info' pages.
Or the habit of some "Unixes" to have no man-page information but
only a stub with the hint to look into the (IMO horrendous) 'info'
pages. (I'm aware that some certainly prefer them; how else would
their proliferation be otherwise explainable.)
Concerning "clickable" information generally, mind the time when
man-pages were introduced!
Yes, I understand that. Isn't time we get something better?
Janis, preferring to avoid mouse-devices in the first place
Sure, point and click can also work without a mouse. Use the cursor to select an entry in the table of contents, then hit enter.
The DEC systems are more fun, have better docs, and much more
software.
I have one recollection of maybe 1983 or 84, visiting the microwave department, which was designing the aerial complex of a new frigate.
They had a model to scale in an anechoic chamber; they had to scale
the frequencies up to match the ship model. And they had an Apple-II
(or was it I?) doing some long calculations.
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index.
Info pages are somewhat better.
For example, if you look at Perl, you will see its documentation is
broken up across a whole series of man pages -- over 200 of them.
(Is that going too far in the other direction, do you think ... ?)
On 2026-10-31 11:49, Nuno Silva wrote:
On 2026-10-31, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
Isn't that more about the viewer using to read the manual pages, than
a flaw of the online manual in itself?
No, man pages do not contain links to section of the document. It is
not structured.
On 2026-10-31 16:57, Janis Papanagnou wrote:
On 2026-10-31 11:40, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
I agree with you on the _horrendous large_ 'man' "pages".
They shouldn't replace exhaustive and thorough documentation, IMO.
But worse I think are the ("maneuverable") 'info' pages.
Or the habit of some "Unixes" to have no man-page information but
only a stub with the hint to look into the (IMO horrendous) 'info'
pages. (I'm aware that some certainly prefer them; how else would
their proliferation be otherwise explainable.)
Concerning "clickable" information generally, mind the time when
man-pages were introduced!
Yes, I understand that. Isn't time we get something better?
--Janis, preferring to avoid mouse-devices in the first place
Sure, point and click can also work without a mouse. Use the cursor to
select an entry in the table of contents, then hit enter.
On 2026-10-31 19:56, Carlos E. R. wrote:
On 2026-10-31 16:57, Janis Papanagnou wrote:
On 2026-10-31 11:40, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
I agree with you on the _horrendous large_ 'man' "pages".
They shouldn't replace exhaustive and thorough documentation, IMO.
But worse I think are the ("maneuverable") 'info' pages.
Or the habit of some "Unixes" to have no man-page information but
only a stub with the hint to look into the (IMO horrendous) 'info'
pages. (I'm aware that some certainly prefer them; how else would
their proliferation be otherwise explainable.)
Concerning "clickable" information generally, mind the time when
man-pages were introduced!
Yes, I understand that. Isn't time we get something better?
But technically we have already "clickable" documents; with HTML for
example.
(Although it should still be available in one page, and ideally also searchable by regexps in easy ways. (As opposed to click on a search
icon, move the mouse to a new sub-window, (optionally) click another
time in that window to "activate" it for input, type in only a fixed non-regexp text (often without possibility to distinguish caps, or,
if you're lucky, only with an extra click in some control-box), and
hit enter (if you're lucky) or have to again move the mouse to click
on some Okay button.) - I certainly hate to have no '/search-pattern'
or '<Ctrl-F>search-pattern' available; with emphasis on "pattern".)
Personally I typically want (at least) three levels of information.
Terse usage information, complete interface description, and thorough explanations of all details in case of the more complex tools.
Mostly we (now) have all that information; if only through the Web.
(But often inconsistent or even unreliable.)
Should we have everything just in one (clickable) format like HTML?
I don't think so. Should we have all the spread existing information rewritten in some other format? Well, who would do that job?
What I (also) criticize is that we don't have a consistent level of information with some newer Linux platforms. I wished for the "second
level" of information that the man page format would have been kept/ maintained instead of trying to deprive that format in favor of the
'info' format.
Janis, preferring to avoid mouse-devices in the first place
Sure, point and click can also work without a mouse. Use the cursor to
select an entry in the table of contents, then hit enter.
Yes, sure. For long documents ("level three" information) with a lot
content where a "table of contents" is sensible that makes completely
sense. (Even pure mouse-control would be fine here [for me] in this
case.)
For the information level where I'd see the man-pages I wouldn't even
need a "table of contents", since that type of information shouldn't
be that voluminous, IMO.
On Sat, 31 Oct 2026 11:07:47 -0400, jayjwa wrote:
The DEC systems are more fun, have better docs, and much more
software.
If you look at all the vintage docs collected at Bitsavers, out of (currently) about 660GB in total, over 40% is from DEC. Number 2 is
IBM, but the DEC collection is bigger by nearly 3:1.
Number 3 is CDC, at about a quarter of the size of the IBM collection.
On 2026-10-31 16:57, Janis Papanagnou wrote:
On 2026-10-31 11:40, Carlos E. R. wrote:
I agree with you on the _horrendous large_ 'man' "pages".
Talking of manuals, I find large man pages very difficult to use, specially >>> when you need something specific. No clickable table of content, no
clickable index. Info pages are somewhat better.
They shouldn't replace exhaustive and thorough documentation, IMO.
But worse I think are the ("maneuverable") 'info' pages.
Or the habit of some "Unixes" to have no man-page information but
only a stub with the hint to look into the (IMO horrendous) 'info'
pages. (I'm aware that some certainly prefer them; how else would
their proliferation be otherwise explainable.)
Concerning "clickable" information generally, mind the time when
man-pages were introduced!
Yes, I understand that. Isn't time we get something better?
On 10/31/26 17:38, Lawrence DrCOOliveiro wrote:
On Sat, 31 Oct 2026 11:07:47 -0400, jayjwa wrote:
The DEC systems are more fun, have better docs, and much more
software.
If you look at all the vintage docs collected at Bitsavers, out of
(currently) about 660GB in total, over 40% is from DEC. Number 2 is
IBM, but the DEC collection is bigger by nearly 3:1.
Number 3 is CDC, at about a quarter of the size of the IBM collection.
The DEC boxes were GREAT !
Not entirely sure why they went out of biz - likely
idiot management. Daddy builds it, Junior TRASHES it.
Still wish someone would modernize VMS - it was a
fantastic/powerful system with a ton of prescient
modern features - could do international internet
before there WAS an internet.
On 2026-10-31 20:08, Carlos E. R. wrote:
I don't remember if my friend punched his own code, or had it punched.
Would depend on how many punch machines they had for lowly students.
We had a "row" of such machines available. Don't recall to have
ever waited for a free place (or the waiting time was negligible).
I remember his saying that he had to submit his stack of cards to a
clerk, then given a paper slip to collect the results another day.
The way it was done in our place was technically similar but did
not require that much time.
The batch cards reader was in an accessible are, you could just
put the cards either directly in the reader device or in a queue
on a nearby desk where an operator did that job then. The reader
was constantly fed with cards by the operators since the queue
was also regularly filled by all the students; hardly more delay
than, maybe, 15 minutes. The printouts were quickly created with
the batch processing by super-fast chain printers, so were also
available in short term. Some delay was due to the fanfold paper
needed to be physically separated (by the jobs or user IDs) to be
associated to the different users. The output landed sorted in a
special shelf where we could fetch them. The whole batch process,
from the punch cards delivery to getting the printout results in
(typically less than) an hour.
A peculiarity of the physical separation of the printouts was the
application of a nice trick; the printing logic for the front page
of the job printed some text (X-es?) across the paper-perforation
so that the individual jobs could quickly be detected visually if
you looked at the paper stack from the side.
"Carlos E. R." <robin_listas@es.invalid> writes:
On 2026-10-31 11:49, Nuno Silva wrote:
On 2026-10-31, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
Isn't that more about the viewer using to read the manual pages, than
a flaw of the online manual in itself?
No, man pages do not contain links to section of the document. It is
not structured.
Nuno is right; a viewer can include links. For an example see man7.org.
On 2026-10-31 19:56, Carlos E. R. wrote:
On 2026-10-31 16:57, Janis Papanagnou wrote:
On 2026-10-31 11:40, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
I agree with you on the _horrendous large_ 'man' "pages".
They shouldn't replace exhaustive and thorough documentation, IMO.
But worse I think are the ("maneuverable") 'info' pages.
Or the habit of some "Unixes" to have no man-page information but
only a stub with the hint to look into the (IMO horrendous) 'info'
pages. (I'm aware that some certainly prefer them; how else would
their proliferation be otherwise explainable.)
Concerning "clickable" information generally, mind the time when
man-pages were introduced!
Yes, I understand that. Isn't time we get something better?
But technically we have already "clickable" documents; with HTML for
example.
(Although it should still be available in one page, and ideally also searchable by regexps in easy ways. (As opposed to click on a search
icon, move the mouse to a new sub-window, (optionally) click another
time in that window to "activate" it for input, type in only a fixed non-regexp text (often without possibility to distinguish caps, or,
if you're lucky, only with an extra click in some control-box), and
hit enter (if you're lucky) or have to again move the mouse to click
on some Okay button.) - I certainly hate to have no '/search-pattern'
or '<Ctrl-F>search-pattern' available; with emphasis on "pattern".)
Personally I typically want (at least) three levels of information.
Terse usage information, complete interface description, and thorough explanations of all details in case of the more complex tools.
Mostly we (now) have all that information; if only through the Web.
(But often inconsistent or even unreliable.)
Should we have everything just in one (clickable) format like HTML?
I don't think so. Should we have all the spread existing information rewritten in some other format? Well, who would do that job?
What I (also) criticize is that we don't have a consistent level of information with some newer Linux platforms. I wished for the "second
level" of information that the man page format would have been kept/ maintained instead of trying to deprive that format in favor of the
'info' format.
Janis, preferring to avoid mouse-devices in the first place
Sure, point and click can also work without a mouse. Use the cursor to
select an entry in the table of contents, then hit enter.
Yes, sure. For long documents ("level three" information) with a lot
content where a "table of contents" is sensible that makes completely
sense. (Even pure mouse-control would be fine here [for me] in this
case.)
For the information level where I'd see the man-pages I wouldn't even
need a "table of contents", since that type of information shouldn't
be that voluminous, IMO.
On 2026-10-31, Carlos E. R. wrote:
On 2026-10-31 16:57, Janis Papanagnou wrote:
On 2026-10-31 11:40, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
I agree with you on the _horrendous large_ 'man' "pages".
They shouldn't replace exhaustive and thorough documentation, IMO.
But worse I think are the ("maneuverable") 'info' pages.
Or the habit of some "Unixes" to have no man-page information but
only a stub with the hint to look into the (IMO horrendous) 'info'
pages. (I'm aware that some certainly prefer them; how else would
their proliferation be otherwise explainable.)
Concerning "clickable" information generally, mind the time when
man-pages were introduced!
Yes, I understand that. Isn't time we get something better?
A question is how much of that is already possible using the current
format.
And another question is whether the current format, which could easily
be printed besides being installed as an online manual, does make for
better documentation to read than something focused on having more
internal references.
On Sat, 31 Oct 2026 11:40:12 +0100, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index.
Most man pages are not supposed to be that large. Of course, there
are some exceptions (like bash). For those I think yourCOll find the
man page was generated from some other, more navigable form anyway.
For example, if you look at Perl, you will see its documentation is
broken up across a whole series of man pages -- over 200 of them.
(Is that going too far in the other direction, do you think ... ?)
Tip: if you prefer another form, like PDF, itrCOs quite easy to generate
that from any man page. I used to use a somewhat convoluted command
for this, but it turns the usual Linux rCLmanrCY command has an option to directly choose the output format, e.g. view a man page directly in
typeset form without saving it to a file:
man -Tpdf 7 unix | okular - &
(Substitute your choice of favourite PDF reader.)
Info pages are somewhat better.
I have tried info pages a few times, but always found them awkward to navigate. IrCOve never heard of anybody who liked them.
On 10/31/26 17:11, Janis Papanagnou wrote:[...]
For the information level where I'd see the man-pages I wouldn't even
need a "table of contents", since that type of information shouldn't
be that voluminous, IMO.
-a But what if it BECAME so ?
-a Shit turns into a lot MORE shit.
-a My input to hypertext in the ancient days was the
-a idea of a sort of 'infinitely-indexed database'
-a where everything of importance somehow linked to
-a a huge tree of related info, which linked to a
-a huge tree of related info .....
-a Weren't as many 'computer geeks' back then.
-a They solved the index-space issue by using html
-a page links. The giant 'tree' wasn't in one place
-a any more. Not really possible before the early-80s
-a alas until DARPA-Net firmly became the Internet.
On 2026-10-31 22:11, Janis Papanagnou wrote:
On 2026-10-31 19:56, Carlos E. R. wrote:
On 2026-10-31 16:57, Janis Papanagnou wrote:
On 2026-10-31 11:40, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
I agree with you on the _horrendous large_ 'man' "pages".
They shouldn't replace exhaustive and thorough documentation, IMO.
But worse I think are the ("maneuverable") 'info' pages.
Or the habit of some "Unixes" to have no man-page information but
only a stub with the hint to look into the (IMO horrendous) 'info'
pages. (I'm aware that some certainly prefer them; how else would
their proliferation be otherwise explainable.)
Concerning "clickable" information generally, mind the time when
man-pages were introduced!
Yes, I understand that. Isn't time we get something better?
But technically we have already "clickable" documents; with HTML for
example.
Yes, and some tools use that. But then, a command like "apropos" does
not find that page.
(Although it should still be available in one page, and ideally also
searchable by regexps in easy ways. (As opposed to click on a search
icon, move the mouse to a new sub-window, (optionally) click another
time in that window to "activate" it for input, type in only a fixed
non-regexp text (often without possibility to distinguish caps, or,
if you're lucky, only with an extra click in some control-box), and
hit enter (if you're lucky) or have to again move the mouse to click
on some Okay button.) - I certainly hate to have no '/search-pattern'
or '<Ctrl-F>search-pattern' available; with emphasis on "pattern".)
Personally I typically want (at least) three levels of information.
Terse usage information, complete interface description, and thorough
explanations of all details in case of the more complex tools.
And an examples section. How to do this or that.
Mostly we (now) have all that information; if only through the Web.
(But often inconsistent or even unreliable.)
Should we have everything just in one (clickable) format like HTML?
I don't think so. Should we have all the spread existing information
rewritten in some other format? Well, who would do that job?
What I (also) criticize is that we don't have a consistent level of
information with some newer Linux platforms. I wished for the "second
level" of information that the man page format would have been kept/
maintained instead of trying to deprive that format in favor of the
'info' format.
Well, I find "info date" (for me "pinfo date") better for finding the information than "man date". Or "pinfo gcc".
[...]--- Synchronet 3.22a-Linux NewsLink 1.2
Frankly, I don't know the installation process of my Linux distro in
detail, but I can say that when I add a new tool-binary I instantly also
have the man page accessible, and I find the respective infos through 'apropos' as well. I don't see why there would be a principal hindrance
to do that for other information sources as well.
Daddy built it, daddy trashed it (Ken Olsen, he was responsible
for both).
On 2026-10-31, Carlos E. R. wrote:
On 2026-10-30 20:24, Charlie Gibbs wrote:
On 2026-10-30, c186282 <c186282@nnada.net> wrote:
More VOLUME of docs these days ... but "As Good" ???
Hardly. Most manuals back in the day had a section titled
"Theory of Operation". I really miss that.
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
Isn't that more about the viewer using to read the manual pages, than a
flaw of the online manual in itself?
On Sat, 31 Oct 2026 21:57:58 -0000 (UTC), Lawrence DrCOOliveiro wrote:
For example, if you look at Perl, you will see its documentation is
broken up across a whole series of man pages -- over 200 of them.
(Is that going too far in the other direction, do you think ... ?)
The record for the number of man pages has to be openssl.
Here is the list on my Gentoo system. Don't bother counting.
No, man pages do not contain links to section of the document. It is not structured.
But then, a command like "apropos" does not find that page.
On 2026-10-31 22:57, Lawrence DrCOOliveiro wrote:
Tip: if you prefer another form, like PDF, itrCOs quite easy to
generate that from any man page.
PDF is fine, but not if the source is a man page, because there are
no cross references that you can click or tap ...
... ditto table of contents and index.
c186282 <c186282@nnada.net> schrieb:
On 10/31/26 17:38, Lawrence DrCOOliveiro wrote:
On Sat, 31 Oct 2026 11:07:47 -0400, jayjwa wrote:
The DEC systems are more fun, have better docs, and much more
software.
If you look at all the vintage docs collected at Bitsavers, out of
(currently) about 660GB in total, over 40% is from DEC. Number 2 is
IBM, but the DEC collection is bigger by nearly 3:1.
Number 3 is CDC, at about a quarter of the size of the IBM collection.
The DEC boxes were GREAT !
Not entirely sure why they went out of biz - likely
idiot management. Daddy builds it, Junior TRASHES it.
Daddy built it, daddy trashed it (Ken Olsen, he was responsible
for both).
If you are interested, you could look up "DEC is Dead, Long Live
DEC", an excellent book.
Still wish someone would modernize VMS - it was a
fantastic/powerful system with a ton of prescient
modern features - could do international internet
before there WAS an internet.
Somebody is trying to do so (see https://vmssoftware.com/) but
that particular train has left the station a long time ago.
On 11/1/26 05:16, Thomas Koenig wrote:[...]
Daddy built it, daddy trashed it (Ken Olsen, he was responsible
for both).
Developed giant coke habit maybe ? ;-)
I have tried info pages a few times, but always found them awkward to navigate. IrCOve never heard of anybody who liked them.
c186282 <c186282@nnada.net> writes:
On 11/1/26 05:16, Thomas Koenig wrote:[...]
Daddy built it, daddy trashed it (Ken Olsen, he was responsible
for both).
Developed giant coke habit maybe ? ;-)
Do you have some basis for this? If it's a joke (yes, I see the
smiley), I fail to see the humor.
[...]
Umm ... NOT so unusual for the times :-)
Gotta wonder. Big powerful cutting-edge corp
with good history suddenly crashes and burns ......
According to c186282 <c186282@nnada.net>:
Umm ... NOT so unusual for the times
Gotta wonder. Big powerful cutting-edge corp with good history
suddenly crashes and burns ......
Sheesh, it's not like it's hard to find out what happened. The world of microprocessors was different from the world of minicomputers, DEC's management never figured that out, and its board of directors didn't
either.
