From Newsgroup: comp.os.vms

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.
--
Waldek Hebisch
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

I suspect very few try