On Mon, 2 Nov 2026 02:16:41 -0000 (UTC), John Levine wrote:
According to c186282 <c186282@nnada.net>:
Umm ... NOT so unusual for the times
Gotta wonder. Big powerful cutting-edge corp with good history
suddenly crashes and burns ......
Sheesh, it's not like it's hard to find out what happened. The world of
microprocessors was different from the world of minicomputers, DEC's
management never figured that out, and its board of directors didn't
either.
The Rainbow was interesting. Almost, but not quite, compatible with PCs.
The proprietary diskettes at $5 a pop didn't go over well either.
Olsen's belief that people didn't need or want a home computer wasn't
unique in that era. Ward Christensen of xmodem fame worked for IBM.
Famously he sent a memo up the chain of command saying IBM should look
into microcomputers. The reply was there was no market for them but if he wanted to screw around with them on his own time feel free.
We contracted to supply cable assemblies to DEC in Augusta ME in the '70s. Their QA was quite demanding but we could pass inspection. Then they gave
the contract to the Passamaquoddy tribe. I don't know how that turned
out.
According to c186282 <c186282@nnada.net>:
Umm ... NOT so unusual for the times :-)
Gotta wonder. Big powerful cutting-edge corp
with good history suddenly crashes and burns ......
Sheesh, it's not like it's hard to find out what happened. The world of microprocessors
was different from the world of minicomputers, DEC's management never figured that out,
and its board of directors didn't either.
Gordon Bell wrote this afterword for the "DEC is Dead" book which explains it well.
https://gordonbell.azurewebsites.net/CGB%20Files/DEC_Is_Dead_Bell_Appendix_Schein_Book.pdf
Windows 95 helped, but as far as I understand it, it really was meant
for corporate environments more than the home on the technical side; the
rest was just marketing.
[...]
-a WOULD love one or two of those big black IBM
-a mainframe units. Not sure what I'd DO with
-a them, but they'd be SO cool !
Anyway they do
-a have plenty of customers - govt/uni and big
-a industry. Sometimes you NEED a 'mainframe'.
[...]
On 11/2/2026 12:48 PM, rbowman wrote:
On Mon, 2 Nov 2026 02:16:41 -0000 (UTC), John Levine wrote:
According to c186282-a <c186282@nnada.net>:
-a-a Umm ... NOT so unusual for the times
-a-a Gotta wonder. Big powerful cutting-edge corp with good history
-a-a suddenly crashes and burns ......
Sheesh, it's not like it's hard to find out what happened.-a The world of >>> microprocessors was different from the world of minicomputers, DEC's
management never figured that out, and its board of directors didn't
either.
The Rainbow was interesting. Almost, but not quite, compatible with PCs.
The proprietary diskettes at $5 a pop didn't go over well either.
Olsen's belief that people didn't need or want a home computer wasn't
unique in that era. Ward Christensen of xmodem fame worked for IBM.
Famously he sent a memo up the chain of command saying IBM should look
into microcomputers. The reply was there was no market for them but if he
wanted to screw around with them on his own time feel free.
We contracted to supply cable assemblies to DEC in Augusta ME in the
'70s.
Their QA was quite demanding but we could pass inspection. Then they gave
the contract to the Passamaquoddy tribe. I don't know how that turned
out.
Nobody can prevent someone else from inventing something that changes
the business rules from under you.-a I'm only theorizing here, but I
would not be surprised [1] if DOOM sold a lot of home PCs in the early
90s, then Quake even more.
Windows 95 helped, but as far as I understand it, it really was meant
for corporate environments more than the home on the technical side; the
rest was just marketing.
The social phenomenon of people building their own home computers, then gathering together to connect them with a LAN, all in order to play DOOM shouldn't be underestimated.
Side note: there's a marketing book called /Guerilla Marketing/ that
touches upon this subject.
Best wishes, and happy victories in DOOM!
[1] And indeed would be surprised it wasn't a significant factor.
Not entirely sure why they went out of biz - likely idiot
management. Daddy builds it, Junior TRASHES it.
Daddy built it, daddy trashed it (Ken Olsen, he was responsible for
both).
Developed giant coke habit maybe ? Efye
c186282 <c186282@nnada.net> writes:
On 11/1/26 05:16, Thomas Koenig wrote:[...]
Daddy built it, daddy trashed it (Ken Olsen, he was responsible
for both).
Developed giant coke habit maybe ? ;-)
Do you have some basis for this? If it's a joke (yes, I see the
smiley), I fail to see the humor.
[...]
I can believe "board of directors" ... but not so much the hands-on
people at DEC.
On 2026-11-02 07:28, c186282 wrote:
[...]
-a-a WOULD love one or two of those big black IBM
-a-a mainframe units. Not sure what I'd DO with
-a-a them, but they'd be SO cool !
:-)
There's a couple hardware types I'd have liked to have -
and I'd also not know what to do with them but to wallow
in memories.
-a-a Anyway they do
-a-a have plenty of customers - govt/uni and big
-a-a industry. Sometimes you NEED a 'mainframe'.
Hmm.. - "need" - really?
During the 1990's we switched our big telcos from mainframes
to Unix servers. The huge sports halls where the mainframes
had been running were almost empty then. It was just more
economic to run Unix servers (mainframe support was costly,
as was just running them), also more ecologic (though I'm
not sure that this mattered that much in practice, but costs
of electricity might have been an issue), and the much less
required physical space as said. Basically all cost factors.
Also computing centers switched. Our LRZ, formerly running
a CDC 175, is meanwhile running various Unix clusters for
"super computing". Our university also said goodbye to its
TR440 and is providing mainly Linux systems to its students
for the scientific oriented departments.
The only "need" for a mainframe I observed in some financial
business area, where they were still running their legacy
software on mainframes obviously with "no chance" to migrate,
or not wanting to take the risk and cost (as long as the old
machinery is doing its task). But it's also difficult to get
spares, and to find people who can and are willing to operate
them. (But that information is now a bit dated and things may
have meanwhile changed even in that business area.)
Lawrence DrCOOliveiro <ldo@nz.invalid> writes:
[...]
I have tried info pages a few times, but always found them awkward to
navigate. IrCOve never heard of anybody who liked them.
You have now: I do. I'm not sure what the objection is.
On Mon, 2 Nov 2026 14:13:06 +0800, Johann 'Myrkraverk' Oskarsson wrote:
Windows 95 helped, but as far as I understand it, it really was meant
for corporate environments more than the home on the technical side; the
rest was just marketing.
Enterprises use NT. It wasn't until Windows XP that the the consumer and business branches were merged. NT 4.0 used the Windows 95 UX but Windows
2000 brought in more features from the consumer lines. I don't think 2000
was marketed in the consumer market. The lucky consumers got Windows Me as
a consolation prize since the XP project was moving slowly.
NT 4.0 was the last release to be called NT but the numbering scheme was carried on so Windows 11 is NT 10.0 formally. It might have been my
machine but NT 4.0 usually greeted my with the BSOD on Monday morning.
2000 was an improvement.
The social phenomenon of people building their own home computers, then
gathering together to connect them with a LAN, all in order to play DOOM
shouldn't be underestimated.
-a It was an important milestone - and boosted
-a the development of a lot of shit. "Games"
-a are, unto themselves, trivial. However the
-a tech/knowledge to do them well drove a lot
-a of tech. A *large* market segment demanded
-a the 'better game computer', so ....
-a NVidia came out of that. Now it's a zillion
-a dollar concern smack in the middle of 'AI'.
On 02/11/2026 08:41, Janis Papanagnou wrote:
On 2026-11-02 07:28, c186282 wrote:Yes. The transition from mainframe, to Unix clusters and then to blade servers shrunk machine rooms to almost nothing, amnd high speed internet moved them to cloud centres
[...]
-a-a WOULD love one or two of those big black IBM
-a-a mainframe units. Not sure what I'd DO with
-a-a them, but they'd be SO cool !
:-)
There's a couple hardware types I'd have liked to have -
and I'd also not know what to do with them but to wallow
in memories.
-a-a Anyway they do
-a-a have plenty of customers - govt/uni and big
-a-a industry. Sometimes you NEED a 'mainframe'.
Hmm.. - "need" - really?
During the 1990's we switched our big telcos from mainframes
to Unix servers. The huge sports halls where the mainframes
had been running were almost empty then. It was just more
economic to run Unix servers (mainframe support was costly,
as was just running them), also more ecologic (though I'm
not sure that this mattered that much in practice, but costs
of electricity might have been an issue), and the much less
required physical space as said. Basically all cost factors.
Also computing centers switched. Our LRZ, formerly running
a CDC 175, is meanwhile running various Unix clusters for
"super computing". Our university also said goodbye to its
TR440 and is providing mainly Linux systems to its students
for the scientific oriented departments.
The only "need" for a mainframe I observed in some financialThat's not quite true. First of all old IBM COBOL and RPG is well
business area, where they were still running their legacy
software on mainframes obviously with "no chance" to migrate,
or not wanting to take the risk and cost (as long as the old
machinery is doing its task). But it's also difficult to get
spares, and to find people who can and are willing to operate
them. (But that information is now a bit dated and things may
have meanwhile changed even in that business area.)
supported on modern -|P based hardware. and second mainframes are still
in use where enormous numbers of I/O operations are called for.
"Mainframes remain the invisible backbone of the global economy in the
21st century, handling 90% of credit card transactions, powering 92 of
the top 100 banks, and managing 68% of production IT workloads globally while accounting for just a fraction of total IT spending."
Mainframes remain cost effective alternatives for very large scale low complexcity transactions
Only SUN really survived. And where are they now?
On 11/2/2026 4:56 PM, c186282 wrote:
The social phenomenon of people building their own home computers, then
gathering together to connect them with a LAN, all in order to play DOOM >>> shouldn't be underestimated.
-a-a It was an important milestone - and boosted
-a-a the development of a lot of shit. "Games"
-a-a are, unto themselves, trivial. However the
-a-a tech/knowledge to do them well drove a lot
-a-a of tech. A *large* market segment demanded
-a-a the 'better game computer', so ....
-a-a NVidia came out of that. Now it's a zillion
-a-a dollar concern smack in the middle of 'AI'.
And now people are dreaming NVidia will disappear, the way of DEC.-a Well maybe it /will/ be sold to HP?
After all, gamers want their affordable graphics cards back.
So if people think mainframes are somehow /obsolete/, it's because they
are extrapolating from their own personal experiences and never look un-
der the hood of the commercial world.
On 11/2/2026 6:47 PM, The Natural Philosopher wrote:
Only SUN really survived. And where are they now?
Well, we can say that Solaris got forked.-a The enterprise behemoth got
the trade mark, the open source edition got the people.
So Illumos is still around, and doing well all things considered.
I personally wish them well.
On 02/11/2026 00:45, Keith Thompson wrote:
c186282 <c186282@nnada.net> writes:During the 90s and noughties, that was the hallmark of a successful >professional [politician]
On 11/1/26 05:16, Thomas Koenig wrote:[...]
Daddy built it, daddy trashed it (Ken Olsen, he was responsible
for both).
Developed giant coke habit maybe ? ;-)
Do you have some basis for this? If it's a joke (yes, I see the
smiley), I fail to see the humor.
NT 4.0 was the last release to be called NT but the numbering scheme was carried on so Windows 11 is NT 10.0 formally. It might have been my
machine but NT 4.0 usually greeted my with the BSOD on Monday morning.
2000 was an improvement.
On 2026-11-02 07:28, c186282 wrote:
[...]
-a WOULD love one or two of those big black IBM
-a mainframe units. Not sure what I'd DO with
-a them, but they'd be SO cool !
:-)
There's a couple hardware types I'd have liked to have -
and I'd also not know what to do with them but to wallow
in memories.
On 11/1/26 21:16, John Levine wrote:
According to c186282 <c186282@nnada.net>:
Umm ... NOT so unusual for the times :-)
Gotta wonder. Big powerful cutting-edge corp
with good history suddenly crashes and burns ......
Sheesh, it's not like it's hard to find out what happened. The world of microprocessors
was different from the world of minicomputers, DEC's management never figured that out,
and its board of directors didn't either.
I can believe "board of directors" ... but not
so much the hands-on people at DEC. ...
Gordon Bell wrote this afterword for the "DEC is Dead" book which explains it well.
https://gordonbell.azurewebsites.net/CGB%20Files/DEC_Is_Dead_Bell_Appendix_Schein_Book.pdf
One guy's opinions, probably had a grudge. As usual,
it's gonna be more complicated, subtle, than that.
Yeah, the impression I've gotten is that Olsen was less the kind of Bad Management that drove e.g. Commodore into the ground and more just theDaddy built it, daddy trashed it (Ken Olsen, he was responsible
for both).
If you talk to the staff who worked under him, you wonrCOt find anybody
with a bad word to say about him. He looked after his people well.
But, he was very much a Unix-hater, describing it as rCLsnake oilrCY at
one point.
After he was ousted, the asset-strippers took charge.
The rise of LSI in the 70s made minicomputers possible. The rise of mass produced microprocessors in the 1980s made them irrelevant.
That's probably true, but I'll point out that not all /businesses/ are
at the enterprise level. Lots and lots of small to medium businesses
had DOS, then Windows machines in the 90s.
And now people are dreaming NVidia will disappear, the way of DEC. Well maybe it /will/ be sold to HP?
After all, gamers want their affordable graphics cards back.
Since DDEML proved unusable for interprocess communication, I tried
various other methods, from shared memory to WM_COPYDATA messages.
Eventually I went to sockets, where I happily am today. (I love being
able to test my programs with telnet.)
And on top of that, the Rainbow was a dud: more expensive than the IBM
PC for similar specs with only somewhat more flexible graphics and CP/M compatibility as its advantages. The Professional at least offered one
of DEC's own architecture, but it was over *twice* the price of the PC.
On 2026-10-31, Nuno Silva <nunojsilva@invalid.invalid> wrote:
On 2026-10-31, Carlos E. R. wrote:
On 2026-10-30 20:24, Charlie Gibbs wrote:
On 2026-10-30, c186282 <c186282@nnada.net> wrote:
More VOLUME of docs these days ... but "As Good" ???
Hardly. Most manuals back in the day had a section titled
"Theory of Operation". I really miss that.
Talking of manuals, I find large man pages very difficult to use,
specially when you need something specific. No clickable table of
content, no clickable index. Info pages are somewhat better.
Isn't that more about the viewer using to read the manual pages, than a
flaw of the online manual in itself?
At least the man command can search a man page for a text string.
It's not perfect, but it's better than nothing.
On 2026-11-01 12:59, Carlos E. R. wrote:
On 2026-10-31 22:11, Janis Papanagnou wrote:
On 2026-10-31 19:56, Carlos E. R. wrote:
On 2026-10-31 16:57, Janis Papanagnou wrote:
On 2026-10-31 11:40, Carlos E. R. wrote:
Talking of manuals, I find large man pages very difficult to use, >>>>>> specially when you need something specific. No clickable table of >>>>>> content, no clickable index. Info pages are somewhat better.
I agree with you on the _horrendous large_ 'man' "pages".
They shouldn't replace exhaustive and thorough documentation, IMO.
But worse I think are the ("maneuverable") 'info' pages.
Or the habit of some "Unixes" to have no man-page information but
only a stub with the hint to look into the (IMO horrendous) 'info'
pages. (I'm aware that some certainly prefer them; how else would
their proliferation be otherwise explainable.)
Concerning "clickable" information generally, mind the time when
man-pages were introduced!
Yes, I understand that. Isn't time we get something better?
But technically we have already "clickable" documents; with HTML for
example.
Yes, and some tools use that. But then, a command like "apropos" does
not find that page.
Frankly, I don't know the installation process of my Linux distro in
detail, but I can say that when I add a new tool-binary I instantly
also have the man page accessible, and I find the respective infos
through 'apropos' as well. I don't see why there would be a principal hindrance to do that for other information sources as well.
(Although it should still be available in one page, and ideally also
searchable by regexps in easy ways. (As opposed to click on a search
icon, move the mouse to a new sub-window, (optionally) click another
time in that window to "activate" it for input, type in only a fixed
non-regexp text (often without possibility to distinguish caps, or,
if you're lucky, only with an extra click in some control-box), and
hit enter (if you're lucky) or have to again move the mouse to click
on some Okay button.) - I certainly hate to have no '/search-pattern'
or '<Ctrl-F>search-pattern' available; with emphasis on "pattern".)
Personally I typically want (at least) three levels of information.
Terse usage information, complete interface description, and thorough
explanations of all details in case of the more complex tools.
And an examples section. How to do this or that.
This is often useful. For powerful large tools this might exceed
what one likes in a man-page, though, and would just contribute
to making the man-page "unreadable" due to its length. Here again
I think that the type of the examples is the discriminating info;
the man-page should certainly contain the technical options and
examples how to call the tool with those, but (having tools like
Imagemagick in mind) not the functional usage examples that would
better be placed in a tutorial type of documentation.
--
Mostly we (now) have all that information; if only through the Web.
(But often inconsistent or even unreliable.)
Should we have everything just in one (clickable) format like HTML?
I don't think so. Should we have all the spread existing information
rewritten in some other format? Well, who would do that job?
What I (also) criticize is that we don't have a consistent level of
information with some newer Linux platforms. I wished for the "second
level" of information that the man page format would have been kept/
maintained instead of trying to deprive that format in favor of the
'info' format.
Well, I find "info date" (for me "pinfo date") better for finding the
information than "man date". Or "pinfo gcc".
One problem I have with 'info' is that I see only a paragraph of
a subsection of a section of the described thing. I certainly want
to see more context! I also like to have the typical chapters that
describe the tools' _interfaces_ (the name with terse description,
the usage synopsis, a description, the list of options, returned
status codes, files that affect the tool in some way, example uses,
and know bugs, for example).
The 'date' manual you gave as example is certainly not one of the pathologically huge man-pages (with its 200 lines, and most of it
is anyway just a table of the format specifiers).
Janis
[...]
On Sun, 1 Nov 2026 13:07:09 +0100, Carlos E. R. wrote:
On 2026-10-31 22:57, Lawrence DrCOOliveiro wrote:
Tip: if you prefer another form, like PDF, itrCOs quite easy to
generate that from any man page.
PDF is fine, but not if the source is a man page, because there are
no cross references that you can click or tap ...
Actually, you *can* include URLs in the source markup, and the PDF
version will make them clickable.
--... ditto table of contents and index.
Yeah, theyrCyre not (normally) supposed to be long enough to need that
...
On 02/11/2026 08:41, Janis Papanagnou wrote:
On 2026-11-02 07:28, c186282 wrote:Yes. The transition from mainframe, to Unix clusters and then to blade servers shrunk machine rooms to almost nothing, amnd high speed internet moved them to cloud centres
[...]
-a-a WOULD love one or two of those big black IBM
-a-a mainframe units. Not sure what I'd DO with
-a-a them, but they'd be SO cool !
:-)
There's a couple hardware types I'd have liked to have -
and I'd also not know what to do with them but to wallow
in memories.
-a-a Anyway they do
-a-a have plenty of customers - govt/uni and big
-a-a industry. Sometimes you NEED a 'mainframe'.
Hmm.. - "need" - really?
During the 1990's we switched our big telcos from mainframes
to Unix servers. The huge sports halls where the mainframes
had been running were almost empty then. It was just more
economic to run Unix servers (mainframe support was costly,
as was just running them), also more ecologic (though I'm
not sure that this mattered that much in practice, but costs
of electricity might have been an issue), and the much less
required physical space as said. Basically all cost factors.
Also computing centers switched. Our LRZ, formerly running
a CDC 175, is meanwhile running various Unix clusters for
"super computing". Our university also said goodbye to its
TR440 and is providing mainly Linux systems to its students
for the scientific oriented departments.
The only "need" for a mainframe I observed in some financialThat's not quite true.
business area, where they were still running their legacy
software on mainframes obviously with "no chance" to migrate,
or not wanting to take the risk and cost (as long as the old
machinery is doing its task). But it's also difficult to get
spares, and to find people who can and are willing to operate
them. (But that information is now a bit dated and things may
have meanwhile changed even in that business area.)
[...]
On 02/11/2026 12:11, Johann 'Myrkraverk' Oskarsson wrote:
So if people think mainframes are somehow /obsolete/, it's because they
are extrapolating from their own personal experiences and never look un-
der the hood of the commercial world.
[...]
On 2026-11-01 23:32, Lawrence DrCOOliveiro wrote:
Actually, you *can* include URLs in the source markup, and the PDF
version will make them clickable.
Yes, that happens when the source is used to create several output
formats. Tpically both man and info, but there can be more.
But not when the pdf is generated using the man as source.
Sometimes a tool has a small man page, then a large structured document
in some other format.
On 2026-11-02 23:32, Carlos E. R. wrote:
Sometimes a tool has a small man page, then a large structured
document in some other format.
I think that's exactly how it should be.
On Mon, 2 Nov 2026 23:35:43 +0100, Carlos E. R. wrote:
On 2026-11-01 23:32, Lawrence DrCOOliveiro wrote:
Actually, you *can* include URLs in the source markup, and the PDF
version will make them clickable.
Yes, that happens when the source is used to create several output
formats. Tpically both man and info, but there can be more.
But not when the pdf is generated using the man as source.
IrCOm not sure what you mean by rCLman as sourcerCY, if not the troff/groff markup. That *is* the required source form for man pages, after all.
On 2026-11-02 12:22, The Natural Philosopher wrote:
On 02/11/2026 08:41, Janis Papanagnou wrote:
On 2026-11-02 07:28, c186282 wrote:Yes. The transition from mainframe, to Unix clusters and then to blade
[...]
-a-a WOULD love one or two of those big black IBM
-a-a mainframe units. Not sure what I'd DO with
-a-a them, but they'd be SO cool !
:-)
There's a couple hardware types I'd have liked to have -
and I'd also not know what to do with them but to wallow
in memories.
-a-a Anyway they do
-a-a have plenty of customers - govt/uni and big
-a-a industry. Sometimes you NEED a 'mainframe'.
Hmm.. - "need" - really?
During the 1990's we switched our big telcos from mainframes
to Unix servers. The huge sports halls where the mainframes
had been running were almost empty then. It was just more
economic to run Unix servers (mainframe support was costly,
as was just running them), also more ecologic (though I'm
not sure that this mattered that much in practice, but costs
of electricity might have been an issue), and the much less
required physical space as said. Basically all cost factors.
Also computing centers switched. Our LRZ, formerly running
a CDC 175, is meanwhile running various Unix clusters for
"super computing". Our university also said goodbye to its
TR440 and is providing mainly Linux systems to its students
for the scientific oriented departments.
servers shrunk machine rooms to almost nothing, amnd high speed
internet moved them to cloud centres
The only "need" for a mainframe I observed in some financialThat's not quite true.
business area, where they were still running their legacy
software on mainframes obviously with "no chance" to migrate,
or not wanting to take the risk and cost (as long as the old
machinery is doing its task). But it's also difficult to get
spares, and to find people who can and are willing to operate
them. (But that information is now a bit dated and things may
have meanwhile changed even in that business area.)
What specifically are you referring to here?
That I observed mainframes still in use in commercial contexts?
Or my speculation that this might meanwhile have changed?
Or that the reason for not having done a change was something else?
I've worked in several business areas were mainframes had been used.
* The computing centers I was aware of changed their architectures
-a from mainframe architectures to clusters.
* The telcos down-scaled the [amount of] metal, yet increased their
-a throughput by switching away from mainframes. They made the change
-a because they also needed a lot contemporary new functions supported
-a and it was far too costly and much more effort to do that on legacy
-a mainframe hardware.
* The business area I worked in where there were still running some
-a mainframes did so primarily because of the mentioned reasons. That
-a was 20 years ago, so that's the reason why I said that things "may"
-a have meanwhile changed. If they need to runs (similar to the telcos)
-a significant amounts of new functions they have to do the calculation
-a whether it pays to do that on the old systems. And they have to take
-a into consideration whether they have (or can find) programmers for
-a old legacy languages as far as the legacy software is concerned; two
-a decades ago we didn't find any COBOL programmers for those tasks and
-a I'd expect that the situation got even worse meanwhile.
* Most (all?) internet providers I've heard of aren't using mainframes
-a but built their systems architecture based on (often Linux-)clusters.
There must be reasons to change a running mainframe system. As said,
such a transition process can be risky and is costly (without instant
gain of any new functions). The reasons I've heard from the financial sectors, i.e. not to switch, were exactly those.
I'm also a bit astonished about your mention of using mainframes for efficient I/O. Networking clusters appear to me to be much better able
than a single mainframe instance. The older mainframe architectures we
used also had a lot of peripheral processors to specifically handle
the I/O.
(But of course I can, as I did, just report about the (certainly non- singular) cases I was involved in or heard from reliable sources.)
Janis
[...]
Heck, Borland in the 90's had an IDE with integrated help that could
find the proper help paragraph by clicking on a keyword in your text.
Seems too difficult to replicate. Not even Lazarus has it.
On Mon, 2 Nov 2026 23:35:43 +0100, Carlos E. R. wrote:
On 2026-11-01 23:32, Lawrence DrCOOliveiro wrote:
Actually, you *can* include URLs in the source markup, and the PDF
version will make them clickable.
Yes, that happens when the source is used to create several output
formats. Tpically both man and info, but there can be more.
But not when the pdf is generated using the man as source.
IrCOm not sure what you mean by rCLman as sourcerCY, if not the troff/groff markup. That *is* the required source form for man pages, after all.
I'm also a bit astonished about your mention of using mainframes for efficient I/O. Networking clusters appear to me to be much better able
than a single mainframe instance. The older mainframe architectures we
used also had a lot of peripheral processors to specifically handle
the I/O.
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
On 2026-11-03 00:26, Janis Papanagnou wrote:
On 2026-11-02 23:32, Carlos E. R. wrote:
Sometimes a tool has a small man page, then a large structured
document in some other format.
I think that's exactly how it should be.
I would prefer a properly indexed global system that can find me
keywords and things across the system, and each document properly
indexed and with proper tables of contents, all linked and browsable as
a modern computer in the XXI can do, not the equivalent of printed paper.
Heck, Borland in the 90's had an IDE with integrated help that could
find the proper help paragraph by clicking on a keyword in your text.
Seems too difficult to replicate. Not even Lazarus has it.
I have seen documentation PDFs that are just the output of man2pdf.
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
On 2026-11-03 00:26, Janis Papanagnou wrote:
On 2026-11-02 23:32, Carlos E. R. wrote:
Sometimes a tool has a small man page, then a large structured
document in some other format.
I think that's exactly how it should be.
I would prefer a properly indexed global system that can find me
keywords and things across the system, and each document properly
indexed and with proper tables of contents, all linked and browsable as
a modern computer in the XXI can do, not the equivalent of printed paper.
Heck, Borland in the 90's had an IDE with integrated help that could
find the proper help paragraph by clicking on a keyword in your text.
Seems too difficult to replicate. Not even Lazarus has it.
On 2026-11-02, Janis Papanagnou <janis_papanagnou+ng@hotmail.com> wrote:
I'm also a bit astonished about your mention of using mainframes for
efficient I/O. Networking clusters appear to me to be much better able
than a single mainframe instance. The older mainframe architectures we
used also had a lot of peripheral processors to specifically handle
the I/O.
True, but back in 1970 there wasn't a lot of networking happening,
largely because people hadn't really figured it out yet. Mainframe
I/O capacity was prodigious at the time - early minis could hardly
manage the megabyte-per-second speeds that mainframes dealt with
routinely. Consider that while the mini people were making a big
thing of DMA, the term was unheard of in the mainframe world -
mainframers just took it for granted.
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
Yes, there is a dichotomy there. Mainframe protocols were designed
for sending files across a continent, while mini protocols were
designed for sending characters across the room. Compromises were
made on both sides.
I would prefer a properly indexed global system that can find me
keywords and things across the system, and each document properly
indexed and with proper tables of contents, all linked and browsable
as a modern computer in the XXI can do, not the equivalent of
printed paper.
Heck, Borland in the 90's had an IDE with integrated help that could
find the proper help paragraph by clicking on a keyword in your
text. Seems too difficult to replicate. Not even Lazarus has it.
According to c186282 <c186282@nnada.net>:
Umm ... NOT so unusual for the times :-)
Gotta wonder. Big powerful cutting-edge corp
with good history suddenly crashes and burns ......
Sheesh, it's not like it's hard to find out what happened. The world of microprocessors
was different from the world of minicomputers, DEC's management never figured that out,
and its board of directors didn't either.
Gordon Bell wrote this afterword for the "DEC is Dead" book which explains it well.
https://gordonbell.azurewebsites.net/CGB%20Files/DEC_Is_Dead_Bell_Appendix_Schein_Book.pdf
On 11/2/26 21:49, Lawrence DrCOOliveiro wrote:
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous
controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
Yep ! Even the old mainframes were "smart" - lots
of optimizations that only later migrated down.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
System interrupts PER-KEYSTROKE ??? Horrible !
The "minis" were pretty good all in all, had a
definite 'place'. Lots made. Had some experience
with the TI-990 minis. Super-funky hardware-
based multi-user scheme with RAM-based registers.
Hey, good at the time !
On Tue, 03 Nov 2026 02:53:45 GMT, Charlie Gibbs wrote:
Yes, there is a dichotomy there. Mainframe protocols were designed
for sending files across a continent, while mini protocols were
designed for sending characters across the room. Compromises were
made on both sides.
ThatrCOs not the dichotomy. The dichotomy was that mainframes were
optimized for throughput at the expense of latency, while
minicomputers concentrated on minimizing latency to maximize
interactive responsiveness at the expense of throughput.
On 2026-11-03 00:26, Janis Papanagnou wrote:
On 2026-11-02 23:32, Carlos E. R. wrote:
Sometimes a tool has a small man page, then a large structuredI think that's exactly how it should be.
document in some other format.
I would prefer a properly indexed global system that can find me
keywords and things across the system, and each document properly
indexed and with proper tables of contents, all linked and browsable
as a modern computer in the XXI can do, not the equivalent of printed
paper.
On 2026-11-01 23:32, Lawrence DrCOOliveiro wrote:
On Sun, 1 Nov 2026 13:07:09 +0100, Carlos E. R. wrote:
On 2026-10-31 22:57, Lawrence DrCOOliveiro wrote:Actually, you *can* include URLs in the source markup, and the PDF
Tip: if you prefer another form, like PDF, itrCOs quite easy to
generate that from any man page.
PDF is fine, but not if the source is a man page, because there are
no cross references that you can click or tap ...
version will make them clickable.
Yes, that happens when the source is used to create several output
formats. Tpically both man and info, but there can be more.
But not when the pdf is generated using the man as source.
On Tue, 03 Nov 2026 02:53:45 GMT, Charlie Gibbs wrote:
Yes, there is a dichotomy there. Mainframe protocols were designed
for sending files across a continent, while mini protocols were
designed for sending characters across the room. Compromises were
made on both sides.
ThatrCOs not the dichotomy. The dichotomy was that mainframes were
optimized for throughput at the expense of latency, while
minicomputers concentrated on minimizing latency to maximize
interactive responsiveness at the expense of throughput.
On 11/2/26 20:57, Carlos E. R. wrote:
On 2026-11-03 00:26, Janis Papanagnou wrote:
On 2026-11-02 23:32, Carlos E. R. wrote:
Sometimes a tool has a small man page, then a large structured
document in some other format.
I think that's exactly how it should be.
I would prefer a properly indexed global system that can find me
keywords and things across the system, and each document properly
indexed and with proper tables of contents, all linked and browsable
as a modern computer in the XXI can do, not the equivalent of printed
paper.
-a I generally agree about how it SHOULD be ... but
-a we're not seeing it and thus I don't think we *will*.
-a Man pages are a 'specialty' kind of doc.
-a Do NOT find them very useful these days.
Heck, Borland in the 90's had an IDE with integrated help that could
find the proper help paragraph by clicking on a keyword in your text.
Seems too difficult to replicate. Not even Lazarus has it.
-a Borland was ahead of its time. Alas few others
-a kept up. Phillipe Kahn was a god !
-a Ah, Turbo Pascal ... bought it CHEAP, with a
-a sent-in paper check initially. Completely
-a CHANGED the development experience ! Do
-a have a few of the old multi-pass M$/IBM
-a compilers in a DOS VM somewhere, even
-a fuck with 'em once in awhile for fun.
-a But for DOING anything - no no no !!!
-a Still strongly pref the Gen-3 development
-a IDEs over anything allegedly "better".
-a Straight-up, no bullshit, everything you
-a need to get it done fast and right.
-a Apparently I'm on of the few who still write
-a Pascal. Love it ! The Lazarus/FPC environment
-a suits me just FINE. Proto in Python, then, if
-a desired, re-write in Pascal - tighter, prettier,
-a much faster.
On Tue, 3 Nov 2026 02:57:27 +0100, Carlos E. R. wrote:
I would prefer a properly indexed global system that can find me
keywords and things across the system, and each document properly
indexed and with proper tables of contents, all linked and browsable
as a modern computer in the XXI can do, not the equivalent of
printed paper.
Given yourCOre basically talking about a search through /usr/share/doc,
that isnrCOt actually that hard to do.
Heck, Borland in the 90's had an IDE with integrated help that could
find the proper help paragraph by clicking on a keyword in your
text. Seems too difficult to replicate. Not even Lazarus has it.
Help on what, though? Just programming APIs? Or other system stuff as
well?
MS-DOS never had an equivalent to /usr/share/doc. Nor does present-day Windows.--
On Tue, 3 Nov 2026 03:02:07 +0100, Carlos E. R. wrote:
I have seen documentation PDFs that are just the output of man2pdf.
Not sure why you need a separate package/command, when the usual rCLmanrCY command is quite capable of doing that itself.
"Carlos E. R." <robin_listas@es.invalid> writes:
On 2026-11-01 23:32, Lawrence DrCOOliveiro wrote:
On Sun, 1 Nov 2026 13:07:09 +0100, Carlos E. R. wrote:
On 2026-10-31 22:57, Lawrence DrCOOliveiro wrote:Actually, you *can* include URLs in the source markup, and the PDF
Tip: if you prefer another form, like PDF, itrCOs quite easy to
generate that from any man page.
PDF is fine, but not if the source is a man page, because there are
no cross references that you can click or tap ...
version will make them clickable.
Yes, that happens when the source is used to create several output
formats. Tpically both man and info, but there can be more.
But not when the pdf is generated using the man as source.
Links work fine in PDFs generated from man page sources. See
link at the bottom of https://terraraq.uk/junk/ccache.pdf to https://ccache.dev/credits.html for an example.
-a WOULD love one or two of those big black IBM
-a mainframe units. Not sure what I'd DO with
-a them, but they'd be SO cool ! Anyway they do
-a have plenty of customers - govt/uni and big
-a industry. Sometimes you NEED a 'mainframe'.
On 11/2/2026 7:22 PM, The Natural Philosopher wrote:
On 02/11/2026 08:41, Janis Papanagnou wrote:
On 2026-11-02 07:28, c186282 wrote:
That's not quite true. First of all old IBM COBOL and RPG is well
supported on modern -|P based hardware. and second mainframes are still
in use where enormous numbers of I/O operations are called for.
"Mainframes remain the invisible backbone of the global economy in the
21st century, handling 90% of credit card transactions, powering 92 of
the top 100 banks, and managing 68% of production IT workloads
globally while accounting for just a fraction of total IT spending."
Mainframes remain cost effective alternatives for very large scale low
complexcity transactions
I've heard -- let's say verbally -- that the relentless transaction pro- cessing done on mainframes is hard to do on different types of machines.
And what is hard?-a I haven't worked with mainframes, but last time I
looked into it, POWER was the only CPU with hardware accelerated decimal arithmetic [1].-a This is the kind of computation universities don't do
as research, so -- based on my stories of academia -- there are none
there for budding CS doctorates to tinker with.
So if people think mainframes are somehow /obsolete/, it's because they
are extrapolating from their own personal experiences and never look un-
der the hood of the commercial world.
Best wishes, and happy mainframes!
[1] I don't have a POWER CPU, and didn't personally verify this.--
I know at least one guy working on IBM mainframes, and he tells me they
are in high need of good assembler programmers.
They need high i/o with very large databases.
What it is exactly they do, I don't know.
On 11/2/26 21:49, Lawrence DrCOOliveiro wrote:
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous
controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
-a Yep ! Even the old mainframes were "smart" - lots
-a of optimizations that only later migrated down.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
-a System interrupts PER-KEYSTROKE ??? Horrible !
On 11/3/26 11:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
What it is exactly they do, I don't know.
I'm too young for mainframes, but I always assumed they were for running legacy systems that people couldn't be bothered to port to new hardware/ architectures.
FWIW, my guess is that "high i/o" and "very large" is only by the
standards of mainframe programmers.
On 2026-11-03 12:49, Pancho wrote:
On 11/3/26 11:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
What it is exactly they do, I don't know.
I'm too young for mainframes, but I always assumed they were for
running legacy systems that people couldn't be bothered to port to new
hardware/ architectures.
FWIW, my guess is that "high i/o" and "very large" is only by the
standards of mainframe programmers.
No, it is really high i/o and very large databases. I saw some figures
once and they were impressive.
On 11/2/26 21:49, Lawrence DrCOOliveiro wrote:
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous
controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
-a Yep ! Even the old mainframes were "smart" - lots
-a of optimizations that only later migrated down.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
-a System interrupts PER-KEYSTROKE ??? Horrible !
-a The "minis" were pretty good all in all, had a
-a definite 'place'. Lots made. Had some experience
-a with the TI-990 minis. Super-funky hardware-
-a based multi-user scheme with RAM-based registers.
-a Hey, good at the time !
-a I guess you COULD call the PDP-11 a "mini", but
-a imho it didn't have quite the "feel". More like
-a a cut-down mainframe. Good peripheral bus though,
-a you could plug in all kinds of goodies.
On 11/3/26 11:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
What it is exactly they do, I don't know.
I'm too young for mainframes, but I always assumed they were for running legacy systems that people couldn't be bothered to port to new hardware/architectures.
FWIW, my guess is that "high i/o" and "very large" is only by the
standards of mainframe programmers.
On 2026-11-03 12:49, Pancho wrote:
On 11/3/26 11:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
What it is exactly they do, I don't know.
I'm too young for mainframes, but I always assumed they were for
running legacy systems that people couldn't be bothered to port to new
hardware/ architectures.
FWIW, my guess is that "high i/o" and "very large" is only by the
standards of mainframe programmers.
No, it is really high i/o and very large databases. I saw some figures
once and they were impressive.
On 2026-11-03 04:53, c186282 wrote:
On 11/2/26 21:49, Lawrence DrCOOliveiro wrote:
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous >>> controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
-a-a Yep ! Even the old mainframes were "smart" - lots
-a-a of optimizations that only later migrated down.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
-a-a System interrupts PER-KEYSTROKE ??? Horrible !
That is what a PC does.
On 2026-11-03 04:53, c186282 wrote:
On 11/2/26 21:49, Lawrence DrCOOliveiro wrote:
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous >>> controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
-a-a Yep ! Even the old mainframes were "smart" - lots
-a-a of optimizations that only later migrated down.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
-a-a System interrupts PER-KEYSTROKE ??? Horrible !
That is what a PC does.
On 03/11/2026 12:02, Carlos E. R. wrote:
On 2026-11-03 12:49, Pancho wrote:
On 11/3/26 11:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
What it is exactly they do, I don't know.
I'm too young for mainframes, but I always assumed they were for
running legacy systems that people couldn't be bothered to port to
new hardware/ architectures.
FWIW, my guess is that "high i/o" and "very large" is only by the
standards of mainframe programmers.
No, it is really high i/o and very large databases. I saw some figures
once and they were impressive.
IBM built a lot of intelligence into the IO subsystem. There are
multiple IO processors, once known as channels, which move data to and
from devices. They are controlled by a channel program which can perform multiple IOs so a single IO request can read or write multiple blocks of data from a tape or disk to or from different locations in memory with
the main CPU only getting involved when they are all complete.
So I don't know if this happens in practice but in theory if you want to page out one chunk of store and page in another you can do that with a single IO request.
On 11/2/26 18:08, Janis Papanagnou wrote:
On 2026-11-02 12:22, The Natural Philosopher wrote:
I'm also a bit astonished about your mention of using mainframes for
efficient I/O. Networking clusters appear to me to be much better able
than a single mainframe instance. The older mainframe architectures we
used also had a lot of peripheral processors to specifically handle
the I/O.
(But of course I can, as I did, just report about the (certainly non-
singular) cases I was involved in or heard from reliable sources.)
Janis
[...]
"Mainframes" do still have their places - major
computing/comm performance. Govt, science, big
international biz.
This may change sooner or later - all replaced
by 'cluster' and "Global AI".
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous >controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
On 03/11/2026 12:02, Carlos E. R. wrote:
On 2026-11-03 12:49, Pancho wrote:
On 11/3/26 11:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
What it is exactly they do, I don't know.
I'm too young for mainframes, but I always assumed they were for
running legacy systems that people couldn't be bothered to port to
new hardware/ architectures.
FWIW, my guess is that "high i/o" and "very large" is only by the
standards of mainframe programmers.
No, it is really high i/o and very large databases. I saw some figures
once and they were impressive.
IBM built a lot of intelligence into the IO subsystem. There are
multiple IO processors, once known as channels, which move data to and
from devices. They are controlled by a channel program which can perform multiple IOs so a single IO request can read or write multiple blocks of data from a tape or disk to or from different locations in memory with
the main CPU only getting involved when they are all complete.
So I don't know if this happens in practice but in theory if you want to page out one chunk of store and page in another you can do that with a single IO request.
Just to tweak things a bit further, if you have multiple disks,
controlled by a single controller, the controller can process multiple
IO requests. To optimise this it also has Rotational Position Sensing so
it can optimise the order in which they are processed.
Modern machines, of course use Fibre to connect the IO devices but back
in the day all this was connected via Bus and Tag channels which ran at 4Mbytes/sec, so 32Mbits with no contention.
Oh and for terminal IO the link to the screen runs at 1.5MBits/sec so
rather faster than any minicomputer link.
of course this complexity cam at a cost...
On 11/3/26 04:50, Carlos E. R. wrote:
On 2026-11-03 04:53, c186282 wrote:
On 11/2/26 21:49, Lawrence DrCOOliveiro wrote:
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous >>>> controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
-a-a Yep ! Even the old mainframes were "smart" - lots
-a-a of optimizations that only later migrated down.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
-a-a System interrupts PER-KEYSTROKE ??? Horrible !
That is what a PC does.
But a PC is, by definition, "personal", so this nonsense is not
affecting anyone else.
On 03/11/2026 11:50, Carlos E. R. wrote:
On 2026-11-03 04:53, c186282 wrote:Indeed it is
On 11/2/26 21:49, Lawrence DrCOOliveiro wrote:
On Tue, 03 Nov 2026 02:42:44 GMT, Charlie Gibbs wrote:
Consider that while the mini people were making a big thing of DMA,
the term was unheard of in the mainframe world - mainframers just
took it for granted.
The mainframe world had rCLchannelsrCY, which were practically autonomous >>>> controllers with their own ability to obey complex sequences of
program instructions while the CPU was busy doing application stuff.
-a-a Yep ! Even the old mainframes were "smart" - lots
-a-a of optimizations that only later migrated down.
The mini people were spending a lot of time building systems which
could efficiently respond to a CPU interrupt on every user keystroke.
Even the thought of that would have filled mainframers with horror.
-a-a System interrupts PER-KEYSTROKE ??? Horrible !
That is what a PC does.
Maybe CP/M did not, but IIRC DOS always did.
On 2026-11-03 09:40, Richard Kettlewell wrote:
"Carlos E. R." <robin_listas@es.invalid> writes:
On 2026-11-01 23:32, Lawrence DrCOOliveiro wrote:
On Sun, 1 Nov 2026 13:07:09 +0100, Carlos E. R. wrote:
On 2026-10-31 22:57, Lawrence DrCOOliveiro wrote:Actually, you *can* include URLs in the source markup, and the PDF
Tip: if you prefer another form, like PDF, itrCOs quite easy to
generate that from any man page.
PDF is fine, but not if the source is a man page, because there are
no cross references that you can click or tap ...
version will make them clickable.
Yes, that happens when the source is used to create several output
formats. Tpically both man and info, but there can be more.
But not when the pdf is generated using the man as source.
Links work fine in PDFs generated from man page sources. See
link at the bottom of https://terraraq.uk/junk/ccache.pdf to
https://ccache.dev/credits.html for an example.
That's not what I mean.
That's just an URL that is detected and activated.
"Carlos E. R." <robin_listas@es.invalid> writes:
On 2026-11-03 09:40, Richard Kettlewell wrote:
"Carlos E. R." <robin_listas@es.invalid> writes:
On 2026-11-01 23:32, Lawrence DrCOOliveiro wrote:
On Sun, 1 Nov 2026 13:07:09 +0100, Carlos E. R. wrote:
On 2026-10-31 22:57, Lawrence DrCOOliveiro wrote:Actually, you *can* include URLs in the source markup, and the PDF
Tip: if you prefer another form, like PDF, itrCOs quite easy to
generate that from any man page.
PDF is fine, but not if the source is a man page, because there are >>>>>> no cross references that you can click or tap ...
version will make them clickable.
Yes, that happens when the source is used to create several output
formats. Tpically both man and info, but there can be more.
But not when the pdf is generated using the man as source.
Links work fine in PDFs generated from man page sources. See
link at the bottom of https://terraraq.uk/junk/ccache.pdf to
https://ccache.dev/credits.html for an example.
That's not what I mean.
Please could you explain clearly what you do mean, then, being specific
about what exactly file formats yourCOre talking about as inputs, what
tools yourCOre talking about for any transformations, etc.
That's just an URL that is detected and activated.
It is not rCLdetectedrCY, it is specifically marked up as a URL with a .URL directive in the source for the man page.
Mainframes in the day used minis as "front-end processors" that
offloaded things like keystroke processing from the main CPU, which
could then concentrate on number crunching.
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
Actually, the numbers decline every year. The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
On 2026-11-03, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 03 Nov 2026 02:53:45 GMT, Charlie Gibbs wrote:
Yes, there is a dichotomy there. Mainframe protocols were designed
for sending files across a continent, while mini protocols were
designed for sending characters across the room. Compromises were
made on both sides.
ThatrCOs not the dichotomy. The dichotomy was that mainframes were
optimized for throughput at the expense of latency, while
minicomputers concentrated on minimizing latency to maximize
interactive responsiveness at the expense of throughput.
Pick your metaphor. In the end the trade-offs are the same.
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe >thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work.
The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
On 03/11/2026 15:13, Scott Lurndal wrote:
Actually, the numbers decline every year. The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys
mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
"Global mainframe market size and IBM Z mainframe revenue trends
highlight a cyclical pattern dominated by multi-year refresh waves (such
as the z16 and z17 cycles):
Global Mainframe Market Size:
Valued at approximately $5.33 billion to $5.71 billion in 2025.
Projected to grow to $5.65 billion to $6.11 billion in 2026.
Forecasted to reach $7.54 billion to $10.53 billion by
2031rCo2034, driven by a compound annual growth rate (CAGR) of roughly
5.95% to 7.05%.
IBM Z Mainframe Performance & Fluctuations:
Late 2025 (z17 Launch / Q4): IBM reported its strongest Q4
mainframe revenue in 20 years, with IBM Z sales surging 67%
year-over-year as enterprises adopted the z17 architecture.
Mid-2026 (Q2): IBM Z mainframe revenue experienced a short-term
42% year-over-year decline in fiscal Q2 2026 due to delayed large
capital expenditure (capex) deals and client shifts toward
supply-constrained servers, though management noted these were deferred >purchases rather than lost contracts (with roughly one-third closing
shortly after the quarter)."
So, as usual, the facts are not correctly represented by the resident >armchair experts
IBM built a lot of intelligence into the IO subsystem. There are
multiple IO processors, once known as channels, which move data to
and from devices. They are controlled by a channel program which can
perform multiple IOs so a single IO request can read or write
multiple blocks of data from a tape or disk to or from different
locations in memory with the main CPU only getting involved when
they are all complete.
And extra cost by ensuring it is reliable.
On 03/11/2026 20:25, Lawrence DrCOOliveiro wrote:
Imagine trying to run a full-screen editor that way: it just
wouldnrCOt work. The app really needed to be able to respond to every
keystroke in order to maintain the necessary level of interactive
response. Minis (and later superminis, like the VAX) had it all
over mainframes in that regard.
Why? I don't need to imagine, I did it all the time. Its not a
problem. The editor generates a screen full of data. The controller
handles most of the character editing so you change the local buffer
which does see every key stroke. At some point you send the modified
screen back to the host.
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but
then I am more familiar with it.
By the time of VMS, there was a new paradigm, where you could push
the pre-processing out to the terminal cluster: The TTY device in
the OS had mode bitmasks, specifying for each character whether it
should be echoed and whether it should generate a read completion
interrupt.
Unix got a similar set of modes controlled with the stty command and
its associated ioctl system calls, but I don't think I have ever
seen these succesfully be used to put the TTY pre-processing in a
remote front-end controller, like VMS could.
The Natural Philosopher <tnp@invalid.invalid> writes:
On 03/11/2026 15:13, Scott Lurndal wrote:
Actually, the numbers decline every year. The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys
mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
"Global mainframe market size and IBM Z mainframe revenue trends
highlight a cyclical pattern dominated by multi-year refresh waves (such
as the z16 and z17 cycles):
Global Mainframe Market Size:
Valued at approximately $5.33 billion to $5.71 billion in 2025.
Projected to grow to $5.65 billion to $6.11 billion in 2026.
Forecasted to reach $7.54 billion to $10.53 billion by
2031rCo2034, driven by a compound annual growth rate (CAGR) of roughly
5.95% to 7.05%.
IBM Z Mainframe Performance & Fluctuations:
Late 2025 (z17 Launch / Q4): IBM reported its strongest Q4
mainframe revenue in 20 years, with IBM Z sales surging 67%
year-over-year as enterprises adopted the z17 architecture.
Mid-2026 (Q2): IBM Z mainframe revenue experienced a short-term
42% year-over-year decline in fiscal Q2 2026 due to delayed large
capital expenditure (capex) deals and client shifts toward
supply-constrained servers, though management noted these were deferred
purchases rather than lost contracts (with roughly one-third closing
shortly after the quarter)."
So, as usual, the facts are not correctly represented by the resident
armchair experts
Dig into a bit and you'll find that those are refreshes for existing customers, not new customers.
Mainframes in the day used minis as "front-end processors" that
offloaded things like keystroke processing from the main CPU, which
could then concentrate on number crunching.
On 2026-11-03 03:50, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 03:02:07 +0100, Carlos E. R. wrote:
I have seen documentation PDFs that are just the output of
man2pdf.
Not sure why you need a separate package/command, when the usual
rCLmanrCY command is quite capable of doing that itself.
And is just as useless.
On 03/11/2026 11:49, Pancho wrote:
On 11/3/26 11:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
What it is exactly they do, I don't know.
I'm too young for mainframes, but I always assumed they were for
running legacy systems that people couldn't be bothered to port to new
hardware/architectures.
FWIW, my guess is that "high i/o" and "very large" is only by the
standards of mainframe programmers.
You would guess wrong then
Think about the number of people a second swiping their cards or
applePay wotsits across terminals...
Each transaction is trivial: post a debit, with details of where the
debt is incurred, and check against customer balance and overdraft
limits, and issue - or deny - authorisation,
But you are talking millions per second. All of which demand instant response. All of which access centralised databases that cant be split across machines
I don't expect new banks, either.
On Tue, 3 Nov 2026 20:53:29 +0000, David Wade wrote:
On 03/11/2026 20:25, Lawrence DrCOOliveiro wrote:
Imagine trying to run a full-screen editor that way: it just
wouldnrCOt work. The app really needed to be able to respond to every
keystroke in order to maintain the necessary level of interactive
response. Minis (and later superminis, like the VAX) had it all
over mainframes in that regard.
Why? I don't need to imagine, I did it all the time. Its not a
problem. The editor generates a screen full of data. The controller
handles most of the character editing so you change the local buffer
which does see every key stroke. At some point you send the modified
screen back to the host.
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but
then I am more familiar with it.
Neither one had an incremental-search function like the one in Emacs
though, did they? That updates the display in real-time, after every keystroke, highlighting all the visible matches for the string as
yourCOve typed it so far.
Try doing that on a block-mode terminal.
Now try doing it with a regular-expression search ...
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
...
Traditional banking transactions almost certainly could be parallelised,
to quite a high degree, whether they were or not, I don't know. They are >nowadays.
The Natural Philosopher <tnp@invalid.invalid> writes:
On 03/11/2026 15:13, Scott Lurndal wrote:
Actually, the numbers decline every year. The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys
mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
"Global mainframe market size and IBM Z mainframe revenue trends
highlight a cyclical pattern dominated by multi-year refresh waves (such
as the z16 and z17 cycles):
Global Mainframe Market Size:
Valued at approximately $5.33 billion to $5.71 billion in 2025.
Projected to grow to $5.65 billion to $6.11 billion in 2026.
Forecasted to reach $7.54 billion to $10.53 billion by
2031rCo2034, driven by a compound annual growth rate (CAGR) of roughly
5.95% to 7.05%.
IBM Z Mainframe Performance & Fluctuations:
Late 2025 (z17 Launch / Q4): IBM reported its strongest Q4
mainframe revenue in 20 years, with IBM Z sales surging 67%
year-over-year as enterprises adopted the z17 architecture.
Mid-2026 (Q2): IBM Z mainframe revenue experienced a short-term
42% year-over-year decline in fiscal Q2 2026 due to delayed large
capital expenditure (capex) deals and client shifts toward
supply-constrained servers, though management noted these were deferred
purchases rather than lost contracts (with roughly one-third closing
shortly after the quarter)."
So, as usual, the facts are not correctly represented by the resident
armchair experts
Dig into a bit and you'll find that those are refreshes for existing customers, not new customers.
I think all this shows is that IBM is able to force expensive
hardware refresh by ending support for old machines and old versions
of software so if you need Payment Card Industry compliance you need supported software.
So, as usual, the facts are not correctly represented by the resident >>armchair experts
Dig into a bit and you'll find that those are refreshes for existing >customers, not new customers.
I guess you COULD call the PDP-11 a "mini", but
imho it didn't have quite the "feel". More like
a cut-down mainframe. Good peripheral bus though,
you could plug in all kinds of goodies.
On 03/11/2026 21:57, Lawrence DrCOOliveiro wrote:
no, but how often do you use that?
Neither one had an incremental-search function like the one in
Emacs though, did they? That updates the display in real-time,
after every keystroke, highlighting all the visible matches for the
string as yourCOve typed it so far.
Now try doing it with a regular-expression search ...
Now that is real geekdom...
According to Pancho <Pancho.Jones@protonmail.com>:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
...
Traditional banking transactions almost certainly could be parallelised,
to quite a high degree, whether they were or not, I don't know. They are
nowadays.
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one succeeded and the other failed*, your bank has a problem. The reason they need very fast very large databases is that updates have to be serialized
and there's no way to make them faster than the underlying system. Same
for airlines allocating seats and any other database reflecting a real world application where there's a finite amount of something to be parcelled out.
* - yes I know about overdrafts, this is an example
On Tue, 03 Nov 2026 07:15:05 GMT, Charlie Gibbs wrote:
Mainframes in the day used minis as "front-end processors" that
offloaded things like keystroke processing from the main CPU, which
could then concentrate on number crunching.
Except that mainframes (of the IBM kind, anyway) were never very good
at number crunching. Their job was rCLbusiness data processingrCY (i.e. COBOL).
If you wanted to do number crunching, you got yourself a CDC machine.
Now *there* was a number cruncher ...
On Tue, 03 Nov 2026 07:15:04 GMT, Charlie Gibbs wrote:
On 2026-11-03, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 03 Nov 2026 02:53:45 GMT, Charlie Gibbs wrote:
Yes, there is a dichotomy there. Mainframe protocols were designed
for sending files across a continent, while mini protocols were
designed for sending characters across the room. Compromises were
made on both sides.
ThatrCOs not the dichotomy. The dichotomy was that mainframes were
optimized for throughput at the expense of latency, while
minicomputers concentrated on minimizing latency to maximize
interactive responsiveness at the expense of throughput.
Pick your metaphor. In the end the trade-offs are the same.
Nothing rCLmetaphoricalrCY about it. Sending data across a continent in real-time became commonplace with the coming of ARPAnet. But those
mainframes couldnrCOt actually connect directly, could they? They had to
go through one of those rCLIMPrCY boxes from BBN. And those were built
around ... minicomputers.
Except that mainframes (of the IBM kind, anyway) were never very good at number crunching. Their job was rCLbusiness data processingrCY (i.e. COBOL).
On 03/11/2026 20:25, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe
thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
Why? I don't need to imagine, I did it all the time. Its not a problem.
The editor generates a screen full of data.
The controller handles most of the character editing so you change the
local buffer which does see every key stroke.
At some point you send the modified screen back to the host.
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but then
I am more familiar with it.
[...]
Recalling at the time that interactive used was typically form-based
(e.g. airline reservation systems, bank teller terminals, etc) and
the block mode terminals of the day typically transferred the contents
of the screen (or the form fields on the screen) in a single transmission.
[...]
-a The age of the "mainframe" may be coming to an
-a end alas. 'Cluster'/AI solutions are going to
-a replace them for most uses. However science/govt
-a still have uses for big ultra-powerful boxes
-a for awhile anyway.
SOME advantage in keeping
-a access 'local' too - as opposed to spread all
-a over the highly-vulnerable net.
[...]
I know at least one guy working on IBM mainframes, and he tells me they
are in high need of good assembler programmers.
[...]
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem. The reason they >> need very fast very large databases is that updates have to be serialized
and there's no way to make them faster than the underlying system. Same
for airlines allocating seats and any other database reflecting a real world >> application where there's a finite amount of something to be parcelled out. >>
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone >else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
On 2026-11-03 04:29, c186282 wrote:
[...]
-a-a The age of the "mainframe" may be coming to an
-a-a end alas. 'Cluster'/AI solutions are going to
-a-a replace them for most uses. However science/govt
-a-a still have uses for big ultra-powerful boxes
-a-a for awhile anyway.
In practice I see no evidence for that here. And why
would it for "science" be better to have a mainframe?
My children actually work "in science" where extremely
demanding complex (hydrodynamic) simulation models are
done; all done on such high-performance clusters. And
this appears to be nowadays quite common as observable.
-a-a SOME advantage in keeping
-a-a access 'local' too - as opposed to spread all
-a-a over the highly-vulnerable net.
There had always been non-local access, even back in
the mainframe days.
There's a more general problem here; whether a single
"box" or a "cluster", if it's connected to the Net it
has to be solidified.
On 11/3/26 22:47, John Levine wrote:
According to Pancho-a <Pancho.Jones@protonmail.com>:
I know at least one guy working on IBM mainframes, and he tells me >>>>>> they are in high need of good assembler programmers.
They need high i/o with very large databases.
...
Traditional banking transactions almost certainly could be parallelised, >>> to quite a high degree, whether they were or not, I don't know. They are >>> nowadays.
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem.-a The reason
they
need very fast very large databases is that updates have to be serialized
and there's no way to make them faster than the underlying system.-a Same
for airlines allocating seats and any other database reflecting a real
world
application where there's a finite amount of something to be parcelled
out.
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
On Tue, 3 Nov 2026 22:51:30 +0000, David Wade wrote:IBM *tends* to Do It Right.
I think all this shows is that IBM is able to force expensive
hardware refresh by ending support for old machines and old versions
of software so if you need Payment Card Industry compliance you need
supported software.
I donrCOt think thatrCOs true. I have hooked up various clients to various payment systems over the years, and there was never a need for any
particular piece of proprietary software at our end.
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
NOTICED - my principal bank - a LOT of transactions
come up with a "pending" label. Might take DAYS. This
is not some 3rd-tier bank either.
According to Scott Lurndal <slp53@pacbell.net>:
So, as usual, the facts are not correctly represented by the resident
armchair experts
Dig into a bit and you'll find that those are refreshes for existing
customers, not new customers.
IBM has two markets for z mainframes, the legacy S/360/370/390 market
and high performance linux servers which they call LinuxONE.
I agree they're not likely to have new legacy customers but I'd be surprised that they put all that effort into LinuxONE without expecting to get new customers out of it.
They just announced that the next generation of z will have a
partnership with ARM so the processors can run ARM64 code as well as
the three legacy 360 modes. That tells me they're trying to get
customers who already have the ARM code running in the cloud at AWS or
Azure.
According to c186282 <c186282@nnada.net>:
I guess you COULD call the PDP-11 a "mini", but
imho it didn't have quite the "feel". More like
a cut-down mainframe. Good peripheral bus though,
you could plug in all kinds of goodies.
I can assure you that PDP-11s were minicomputers. I spent quite a while programming them back in the 1970s.
I can assure you that PDP-11s were minicomputers. I spent quite a while
programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
DEC had the VAX line of mainframes ...
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
The PDP-8 I used had "three cycle data break" which
did data transfers into or out of memory (that was
one cycle) using address and word counts also in
memory (those were the other two cycles.) But that
was only for the disks, and didn't do all of the other
chaining and control stuff that channels did.
DEC had the VAX line of mainframes ...
A VAX was a super-mini, not a mainframe. I was there.
Again, simpler I/O even though it was a lot bigger than
a PDP-8 or PDP-11.
If you look through the comp.arch archives you'll find arguments
about whether a PDP-10 was a mainframe or a very large mini. I
claim it was a mini because its I/O was similar to DEC's other
minis, including a hack reminiscent of three-cycle data break.
Others say it was a mainframe because it was marketed that way.
I will admit that the later KL with a PDP-11 front end and MASSBUS
disk I/O was a lot more like a mainframe than the original KA-10.
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
Mainframes were record-oriented, while minis were byte-oriented.
According to Pancho <Pancho.Jones@protonmail.com>:
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem. The reason they >>> need very fast very large databases is that updates have to be serialized >>> and there's no way to make them faster than the underlying system. Same >>> for airlines allocating seats and any other database reflecting a real world
application where there's a finite amount of something to be parcelled out. >>>
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
Sure, but you're not going to put every account on a different disk. Real transactions are more complicated than that -- when they take the $75 out of my
account they have to credit it somewhere else, and that debit/credit pair has to
be atomic, both work or both fail.
Banks are conservative but they are not stupid. There's a lot of stuff they can
push out onto cheap micro based servers, but there's a certain amount of key business logic that Absolutely Has To Work.
According to c186282 <c186282@nnada.net>:
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with >>> me, it can be done in parallel.
NOTICED - my principal bank - a LOT of transactions
come up with a "pending" label. Might take DAYS. This
is not some 3rd-tier bank either.
That has nothing to do with your bank's database and everything to do
with the way that credit card charges work. First the charge is
approved for an approximate amount which is pretty quick, then the
actual charge is posted which can take anywhere from moments to days.
The amounts are often different, e.g., when you pump gas it approves
an amount big enough to cover the largest plausible fill-up, then when
you're done it posts the final amount. Or at a restaurant, it approves
the amount on the check, then posts that plus whatever tip you add.
For historic reasons it's common to wait to post all of a day's
charges at the end of the day, or even later which is why they can sit
there pending for a while.
According to Pancho <Pancho.Jones@protonmail.com>:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
...
Traditional banking transactions almost certainly could be parallelised,
to quite a high degree, whether they were or not, I don't know. They are
nowadays.
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one succeeded and the other failed*, your bank has a problem. The reason they need very fast very large databases is that updates have to be serialized
and there's no way to make them faster than the underlying system. Same
for airlines allocating seats and any other database reflecting a real world application where there's a finite amount of something to be parcelled out.
* - yes I know about overdrafts, this is an example--
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should
meanwhile be at "best" a niche, and systems programming
with "C" (or similar).
But, yes. It was now for long a problem getting staff for
legacy things. Those companies who are too inert to adapt
[to the technical evolution] will pay, suffer, or demise.
Janis
[...]
On 2026-11-03, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 03 Nov 2026 07:15:04 GMT, Charlie Gibbs wrote:
On 2026-11-03, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 03 Nov 2026 02:53:45 GMT, Charlie Gibbs wrote:
Yes, there is a dichotomy there. Mainframe protocols were designed
for sending files across a continent, while mini protocols were
designed for sending characters across the room. Compromises were
made on both sides.
ThatrCOs not the dichotomy. The dichotomy was that mainframes were
optimized for throughput at the expense of latency, while
minicomputers concentrated on minimizing latency to maximize
interactive responsiveness at the expense of throughput.
Pick your metaphor. In the end the trade-offs are the same.
Nothing rCLmetaphoricalrCY about it. Sending data across a continent in
real-time became commonplace with the coming of ARPAnet. But those
mainframes couldnrCOt actually connect directly, could they? They had to
go through one of those rCLIMPrCY boxes from BBN. And those were built
around ... minicomputers.
That's not the only way to get data across the continent. In
the late '70s our mainframe in Vancouver was driving block-mode
multidropped terminals in Toronto through a leased line at a
blazing 2400 bps. From a software standpoint they looked just
like our local terminals, except that the wires were much longer
and the speed slower. (Our local terminals ran at 9600 bps.)
In those days modems cost about a dollar per bps - that's right,
$2500 for a Bell 202-style modem running synchronously at 2400 bps.
You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either. As ever, we balanced
price against performance.
On Tue, 3 Nov 2026 22:44:10 +0000, David Wade wrote:
On 03/11/2026 21:57, Lawrence DrCOOliveiro wrote:
no, but how often do you use that?
Neither one had an incremental-search function like the one in
Emacs though, did they? That updates the display in real-time,
after every keystroke, highlighting all the visible matches for the
string as yourCOve typed it so far.
Every day.
So, as usual, the facts are not correctly represented by the residentDig into a bit and you'll find that those are refreshes for existing customers, not new customers.
armchair experts
So, as usual, the facts are not correctly represented by the resident
armchair experts
Dig into a bit and you'll find that those are refreshes for existing
customers, not new customers.
I don't expect new banks, either.
for one type of technology, it was...You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either.-a As ever, we balanced
price against performance.
And we were told that 9600 was the limit, the theory said so. I think it
was 9600.
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should
meanwhile be at "best" a niche, and systems programming
with "C" (or similar).
But, yes. It was now for long a problem getting staff for
legacy things. Those companies who are too inert to adapt
[to the technical evolution] will pay, suffer, or demise.
Janis
[...]
On Wed, 04 Nov 2026 06:21:03 GMT, Charlie Gibbs wrote:
Mainframes were record-oriented, while minis were byte-oriented.
Maybe better to say rCLminis were character-orientedrCY. Remember DEC had 18-bit and 12-bit minicomputer architectures before IBM even invented
the term rCLbyterCY.
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
On 03/11/2026 15:13, Scott Lurndal wrote:
Actually, the numbers decline every year.-a-a The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys
mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
"Global mainframe market size and IBM Z mainframe revenue trends
highlight a cyclical pattern dominated by multi-year refresh waves (such
as the z16 and z17 cycles):
Global Mainframe Market Size:
-a-a-a-a-a-a-a Valued at approximately $5.33 billion to $5.71 billion in 2025.
-a-a-a-a-a-a-a Projected to grow to $5.65 billion to $6.11 billion in 2026.
-a-a-a-a-a-a-a Forecasted to reach $7.54 billion to $10.53 billion by 2031rCo2034, driven by a compound annual growth rate (CAGR) of roughly
5.95% to 7.05%.
IBM Z Mainframe Performance & Fluctuations:
-a-a-a-a-a-a-a Late 2025 (z17 Launch / Q4): IBM reported its strongest Q4 mainframe revenue in 20 years, with IBM Z sales surging 67%
year-over-year as enterprises adopted the z17 architecture.
-a-a-a-a-a-a-a Mid-2026 (Q2): IBM Z mainframe revenue experienced a short-term
42% year-over-year decline in fiscal Q2 2026 due to delayed large
capital expenditure (capex) deals and client shifts toward
supply-constrained servers, though management noted these were deferred purchases rather than lost contracts (with roughly one-third closing
shortly after the quarter)."
So, as usual, the facts are not correctly represented by the resident armchair experts
On Tue, 3 Nov 2026 12:45:09 +0100, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
Something not quite right there. Hand-coded assembly language only
makes sense for minimizing CPU usage. If yourCOre doing lots of I/O,
then the CPU is not the bottleneck, so trying to hand-tune assembly
code is a waste of time.
On 03/11/2026 20:25, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe
thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
Why? I don't need to imagine, I did it all the time. Its not a problem.
The editor generates a screen full of data.
The controller handles most of the character editing so you change the
local buffer which does see every key stroke.
At some point you send the modified screen back to the host.
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but then
I am more familiar with it.
Dave
On Tue, 3 Nov 2026 18:06:33 +0100, Carlos E. R. wrote:
And extra cost by ensuring it is reliable.
Very expensive way of ensuring reliability. Google ensures reliability
by running (as of a count in a paper published a decade or two ago)
about half a million servers. So, at any moment, a dozen or a hundred
of them are down; so what? The customers donrCOt notice.
Neither one had an incremental-search function like the one in Emacs
though, did they? That updates the display in real-time, after every keystroke, highlighting all the visible matches for the string as
yourCOve typed it so far.
Now try doing it with a regular-expression search ...
If you want heavy usage, I would guess bank risk calcs, weather calcs,
lots of science applications, running on compute grids, (or whatever the modern word is), blow this mainframe high i/o away.
According to Scott Lurndal <slp53@pacbell.net>:
So, as usual, the facts are not correctly represented by the resident >>>armchair experts
Dig into a bit and you'll find that those are refreshes for existing >>customers, not new customers.
IBM has two markets for z mainframes, the legacy S/360/370/390 market
and high performance linux servers which they call LinuxONE.
I agree they're not likely to have new legacy customers but I'd be surprised >that they put all that effort into LinuxONE without expecting to get new >customers out of it.
They just announced that the next generation of z will have a
partnership with ARM so the processors can run ARM64 code as well as
the three legacy 360 modes. That tells me they're trying to get
customers who already have the ARM code running in the cloud at AWS or
Azure.
On Tue, 3 Nov 2026 20:53:29 +0000, David Wade wrote:
On 03/11/2026 20:25, Lawrence DrCOOliveiro wrote:
Imagine trying to run a full-screen editor that way: it just
wouldnrCOt work. The app really needed to be able to respond to every
keystroke in order to maintain the necessary level of interactive
response. Minis (and later superminis, like the VAX) had it all
over mainframes in that regard.
Why? I don't need to imagine, I did it all the time. Its not a
problem. The editor generates a screen full of data. The controller
handles most of the character editing so you change the local buffer
which does see every key stroke. At some point you send the modified
screen back to the host.
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but
then I am more familiar with it.
Neither one had an incremental-search function like the one in Emacs
though, did they?
On Tue, 3 Nov 2026 05:28:52 -0800, Lars Poulsen wrote:
By the time of VMS, there was a new paradigm, where you could push
the pre-processing out to the terminal cluster: The TTY device in
the OS had mode bitmasks, specifying for each character whether it
should be echoed and whether it should generate a read completion
interrupt.
You realize, all this was done in *software* in the terminal driver,
not in the terminal controller?
So yes, the CPU was still getting an interrupt on every incoming
character.
Unix got a similar set of modes controlled with the stty command and
its associated ioctl system calls, but I don't think I have ever
seen these succesfully be used to put the TTY pre-processing in a
remote front-end controller, like VMS could.
Unix never had a terminal driver as advanced as VMS did, that is true.
According to Pancho <Pancho.Jones@protonmail.com>:
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem. The reason they >>> need very fast very large databases is that updates have to be serialized >>> and there's no way to make them faster than the underlying system. Same >>> for airlines allocating seats and any other database reflecting a real world
application where there's a finite amount of something to be parcelled out. >>>
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with
me, it can be done in parallel.
Sure, but you're not going to put every account on a different disk. Real transactions are more complicated than that -- when they take the $75 out of my
account they have to credit it somewhere else, and that debit/credit pair has to
be atomic, both work or both fail.
Banks are conservative but they are not stupid. There's a lot of stuff they can
push out onto cheap micro based servers, but there's a certain amount of key business logic that Absolutely Has To Work.
On 11/3/26 18:18, Pancho wrote:
On 11/3/26 22:47, John Levine wrote:
According to Pancho-a <Pancho.Jones@protonmail.com>:
I know at least one guy working on IBM mainframes, and he tells me >>>>>>> they are in high need of good assembler programmers.
They need high i/o with very large databases.
...
Traditional banking transactions almost certainly could be
parallelised,
to quite a high degree, whether they were or not, I don't know. They
are
nowadays.
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one
succeeded and the other failed*, your bank has a problem.-a The reason
they
need very fast very large databases is that updates have to be
serialized
and there's no way to make them faster than the underlying system.-a Same >>> for airlines allocating seats and any other database reflecting a
real world
application where there's a finite amount of something to be
parcelled out.
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do
with me, it can be done in parallel.
-a NOTICED - my principal bank - a LOT of transactions
-a come up with a "pending" label. Might take DAYS. This
-a is not some 3rd-tier bank either.
-a Two or more larger credit/debit card charges, not
-a sure HOW they 'compete' in that system. If too
-a near each other one MIGHT get bumped. Problem !
-a In short, even 'online/instant' banking may not
-a be THAT instant.
On 11/3/26 13:27, The Natural Philosopher wrote:
On 03/11/2026 15:13, Scott Lurndal wrote:
Actually, the numbers decline every year.-a-a The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys
mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
"Global mainframe market size and IBM Z mainframe revenue trends
highlight a cyclical pattern dominated by multi-year refresh waves (such
as the z16 and z17 cycles):
Global Mainframe Market Size:
-a-a-a-a-a-a-a-a Valued at approximately $5.33 billion to $5.71 billion in 2025.
-a-a-a-a-a-a-a-a Projected to grow to $5.65 billion to $6.11 billion in 2026.
-a-a-a-a-a-a-a-a Forecasted to reach $7.54 billion to $10.53 billion by
2031rCo2034, driven by a compound annual growth rate (CAGR) of roughly
5.95% to 7.05%.
IBM Z Mainframe Performance & Fluctuations:
-a-a-a-a-a-a-a-a Late 2025 (z17 Launch / Q4): IBM reported its strongest Q4 >> mainframe revenue in 20 years, with IBM Z sales surging 67%
year-over-year as enterprises adopted the z17 architecture.
-a-a-a-a-a-a-a-a Mid-2026 (Q2): IBM Z mainframe revenue experienced a short-term
42% year-over-year decline in fiscal Q2 2026 due to delayed large
capital expenditure (capex) deals and client shifts toward
supply-constrained servers, though management noted these were deferred
purchases rather than lost contracts (with roughly one-third closing
shortly after the quarter)."
So, as usual, the facts are not correctly represented by the resident
armchair experts
My only question is how many mainframe shipments represent NEW
customers, as opposed to additional horsepower for current customers.
I suppose, given that the trend today is to consolidation of everything, that eventually everything will be owned by BigCorp, who will then be
the only mainframe customer.
On 2026-11-04 00:05, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 22:44:10 +0000, David Wade wrote:
On 03/11/2026 21:57, Lawrence DrCOOliveiro wrote:
no, but how often do you use that?
Neither one had an incremental-search function like the one in
Emacs though, did they? That updates the display in real-time,
after every keystroke, highlighting all the visible matches for the
string as yourCOve typed it so far.
Every day.
I'm told that most people do not work directly on a terminal on the mainframe. There is not an OS proper...
On 11/3/26 13:25, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 09:28:46 +0000, David Wade wrote:
I think you misunderstand "mainframes" or at least IBM Mainframes.
The Mainframe never "saw" keystrokes, they were handled in the
terminal controller so there never was any possibility of the
dreaded "type-behind" one saw on an over-loaded VAX.
ThatrCOs part of what contributed to mainframes having less support
for interactive operation compared to minis.
Because the Mainframe only sees blocks of typing there are fewer
interrupts and less paging so interactive response is generally as
good as or better than a mini.
ThatrCOs only if you assume rCLblock moderCY operation. Typical mainframe
thinking: more efficient use of the computer, less efficient for the
human user.
Imagine trying to run a full-screen editor that way: it just wouldnrCOt
work. The app really needed to be able to respond to every keystroke
in order to maintain the necessary level of interactive response.
Minis (and later superminis, like the VAX) had it all over mainframes
in that regard.
I violently disagree. Full screen editors provide instant response to
most operations: Move the cursor around, add, delete, and change text. There's only a delay when you're done editing that screen and want to
move on to something else. On a well-tuned system this is a slight delay.
It just occurred to me that I'm typing this text into a mail app which
is basically the same thing. I can edit all I want, and then press SEND
when I'm done.
Maybe it's my mainframe background, but I think having "the computer" respond instantly to every keystroke is just stupid.
Peter Flass wrote this screed in ALL-CAPS (fixed):[...]
It just occurred to me that I'm typing this text into a mail app which
is basically the same thing. I can edit all I want, and then press SEND
when I'm done.
Maybe it's my mainframe background, but I think having "the computer"
respond instantly to every keystroke is just stupid.
I demand instant response!
On 11/3/26 21:16, Janis Papanagnou wrote:[...]
My children actually work "in science" where extremely
demanding complex (hydrodynamic) simulation models are
done; all done on such high-performance clusters. And
this appears to be nowadays quite common as observable.
-a Umm ... your KIDS are into hydrodynamics ? Are
-a they 'kids' anymore ?
There had always been non-local access, even back in
the mainframe days.
-a Yep - happened. However it's easier to guard/armor
-a ONE mainframe than a vast network.
There's a more general problem here; whether a single
"box" or a "cluster", if it's connected to the Net it
has to be solidified.
-a DOES require a lot of 'solidification' - esp now.
-a Otherwise you're just working for your worst enemies.
On 11/3/26 23:15, John Levine wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
Kinda true.
However 'minis' were SUPPOSED to do everything
mainframes could do - just a a DISCOUNT PRICE.
A VAX was a super-mini, not a mainframe. I was there.
Again, simpler I/O even though it was a lot bigger than
a PDP-8 or PDP-11.
Ummmmm ... I suspect we're dealing with
slightly different defs of 'mainframe'
and 'mini' and whatever. There is NO
clear line.
On 11/4/26 02:10, Lawrence DrCOOliveiro wrote:
On Wed, 04 Nov 2026 06:21:03 GMT, Charlie Gibbs wrote:
Mainframes were record-oriented, while minis were byte-oriented.
Maybe better to say rCLminis were character-orientedrCY. Remember DEC had
18-bit and 12-bit minicomputer architectures before IBM even invented
the term rCLbyterCY.
Well, the 'standard' 8-bit byte kind of evolved, or converged.
On 04/11/2026 02:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:You don't know what IBM assembler is, do you?
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should
meanwhile be at "best" a niche, and systems programming
with "C" (or similar).
But, yes. It was now for long a problem getting staff forYou don't really know what you are talking about, do you?
legacy things. Those companies who are too inert to adapt
[to the technical evolution] will pay, suffer, or demise.
On 2026-11-04 12:25, The Natural Philosopher wrote:
On 04/11/2026 02:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:You don't know what IBM assembler is, do you?
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should
meanwhile be at "best" a niche, and systems programming
with "C" (or similar).
"IBM assembler" specifically? No. - I know (and programmed in) a
couple other "high-level" macro-assemblers. - Do you imply that
the "IBM assembler" is somehow special, and so different that it
would somehow invalidate what I said?
But, yes. It was now for long a problem getting staff forYou don't really know what you are talking about, do you?
legacy things. Those companies who are too inert to adapt
[to the technical evolution] will pay, suffer, or demise.
I certainly do. - But, given your post, I doubt you intend to
contribute any substance and prefer just playing the troll.
(With your stance I don't think it's worthwhile to respond to
you in future.) Good luck.
Janis
DOOM and DUKE NUKEM were great. QUAKE ... it
was mostly BLACK SCREEN shit, why bother having
a monitor ???
Still have DUKE stashed away somewhere, runs in
DOSBox. Lots of user-created episodes too. For one
I had to write a 256 bit binary-2-decimal calculator :-)
Peter Flass wrote this screed in ALL-CAPS (fixed):
Maybe it's my mainframe background, but I think having "the computer"
respond instantly to every keystroke is just stupid.
I demand instant response!
And if you think the fruit vendor is better, you can rest easy, because
at one time they refused to update my card number. Actually refused.
After reaching the /right/ support person, that person could not help
figure out why the website told me the card number had been updated, but
then the TV application always tried to use the old card number.
On 11/3/26 13:29, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 12:45:09 +0100, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
Something not quite right there. Hand-coded assembly language only
makes sense for minimizing CPU usage. If yourCOre doing lots of I/O,
then the CPU is not the bottleneck, so trying to hand-tune assembly
code is a waste of time.
I kind of agree. As a former assembler programmer - at one time it was
my first language - I'd like it to be true (although I'm out of the job market), but I can't see it. These days most HLL code is more optimized than a lot of assembler. A good assembler programmer would not make a
lot of these optimizations due to lack of readability.
On 04/11/2026 09:58, Carlos E. R. wrote:
You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either.-a As ever, we balanced
price against performance.
And we were told that 9600 was the limit, the theory said so. I think it
was 9600.
for one type of technology, it was...
On 11/3/26 13:53, David Wade wrote:
Personally I prefer the XEDIT the VM/CMS editor to the VMS one, but then
I am more familiar with it.
I also like XEdit, in fact I run THE on my PC, although it is not my
primary editor. VAX TPU is also quite good, however, but we had great response time on our VAX.
On 11/3/26 15:15, Pancho wrote:
If you want heavy usage, I would guess bank risk calcs, weather calcs,
lots of science applications, running on compute grids, (or whatever
the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
On 2026-11-04, Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> wrote:
And if you think the fruit vendor is better, you can rest easy, because
at one time they refused to update my card number. Actually refused.
After reaching the /right/ support person, that person could not help
figure out why the website told me the card number had been updated, but
then the TV application always tried to use the old card number.
Ah, you've been around the mulberry bush with Netflix too, I presume.
We recently found we couldn't log in to Netflix because they didn't
like our card number - and quoted the last four digits and date of
a card that expired seven years ago. I wonder how they had managed
to bill us the previous month.
I called their customer support line, and the droid became downright belligerent, claiming that it wasn't their fault but our bank's.
I wound up hanging up on him, and we cancelled our Netflix account.
On 2026-11-04, Chris Ahlstrom <OFeem1987@teleworm.us> wrote:
Peter Flass wrote this screed in ALL-CAPS (fixed):
Maybe it's my mainframe background, but I think having "the computer"
respond instantly to every keystroke is just stupid.
I can understand that. The overhead involved with dealing with every character coming in from multiple terminals can be considerable.
Given the CPU power we had back in the day (maybe 130 KIPS for a small-to-middling mainframe), it would bring the machine to its knees.
I demand instant response!
Given typical mainframe protocols of the day (bisync, half-duplex, multidropped terminals that get polled maybe once a second), you're
not going to get that even if you had all the CPU power in the world.
I remember reading about an experiment done in an IBM shop where
they shoveled in enough hardware and tweaked the protocols so that
terminal users got sub-second response time. Apparently it increased throughput even more than one would expect by just scaling the
response time - users could develop a rhythm and not have their
minds wander while waiting for a response.
On the other side of the coin, someone at Univac decided that
the key to productivity was not so much a quick response time
as a consistent one - which led to their system inserting
delays if a response was ready too soon. :-p
If you look through the comp.arch archives you'll find arguments about whether a PDP-10 was a mainframe or a very large mini. I claim it was a mini because its I/O was similar to DEC's other minis, including a hack reminiscent of three-cycle data break. Others say it was a mainframe because it was marketed that way.
I will admit that the later KL with a PDP-11 front end and MASSBUS disk I/O was a lot more like a mainframe than the original KA-10.
You can't distribute an airline reservation system, the data has to
live in one place.
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
On Wed, 04 Nov 2026 06:21:03 GMT, Charlie Gibbs wrote:
Mainframes were record-oriented, while minis were byte-oriented.
Maybe better to say "minis were character-oriented". Remember DEC had
18-bit and 12-bit minicomputer architectures before IBM even invented
the term "byte".
On 11/3/26 14:57, Lawrence DrCOOliveiro wrote:
Neither one had an incremental-search function like the one in
Emacs though, did they? That updates the display in real-time,
after every keystroke, highlighting all the visible matches for the
string as yourCOve typed it so far.
I find this more distracting than useful.
Let me type my search, including correcting inevitable typos, etc.
and then run it instead of continually flashing the screen at me as
I type.
Now try doing it with a regular-expression search ...
Why would this be a problem? "Yes, ISPF View and Edit support regular expressions."
And we were told that 9600 was the limit, the theory said so. I
think it was 9600.
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written
in C, with very little assembly language.
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written
in C, with very little assembly language.
On 2026-11-04, The Natural Philosopher <tnp@invalid.invalid> wrote:
On 04/11/2026 09:58, Carlos E. R. wrote:
You could get 9600-bps modems, but they were ten grand a pop.
And the leased line wasn't cheap either.-a As ever, we balanced
price against performance.
And we were told that 9600 was the limit, the theory said so. I think it >>> was 9600.
for one type of technology, it was...
The mainframe I worked on could do async (although nobody
ever did) - but since Everybody Knows that async is a
low-speed protocol the ports weren't designed to run async
any faster than 2400 bps, even for direct connections.
Didn't part of the RS-232 spec say that wiring had to handle
50-foot runs at 9600 bps? Given cable capacitance it could
be a crap shoot, but we managed to get async terminals
(or emulators on PCs) to talk to a Sun workstation at the
other end of the building at 38400 bps.
Also, the disks do not have a filesystem, either (I also asked).
Not something I ever did, but I strongly suspect it is message
queues, including acknowledgements. Tricky to get perfect, but
standard IT stuff.
I got my PC to talk to another PC at 118K with a serial port across
perhaps 20 metres, in the 90's.
On 2026-11-04, Peter Flass <Peter@Iron-Spring.com> wrote:
On 11/3/26 13:29, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 12:45:09 +0100, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
Something not quite right there. Hand-coded assembly language only
makes sense for minimizing CPU usage. If yourCOre doing lots of I/O,
then the CPU is not the bottleneck, so trying to hand-tune assembly
code is a waste of time.
I kind of agree. As a former assembler programmer - at one time it was
my first language - I'd like it to be true (although I'm out of the job
market), but I can't see it. These days most HLL code is more optimized
than a lot of assembler. A good assembler programmer would not make a
lot of these optimizations due to lack of readability.
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
On 11/4/26 14:54, Peter Flass wrote:
On 11/3/26 15:15, Pancho wrote:
If you want heavy usage, I would guess bank risk calcs, weather
calcs, lots of science applications, running on compute grids, (or
whatever the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
Not really,-quants used Excel on PCs. Lotus123 ran on Sun, but it was
second rate. I think prior to Windows NT some IT departments gambled on
Sun, which had a proper OS, rather than PCs, but it quickly became
obvious it was a poor choice.
There was also a buzz about RISC cpus replacing x86, which to be fair
now seems to be happening, just 35 years late.
x86 is RISC under the hood.
On Wed, 4 Nov 2026 19:04:22 -0700, Peter Flass wrote:
x86 is RISC under the hood.
ThatrCOs nonsense. ThatrCOs Intel/AMD PR talking.
On 11/4/26 03:12, John Levine wrote:
According to Pancho <Pancho.Jones@protonmail.com>:
Only up to a point. Making the read part of databases run in parallel
is easy. Making the write part is not.
If you have $100 in your bank account, and two $75 debit transactions
arrive at the same time, if the result is anything other than that one >>>> succeeded and the other failed*, your bank has a problem. The reason they >>>> need very fast very large databases is that updates have to be serialized >>>> and there's no way to make them faster than the underlying system. Same >>>> for airlines allocating seats and any other database reflecting a real world
application where there's a finite amount of something to be parcelled out.
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with >>> me, it can be done in parallel.
Sure, but you're not going to put every account on a different disk. Real
transactions are more complicated than that -- when they take the $75 out of my
account they have to credit it somewhere else, and that debit/credit pair has to
be atomic, both work or both fail.
Not something I ever did, but I strongly suspect it is message queues, including acknowledgements. Tricky to get perfect, but standard IT
stuff.
Banks are conservative but they are not stupid. There's a lot of stuff they can
push out onto cheap micro based servers, but there's a certain amount of key >> business logic that Absolutely Has To Work.
Absolutely has to work, means testing. That is why they keep the old
stuff, it has been tested, in production.
Newer system reliability will be almost entirely about software
design, better techniques, but testing is difficult.
We recently had a scandal in the UK, the Post Office, shocking flaws
in design, from the mid 1990s, but over the years they caught most of
the bugs and made it work.
A superficial look at the news reports--
showed they weren't using things like idempotent operations. Modern
systems are designed to be much more tolerant of hardware glitches.
According to c186282 <c186282@nnada.net>:
* - yes I know about overdrafts, this is an example
Erm, yep, sequential transactions on my account, but why does everyone
else's account need to be on the same database/compute server as my
account? If A and B are doing a transaction, that has nothing to do with >>> me, it can be done in parallel.
NOTICED - my principal bank - a LOT of transactions
come up with a "pending" label. Might take DAYS. This
is not some 3rd-tier bank either.
That has nothing to do with your bank's database and everything to do
with the way that credit card charges work. First the charge is
approved for an approximate amount which is pretty quick, then the
actual charge is posted which can take anywhere from moments to days.
The amounts are often different, e.g., when you pump gas it approves
an amount big enough to cover the largest plausible fill-up, then when
you're done it posts the final amount. Or at a restaurant, it approves
the amount on the check, then posts that plus whatever tip you add.
For historic reasons it's common to wait to post all of a day's--
charges at the end of the day, or even later which is why they can sit
there pending for a while.
On Wed, 4 Nov 2026 07:52:35 -0700, Peter Flass wrote:
On 11/3/26 14:57, Lawrence DrCOOliveiro wrote:
Neither one had an incremental-search function like the one in
Emacs though, did they? That updates the display in real-time,
after every keystroke, highlighting all the visible matches for the
string as yourCOve typed it so far.
I find this more distracting than useful.
You do, others donrCOt.
Let me type my search, including correcting inevitable typos, etc.
and then run it instead of continually flashing the screen at me as
I type.
Except you may not notice your typos until you see what does or
doesnrCOt match the search. ThatrCOs why the immediate feedback is useful.
--Now try doing it with a regular-expression search ...
Why would this be a problem? "Yes, ISPF View and Edit support regular
expressions."
But they are even more difficult to get right. Which is why the
immediate feedback becomes even more useful.
We recently had a scandal in the UK, the Post Office, shocking flaws
in design, from the mid 1990s, but over the years they caught most of
the bugs and made it work.
At the same time they kept prosecuting innocent people for it, right?
On Thu, 5 Nov 2026 00:14:16 +0100, Carlos E. R. wrote:
Also, the disks do not have a filesystem, either (I also asked).
Programming like itrCOs 1975 <https://vinyl-cache.org/docs/9.1/phk/notes.html>:
So whatrCOs wrong with 1975 programming ?
The really short answer is that computers do not have two kinds of
storage anymore.
...
Virtual memory was meant to make it easier to program when data
was larger than the physical memory, but people have still not
caught on.
On 11/4/26 12:51, Charlie Gibbs wrote:
On 2026-11-04, Peter Flass <Peter@Iron-Spring.com> wrote:
On 11/3/26 13:29, Lawrence DrCOOliveiro wrote:
On Tue, 3 Nov 2026 12:45:09 +0100, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
They need high i/o with very large databases.
Something not quite right there. Hand-coded assembly language only
makes sense for minimizing CPU usage. If yourCOre doing lots of I/O,
then the CPU is not the bottleneck, so trying to hand-tune assembly
code is a waste of time.
I kind of agree. As a former assembler programmer - at one time it was
my first language - I'd like it to be true (although I'm out of the job
market), but I can't see it. These days most HLL code is more optimized >>> than a lot of assembler. A good assembler programmer would not make a
lot of these optimizations due to lack of readability.
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
1. "standard" things like converting a branch to a nop or back. I know
1401 Autocoder people used to do horrible things like using part of an >instruction as a constant.
On 11/4/26 12:51, Charlie Gibbs wrote:
On 2026-11-04, Peter Flass <Peter@Iron-Spring.com> wrote:
I kind of agree. As a former assembler programmer - at one time it was
my first language - I'd like it to be true (although I'm out of the job
market), but I can't see it. These days most HLL code is more optimized >>> than a lot of assembler. A good assembler programmer would not make a
lot of these optimizations due to lack of readability.
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
1. "standard" things like converting a branch to a nop or back.
I know 1401 Autocoder people used to do horrible things like using
part of an instruction as a constant.
2. You can do anything as long as you comment it. I once had a sequence
to test-store-conditional branch, for some reason I don't recall, bit I tagged it with comments saying "these three instructions must be kept together".
On 2026-11-04, John Levine <johnl@taugh.com> wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
I agree. I've always believed that the difference between
a mainframe and a mini had nothing to do with the size of
the box, but everything to do with architecture, particularly
I/O. Mainframes were record-oriented, while minis were
byte-oriented.
On 11/3/26 13:27, The Natural Philosopher wrote:
On 03/11/2026 15:13, Scott Lurndal wrote:
Actually, the numbers decline every year.-a-a The vast majority are
legacy installations (particularly for the Burroughs/Univac/Unisys
mainframes) and there are very few, if any, new customers adopting
mainframes of any flavor.
"Global mainframe market size and IBM Z mainframe revenue trends
highlight a cyclical pattern dominated by multi-year refresh waves (such
as the z16 and z17 cycles):
Global Mainframe Market Size:
-a-a-a-a-a-a-a-a Valued at approximately $5.33 billion to $5.71 billion in 2025.
-a-a-a-a-a-a-a-a Projected to grow to $5.65 billion to $6.11 billion in 2026.
-a-a-a-a-a-a-a-a Forecasted to reach $7.54 billion to $10.53 billion by
2031rCo2034, driven by a compound annual growth rate (CAGR) of roughly
5.95% to 7.05%.
IBM Z Mainframe Performance & Fluctuations:
-a-a-a-a-a-a-a-a Late 2025 (z17 Launch / Q4): IBM reported its strongest Q4 >> mainframe revenue in 20 years, with IBM Z sales surging 67%
year-over-year as enterprises adopted the z17 architecture.
-a-a-a-a-a-a-a-a Mid-2026 (Q2): IBM Z mainframe revenue experienced a short-term
42% year-over-year decline in fiscal Q2 2026 due to delayed large
capital expenditure (capex) deals and client shifts toward
supply-constrained servers, though management noted these were deferred
purchases rather than lost contracts (with roughly one-third closing
shortly after the quarter)."
So, as usual, the facts are not correctly represented by the resident
armchair experts
My only question is how many mainframe shipments represent NEW
customers, as opposed to additional horsepower for current customers.
I suppose, given that the trend today is to consolidation of everything, that eventually everything will be owned by BigCorp, who will then be
the only mainframe customer.
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
On 11/3/26 15:15, Pancho wrote:
If you want heavy usage, I would guess bank risk calcs, weather calcs,
lots of science applications, running on compute grids, (or whatever
the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
Do I have something to say on this subject?-a No, not really; I just find
it amusing when trolls accuse other people of being trolls, and then, presumably, killfile the person.-a Consider it "trash taking itself out."
Best wishes, and happy philosophering of natural phenomena!
Lawrence =?iso-8859-13?q?D=FFOliveiro?= <ldo@nz.invalid> writes:
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written >> in C, with very little assembly language.
Again with the incorrect characterizations.
The fastest supercomputers are running massive amounts of
modern (and some ancient) Fortran code.
On 2026-11-04 22:47, Lawrence DrCOOliveiro wrote:
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me
they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written >> in C, with very little assembly language.
Well, I know for a fact that this machine doesn't run any standard OS. I asked him. Also, the disks do not have a filesystem, either (I also
asked). Of course most of the programmers work in some high level
language, but there are certain critical operations that are optimized
in assembler. I don't remember which, and obviously I'm not going to
quote him breaking confidence. But if anyone is a good programmer in
those machines, in assembler, they are hiring.
On 04/11/2026 23:14, Carlos E. R. wrote:
On 2026-11-04 22:47, Lawrence DrCOOliveiro wrote:My mathematical friend needs to use assembler routines to access some of
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me >>>>>> they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written >>> in C, with very little assembly language.
Well, I know for a fact that this machine doesn't run any standard OS.
I asked him. Also, the disks do not have a filesystem, either (I also
asked). Of course most of the programmers work in some high level
language, but there are certain critical operations that are optimized
in assembler. I don't remember which, and obviously I'm not going to
quote him breaking confidence. But if anyone is a good programmer in
those machines, in assembler, they are hiring.
the vector processing that the chipset, but not available-a C libraries, supports.
IBM Macro-a assembler is practically a high level language, given the
scale of macro usage in legacy systems
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
I agree. I've always believed that the difference between
a mainframe and a mini had nothing to do with the size of
the box, but everything to do with architecture, particularly
I/O. Mainframes were record-oriented, while minis were
byte-oriented.
The PDP-8 was anything but byte-oriented, it was a 12-bit machine.
Plus, DECTape was block-oriented.
On 04/11/2026 19:51, Charlie Gibbs wrote:
A mediocre assembly language programmer would not let lack of
readability stand in the way. And once you get into self-modifying
code there are all sorts of horrible things you can do.
IBM macro assembler is about as machine code as COBOL
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-04, John Levine <johnl@taugh.com> wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>>>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
I agree. I've always believed that the difference between
a mainframe and a mini had nothing to do with the size of
the box, but everything to do with architecture, particularly
I/O. Mainframes were record-oriented, while minis were
byte-oriented.
The PDP-8 was anything but byte-oriented, it was a 12-bit machine.
Plus, DECTape was block-oriented.
On 04/11/2026 23:14, Carlos E. R. wrote:
On 2026-11-04 22:47, Lawrence DrCOOliveiro wrote:My mathematical friend needs to use assembler routines to access some of
On Wed, 4 Nov 2026 10:53:41 +0100, Carlos E. R. wrote:
On 2026-11-04 03:23, Janis Papanagnou wrote:
On 2026-11-03 12:45, Carlos E. R. wrote:
I know at least one guy working on IBM mainframes, and he tells me >>>>>> they are in high need of good assembler programmers.
Wow! - I'd have thought that assembler programmers should meanwhile
be at "best" a niche, and systems programming with "C" (or
similar).
They need speed that compilers can not achieve.
The worldrCOs fastest supercomputers run Linux. And that is mostly written >>> in C, with very little assembly language.
Well, I know for a fact that this machine doesn't run any standard OS.
I asked him. Also, the disks do not have a filesystem, either (I also
asked). Of course most of the programmers work in some high level
language, but there are certain critical operations that are optimized
in assembler. I don't remember which, and obviously I'm not going to
quote him breaking confidence. But if anyone is a good programmer in
those machines, in assembler, they are hiring.
the vector processing that the chipset, but not available-a C libraries, supports.
IBM Macro-a assembler is practically a high level language, given the
scale of macro usage in legacy systems
On 2026-11-05, Thomas Koenig <tkoenig@netcologne.de> wrote:
Charlie Gibbs <cgibbs@kltpzyxm.invalid> schrieb:
On 2026-11-04, John Levine <johnl@taugh.com> wrote:
According to c186282 <c186282@nnada.net>:
I can assure you that PDP-11s were minicomputers. I spent quite a while >>>>>> programming them back in the 1970s.
AT THE TIME, characterizing the PDP-11 as
a "mini" is not wrong. I view a "mini" as
a 'cut-down' mainframe.
A mini was never a cut-down mainframe. Mainframes
had channels, minis had simple I/O and fast interrupts.
A 360/30 was considerably slower than a PDP-8, but its
I/O architecture clearly made it a mainframe.
I agree. I've always believed that the difference between
a mainframe and a mini had nothing to do with the size of
the box, but everything to do with architecture, particularly
I/O. Mainframes were record-oriented, while minis were
byte-oriented.
The PDP-8 was anything but byte-oriented, it was a 12-bit machine.
Plus, DECTape was block-oriented.
You've got me there. It's amazing how slippery definitions become,
even when starting from "I know one when I see one." I could say
that minis tend to process a character at a time (e.g. a PDP-11
driving a VT50) while mainframes tend to process a record at a
time (e.g. a 360 filling a 2260's screen in a single operation).
But exceptions (e.g. your DECTape) creep in...
I do admit that the other extreme can become hairy. I once saw a
recursive macro that solved the Towers of Hanoi. It generated no
code; the solution was in the expanded source listing.
I usually just felt like mini=16 bits, mainframe=32 bits or more.
On Thu, 05 Nov 2026 21:57:51 GMT, Charlie Gibbs wrote:
I do admit that the other extreme can become hairy. I once saw a
recursive macro that solved the Towers of Hanoi. It generated no
code; the solution was in the expanded source listing.
String-based macro languages in general can be pretty hairy.
Towers of Hanoi wouldnrCOt even be that hard.
----
dnl
dnl Example m4 implementation of the GPM example output shown on page 64
dnl of Bryan Higman's "A Comparative Study of Programming Languages", 1st ed. dnl
dnl Created 2008 September 10 by Lawrence D'Oliveiro.
dnl
define(`repeat', `ifelse($2, 0, `', $2, 1, `$1', `$1`'repeat(`$1', eval($2 - 1))')')dnl
dnl repeat(str, n) outputs n consecutive repetitions of str
define(`n', `regexp(` x one two three four five six seven eight nine ten', `^\'repeat(` [^ ]+', $1)` \([^ ]+\)', `\1')')dnl
dnl n(i) returns the word representation of i. Need to use the "repeat" macro dnl because m4's regexps don't support \{n\} repeat coefficients. define(`cap', `regexp(`$1', `^\(.\)\(.*\)$', `translit(`\1', `a-z', `A-Z')\2')')dnl
dnl cap(s) returns s with the first letter capitalized.
dnl
define(`nrverses', 10)dnl
define(`nmen', `cap(n($1)) ifelse($1, 1, man, men)')dnl
dnl nmen(i) outputs the phrase "<i> man/men"
define(`tomow', ` went to mow')dnl
define(`mow1', `tomow a meadow')dnl
define(`firstline', `nmen($1)`'tomow,mow1')dnl
dnl firstline(i) outputs the first line of verse i.
define(`restlines', `ifelse($1, 1, `nmen(1) and his dog,mow1', `nmen($1) restlines(eval($1 - 1))')')dnl
dnl restlines(i) outputs the remaining lines of verse i
define(`verse', `firstline($1)
restlines($1)')dnl
dnl verse(i) outputs the complete verse i
define(`verses', `verse($1)`'ifelse($1, nrverses, `', `
verses(eval($1 + 1))')')dnl
dnl verses(i) outputs the verses from i up to the last one.
dnl here we go...
verses(1)
On the other side of the coin, someone at Univac decided that
the key to productivity was not so much a quick response time
as a consistent one - which led to their system inserting
delays if a response was ready too soon. :-p
The PDP-8 was anything but byte-oriented, it was a 12-bit machine.
Plus, DECTape was block-oriented.
You've got me there. It's amazing how slippery definitions become,
even when starting from "I know one when I see one." I could say
that minis tend to process a character at a time (e.g. a PDP-11
driving a VT50) while mainframes tend to process a record at a
time (e.g. a 360 filling a 2260's screen in a single operation).
But exceptions (e.g. your DECTape) creep in...
I usually just felt like mini=16 bits, mainframe=32 bits or more.
I remember reading about an experiment done in an IBM shop where
they shoveled in enough hardware and tweaked the protocols so that
terminal users got sub-second response time. Apparently it increased throughput even more than one would expect by just scaling the
response time - users could develop a rhythm and not have their
minds wander while waiting for a response.
On the other side of the coin, someone at Univac decided that the
key to productivity was not so much a quick response time as a
consistent one - which led to their system inserting delays if a
response was ready too soon. :-p
IBM Macro assembler is practically a high level language, given the
scale of macro usage in legacy systems
On 2026-11-04, Johann 'Myrkraverk' Oskarsson <johann@myrkraverk.invalid> wrote:
And if you think the fruit vendor is better, you can rest easy, because
at one time they refused to update my card number. Actually refused.
After reaching the /right/ support person, that person could not help
figure out why the website told me the card number had been updated, but
then the TV application always tried to use the old card number.
Ah, you've been around the mulberry bush with Netflix too, I presume.
We recently found we couldn't log in to Netflix because they didn't
like our card number - and quoted the last four digits and date of
a card that expired seven years ago. I wonder how they had managed
to bill us the previous month.
I called their customer support line, and the droid became downright belligerent, claiming that it wasn't their fault but our bank's.
I wound up hanging up on him, and we cancelled our Netflix account.
On 11/3/26 20:30, c186282 wrote:
-a NOTICED - my principal bank - a LOT of transactions
-a come up with a "pending" label. Might take DAYS. This
-a is not some 3rd-tier bank either.
-a Two or more larger credit/debit card charges, not
-a sure HOW they 'compete' in that system. If too
-a near each other one MIGHT get bumped. Problem !
-a In short, even 'online/instant' banking may not
-a be THAT instant.
Some of this is communication between banks, and some is your bank
milking the float for as long as is legally allowed.
I got in serious trouble for this,
because the process scheduler assumed that 500 ms was "a long time", and
a program that was going to wait for a long time should be swapped out
of memory.
On 04/11/2026 14:54, Peter Flass wrote:
On 11/3/26 15:15, Pancho wrote:Not for transactuon processing
If you want heavy usage, I would guess bank risk calcs, weather calcs,
lots of science applications, running on compute grids, (or whatever
the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
And hererCOs another tale: this was probably from the very earliest days
of computers, but the story I remember reading involves users entering
a query, and how long it took to get back a response. Apparently, if
the response came back *too* quickly, the users felt that the computer somehow wasnrCOt treating their question with the importance it
deserved.
On 11/4/2026 11:51, Charlie Gibbs wrote:
On the other side of the coin, someone at Univac decided that
the key to productivity was not so much a quick response time
as a consistent one - which led to their system inserting
delays if a response was ready too soon. :-p
Back in my University Computer Center days around 1972, I was hacking
a local email feature into the EXEC-8 editor, and decided that less
variation in response times would be a good thing, so I added a 1/2
second wait before each editor command prompt. I got in serious trouble
for this, because the process scheduler assumed that 500 ms was "a long time", and a program that was going to wait for a long time should be swapped out of memory.
On 2026-11-06, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
And hererCOs another tale: this was probably from the very earliest days
of computers, but the story I remember reading involves users entering
a query, and how long it took to get back a response. Apparently, if
the response came back *too* quickly, the users felt that the computer
somehow wasnrCOt treating their question with the importance it
deserved.
I wrote a program that searches a large collection of files for
various data. I noticed the same thing: it ran so fast that
people weren't taking it seriously. It didn't help that the
search parameter screen was too easy to use, either.
On 2026-11-06, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
And hererCOs another tale: this was probably from the very earliest days
of computers, but the story I remember reading involves users entering
a query, and how long it took to get back a response. Apparently, if
the response came back *too* quickly, the users felt that the computer
somehow wasnrCOt treating their question with the importance it
deserved.
I wrote a program that searches a large collection of files for
various data. I noticed the same thing: it ran so fast that
people weren't taking it seriously. It didn't help that the
search parameter screen was too easy to use, either.
I guess that in this modern age, where complexity is equated
with sophistication, any sign of simplicity or efficiency
is suspect.
On 2026-11-06, Charlie Gibbs <cgibbs@kltpzyxm.invalid> wrote:
On 2026-11-06, Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
And hererCOs another tale: this was probably from the very earliest days >>> of computers, but the story I remember reading involves users entering
a query, and how long it took to get back a response. Apparently, if
the response came back *too* quickly, the users felt that the computer
somehow wasnrCOt treating their question with the importance it
deserved.
I wrote a program that searches a large collection of files for
various data. I noticed the same thing: it ran so fast that
people weren't taking it seriously. It didn't help that the
search parameter screen was too easy to use, either.
That's much the same like people not taking a pair of headphones
seriously if they are light; so Beats famously added hidden metal
weights to them to create a more premium feel.
I've heard that car manufacturers put padding into the doors
to give a satisfying thump rather than a clank.
The Natural Philosopher <tnp@invalid.invalid> schrieb:
IBM Macro assembler is practically a high level language, given the
scale of macro usage in legacy systems
High-level languages abstract away the underlying machine, which
I presume the Macro assembler does not.
Plus, does it do register allocation?
(OK, then Wirth's PL/360 creeps in, which does not do so either,
but then I would argue that that is a glorified assembler - it also
does not do operator precenence).
On 2026-11-07 02:22, Charlie Gibbs wrote:
I've heard that car manufacturers put padding into the doors
to give a satisfying thump rather than a clank.
Unless the various sounds in cars are created artificially!
(Not joking. - Inspect the automotive business job offers,
specifically sound designers and sound engineers, and also
inspect the information on the Net about where artificial
sound is actually synthesized and applied to, including the
doors.)
On 2026-11-07, Janis Papanagnou <janis_papanagnou+ng@hotmail.com> wrote:
On 2026-11-07 02:22, Charlie Gibbs wrote:
I've heard that car manufacturers put padding into the doors
to give a satisfying thump rather than a clank.
Unless the various sounds in cars are created artificially!
(Not joking. - Inspect the automotive business job offers,
specifically sound designers and sound engineers, and also
inspect the information on the Net about where artificial
sound is actually synthesized and applied to, including the
doors.)
I did read recently that they have staffers charged with
designing the weird noises electric cars make at low speeds.
The Natural Philosopher <tnp@invalid.invalid> writes:
On 04/11/2026 14:54, Peter Flass wrote:
On 11/3/26 15:15, Pancho wrote:Not for transactuon processing
If you want heavy usage, I would guess bank risk calcs, weather calcs, >>>> lots of science applications, running on compute grids, (or whatever
the modern word is), blow this mainframe high i/o away.
I think back in the day the quants preferred Sun.
You clearly don't know what a "quant" is, in this context.
I did read recently that they have staffers charged with
designing the weird noises electric cars make at low speeds.
ASM was also *the* way for the older micro-controllers
-a for a long time. However some of the optimized 'C'
-a compilers can now rival ASM for speed and size and
-a you don't have to hand-code bit-banging serial either
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do it
in 'C'.
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler called "Aztec C" that had almost no overhead in excess libraries, start-
up code, etc., unless you asked for them.
On 11/7/2026 10:37 PM, Peter Flass wrote:
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler
called "Aztec C" that had almost no overhead in excess libraries, start-
up code, etc., unless you asked for them.
And nonconsensual AVX code.
I've heard some CPUs have a lower maximum
Hz when AVX is enabled.
There is the tool "pinfo", it can highlight somethings like the name of another man page, and jump to it.
Still, no table of content. No jumping from one section while browsing
the man page.
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler called "Aztec C" that had almost no overhead in excess libraries, start-
up code, etc., unless you asked for them.
On 11/7/26 09:37, Peter Flass wrote:
On 11/7/26 04:45, The Natural Philosopher wrote:
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least). When PCs were new I used a compiler
called "Aztec C" that had almost no overhead in excess libraries,
start- up code, etc., unless you asked for them.
-a Used Aztec ... indeed HAVE the compiler and EXTENSIVE
-a docs stashed away somewhere. Will have to make a new
-a CP/M-86 VM for it first. Not sure you'd get best effect
-a on a simulated Z80. Have seen some 'kit' Z80 boards.
-a There's also a DOS version - but it's apparently meant
-a to run under XP.
-a The docs are IMPRESSIVE - endless bit-level and flag
-a stuff - exactly what you needed to know if trying to
-a shoehorn something into a small system.
-a A few versions of Turbo Pascal allowed code overlays
-a in a fairly civil fashion, another way to squeeze a
-a big pgm into a small machine. But, better have a HDD,
-a pulling overlays from a floppy - you don't wanna
-a go there :-) Did a couple of large pgms where overlays
-a were needed.
-a Anyway, if interested in Aztec, try :
https://www.aztecmuseum.ca/compilers.htm
But then you add all the cruft that GCC drags in ...
On 11/7/26 04:45, The Natural Philosopher wrote:Hardly.
On 07/11/2026 04:09, c186282 wrote:
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C'
-a-a compilers can now rival ASM for speed and size and
-a-a you don't have to hand-code bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
But then you add all the cruft that GCC drags in and you're back where
you started (size-wise, at least).
When PCs were new I used a compiler
called "Aztec C" that had almost no overhead in excess libraries,
start-up code, etc., unless you asked for them.
On 07/11/2026 14:37, Peter Flass wrote:
On 11/7/26 04:45, The Natural Philosopher wrote:Hardly.
On 07/11/2026 04:09, c186282 wrote:But then you add all the cruft that GCC drags in and you're back where
ASM was also *the* way for the older micro-controllers
-a-a for a long time. However some of the optimized 'C' compilers can
-a-a now rival ASM for speed and size and you don't have to hand-code
-a-a bit-banging serial either
At some point I idly compiled some gcc C to assembler . I spent a
pointless half hour trying to work out if I could have done it better.
I couldn't..
In terms of modern micros I wouldn't even bother unless I couldn't do
it in 'C'.
you started (size-wise, at least).
When PCs were new I used a compiler called "Aztec C" that had almost noWell in general you DO ask for them
overhead in excess libraries, start-up code, etc., unless you asked for
them.
The 'mainframe' era is nearly over - to be replaced
by cluster/distributed/AI.
However IBM is already into all that, WILL make money.
Many baskets ... How To Survive/Thrive.
Hmmmm ... do remember when IBM was "typewriters" ...
1988, Nick Donofrio approves HA/6000, originally for NYTimes to move
their newspaper system ("ATEX") off DEC VAXCluster to RS/6000 (running project out at Los Gatos lab). I rename it HA/CMP https://en.wikipedia.org/wiki/IBM_High_Availability_Cluster_Multiprocessing when I start doing technical/scientific cluster scale-up with national
labs (LANL, LLNL, NCAR, etc) and commercial cluster scale-up with RDBMS vendors (Oracle, Sybase, Ingres, Informix) with VAXCluster support in
same source base with UNIX (planning on using Hursley 9333 for mid-range
and full FCS for both technical and commercial scaleup). DB2 was still
all mainframe source, IBM non-mainframe "DB2" had yet to be developed.
IBM S/88 (relogo'ed Stratus) Product Administrator started taking us
around to their customers and also had me write a section for the
corporate continuous availability strategy document (it gets pulled when
both AS400/Rochester and mainframe/POK complain they couldn't meet requirements). Had coined "disaster survivability" and "geographic survivability" (as counter to disaster/recovery) when out marketing
HA/CMP.
Around 2001, my group had developed a T1-over-UDP-over-Ethernet bridge
to use in tandem with our wireless ethernet bridge product. After we
spun out from WiLan, and became Afar, we also marketed it
separately. Our largest sale was to NYT, to replace the dual T1 lines
they had to their (16?) regional printing plants, connected to the phototypesetters in the print shops. At 42nd street they had a rack
with 60(?) of our boxes. While they wanted the two lines for each
remote site to be on separate boxes connected to separate IP networks,
two lines to different sites could share one of our units, which had
two T1 jacks.
System worked well; I once got to visit the Old Lady shortly before
they moved down the street.
Latest public benchmark I've found is 2010 z196 "Peak I/O" getting 2M
IOPS using 104 FICON (20K IOPS/FICON). About the same time a native FCS
is announced for E5-2600 server blades getting over million IOPS (two
such FCS higher throughput than 104 FICON running over FCS). Note IBM
docs has SAPs (system assist processors that do actual I/O) CPU be kept
to 70% ... or 1.5M IOPS ... also no CKD DASD have been made for decades
(just simulated with industry standard fixed-block devices).
... from which I learned that Stratus and Tandem both had figured
out how to do rolling upgrades in a non-stop environment; I had
thought that was an IBM mainframe invention.
Many techies despise System/Z as an overpriced dinosaur with an
obsolete architechture. But I wonder if it is any less expensive to
dip the Intel processors in that magic sauce. I bet IBM has better
scale to their builds.
There was then industry articles that server chip makers were
shipping half their products directly to megadatacenters. Shortly
later, IBM sells off its server blade business.
There was a paper on how Google did it, published 10 or 20 years ago.
At that point, they had a total complement of about 460,000 x86
servers. They would buy the cheapest component they could for every
component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are broken
at any moment? The users certainly wonrCOt notice.
Mainframes stopped being relevant back in the 1980s; every further development since then has just been driven by customer inertia.
Which kind of begs the question, doesnrCOt it: who bought it, and why?
If we define mainframes a large, very robust systems, they are still
very relevant. Although it may be true that the major technology was
in place before 1990 (non-stop cluster systems, transaction locking
with completion/rollback and strategic redundancy). What you call
customer inertia is also known as vendor lock-in.
On 11/8/2026 13:31, Lawrence DrCOOliveiro wrote:
There was a paper on how Google did it, published 10 or 20 years
ago. At that point, they had a total complement of about 460,000
x86 servers. They would buy the cheapest component they could for
every component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are
broken at any moment? The users certainly wonrCOt notice.
The parallelism of a Google is a very different beast from the
robust parallelism of a banking mainframe.
That is very different from transactions that must execute exactly
once.
Mainframes stopped being relevant back in the 1980s; every further
development since then has just been driven by customer inertia.
If we define mainframes a large, very robust systems, they are still
very relevant.
On Sun, 08 Nov 2026 08:32:30 -1000, Lynn Wheeler wrote:
There was then industry articles that server chip makers were
shipping half their products directly to megadatacenters. Shortly
later, IBM sells off its server blade business.
Which kind of begs the question, doesnrCOt it: who bought it, and why?
On Sun, 8 Nov 2026 13:45:03 -0800, Lars Poulsen wrote:
On 11/8/2026 13:31, Lawrence DrCOOliveiro wrote:
There was a paper on how Google did it, published 10 or 20 years
ago. At that point, they had a total complement of about 460,000
x86 servers. They would buy the cheapest component they could for
every component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are
broken at any moment? The users certainly wonrCOt notice.
The parallelism of a Google is a very different beast from the
robust parallelism of a banking mainframe.
Mainframes have never been that rCLrobustrCY. YourCOve mentioned elsewhere learning about Stratus and Tandem, who took robustness to a new level
back in the 1980s, beyond anything the mainframe world could match.
Think of that kind of redundancy multiplied tenfold, a hundredfold, a thousandfold, all built on open-source technologies: thatrCOs the kind
of world we live in today.
On 11/8/26 16:51, Lawrence DrCOOliveiro wrote:
On Sun, 8 Nov 2026 13:45:03 -0800, Lars Poulsen wrote:
On 11/8/2026 13:31, Lawrence DrCOOliveiro wrote:
There was a paper on how Google did it, published 10 or 20 years
ago. At that point, they had a total complement of about 460,000
x86 servers. They would buy the cheapest component they could for
every component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are
broken at any moment? The users certainly wonrCOt notice.
The parallelism of a Google is a very different beast from the
robust parallelism of a banking mainframe.
Mainframes have never been that rCLrobustrCY. YourCOve mentioned elsewhere >> learning about Stratus and Tandem, who took robustness to a new level
back in the 1980s, beyond anything the mainframe world could match.
Think of that kind of redundancy multiplied tenfold, a hundredfold, a
thousandfold, all built on open-source technologies: thatrCOs the kind
of world we live in today.
Mainframes typically achieve an uptime percentage between 99.999% (five nines) and 99.9999999% (nine nines)
Mainframes typically achieve an uptime percentage between 99.999%
(five nines) and 99.9999999% (nine nines)
On Sun, 8 Nov 2026 20:33:45 -0700, Peter Flass wrote:
Mainframes typically achieve an uptime percentage between 99.999%
(five nines) and 99.9999999% (nine nines)
The hyperscalers probably manage that well, if not better.
There is an article from 1986 on Bitsavers which describes having to
reboot an IBM mainframe to turn daylight saving on and off.
Mainframes typically achieve an uptime percentage between 99.999% (five nines) and 99.9999999% (nine nines)
On Sun, 8 Nov 2026 13:45:03 -0800, Lars Poulsen wrote:
On 11/8/2026 13:31, Lawrence DrCOOliveiro wrote:
There was a paper on how Google did it, published 10 or 20 years
ago. At that point, they had a total complement of about 460,000
x86 servers. They would buy the cheapest component they could for
every component but one: the power supply.
With that kind of scale, who cares if 100 or 1000 machines are
broken at any moment? The users certainly wonrCOt notice.
The parallelism of a Google is a very different beast from the
robust parallelism of a banking mainframe.
Mainframes have never been that rCLrobustrCY
learning about Stratus and Tandem, who took robustness to a new level
back in the 1980s, beyond anything the mainframe world could match.
Noted in this group though - I/O arch diffs between
mainframes and minis.
Noted in this group though - I/O arch diffs between
mainframes and minis.
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to
take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of
mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to
take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers
are built on an entirely different design philosophy:
- _everything_ fails all the time (law of big numbers: take the MTBF
for a compute component, run a lot of them and you've got a steady
rain of failures, e.g. say you've got an MTBF of 1 million hours and
you run 2 million of these devices, then on average you'll have a
failure every half hour)
c186282 <c186282@nnada.net> schrieb:
Noted in this group though - I/O arch diffs between
mainframes and minis.
I have a somewhat personal definition of a mainframe, which I also
includes the classical ones.
A mainframe is a computer that
- is very expensive
- takes large resources to run (a computer center)
- has very good performance in at least one field
- is not a cluster, massively parallel etc
On 11/10/26 02:50, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of
mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to >>> take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers
are built on an entirely different design philosophy:
- _everything_ fails all the time (law of big numbers: take the MTBF
for a compute component, run a lot of them and you've got a steady
rain of failures, e.g. say you've got an MTBF of 1 million hours and
you run 2 million of these devices, then on average you'll have a
failure every half hour)
Didn't I read somewhere that techs in large datacenters get around on >rollerblades to swap out failing components?
I read that, when minis came along, their first defining
characteristic was that they didn't need computer center resources
and fit in a single room.
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but
they have extremely high reliability (you can literally fire a
bullet through a CPU, and things will migrate smoothly - if you
don't hit the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much
higher reliability through massive parallelism, based on cheap, yet
highly redundant hardware running a Free Software stack that is not
priced per CPU. That is much more cost-effective and any equivalent
number of mainframes. And there is no rCLRecovery UnitrCY that you can
shoot at to take it out.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
On 2026-11-10, Thomas Koenig <tkoenig@netcologne.de> wrote:
I read that, when minis came along, their first defining
characteristic was that they didn't need computer center resources
and fit in a single room.
And you could plug them into a wall socket.
On 2026-11-09, Lawrence DrCOOliveiro wrote:
On Sun, 8 Nov 2026 20:33:45 -0700, Peter Flass wrote:
Mainframes typically achieve an uptime percentage between 99.999%
(five nines) and 99.9999999% (nine nines)
The hyperscalers probably manage that well, if not better.
There is an article from 1986 on Bitsavers which describes having to
reboot an IBM mainframe to turn daylight saving on and off.
*That* sounds like complicating a problem that would be made simpler by
not using the local time, or at least DST-adjusted local time.
On 11/10/26 02:50, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of
mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to >>> take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers
are built on an entirely different design philosophy:
- _everything_ fails all the time (law of big numbers: take the MTBF
for a compute component, run a lot of them and you've got a steady
rain of failures, e.g. say you've got an MTBF of 1 million hours and
you run 2 million of these devices, then on average you'll have a
failure every half hour)
Didn't I read somewhere that techs in large datacenters get around on rollerblades to swap out failing components?
Didn't I read somewhere that techs in large datacenters get around on >rollerblades to swap out failing components?
On 11/10/26 02:50, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of
mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to >>> take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY
setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers
are built on an entirely different design philosophy:
-a - _everything_ fails all the time (law of big numbers: take the MTBF
-a-a-a for a compute component, run a lot of them and you've got a steady
-a-a-a rain of failures, e.g. say you've got an MTBF of 1 million hours and >> -a-a-a you run 2 million of these devices, then on average you'll have a
-a-a-a failure every half hour)
Didn't I read somewhere that techs in large datacenters get around on rollerblades to swap out failing components?
On 11/10/26 09:37, Peter Flass wrote:
On 11/10/26 02:50, Alexander Schreiber wrote:
Lawrence DrCOOliveiro <ldo@nz.invalid> wrote:
On Tue, 10 Nov 2026 07:09:17 -0000 (UTC), Thomas Koenig wrote:
Today's mainframes can no longer compete on compute power, but they
have extremely high reliability (you can literally fire a bullet
through a CPU, and things will migrate smoothly - if you don't hit
the Recovery Unit) ...
Cloud providers like Amazon, Google, Microsoft etc offer much higher
reliability through massive parallelism, based on cheap, yet highly
redundant hardware running a Free Software stack that is not priced
per CPU. That is much more cost-effective and any equivalent number of >>>> mainframes. And there is no rCLRecovery UnitrCY that you can shoot at to >>>> take it out.
The nice thing is, you can run the same sort of rCLenterprise-classrCY >>>> setup that the big folks operate in your own homelab, because all the
foundation software they use is available to you, too
<https://docs.xcp-ng.org/vms/vm-migration/>.
Not exactly the same. The mainframe design philosophy includes "The
machine MUST NOT fail", for which one pays in complexity and system
cost. As long as your workload fits on that single machine (or a
small cluster (Parallel Sysplex)) and is of the "MUST NOT fail"
nature, fine.
Highly distributed systems like those run and offered by the hyperscalers >>> are built on an entirely different design philosophy:
-a - _everything_ fails all the time (law of big numbers: take the MTBF
-a-a-a for a compute component, run a lot of them and you've got a steady >>> -a-a-a rain of failures, e.g. say you've got an MTBF of 1 million hours and >>> -a-a-a you run 2 million of these devices, then on average you'll have a >>> -a-a-a failure every half hour)
Didn't I read somewhere that techs in large datacenters get around on
rollerblades to swap out failing components?
IS an issue !!!
Linux/UNIX are 'better' at this.
As for 'failures every half hour' ... can't
HAVE that !
According to Peter Flass <Peter@Iron-Spring.com>:
Didn't I read somewhere that techs in large datacenters get around on >>rollerblades to swap out failing components?
I don't think so. I've read that Google has people wandering around with grocery carts full of spare parts that they swap into failed servers.
| Sysop: | Amessyroom |
|---|---|
| Location: | Fayetteville, NC |
| Users: | 76 |
| Nodes: | 6 (0 / 6) |
| Uptime: | 54:21:39 |
| Calls: | 1,265 |
| Calls today: | 1 |
| Files: | 1,362 |
| D/L today: |
3 files (2,684K bytes) |
| Messages: | 285,563 |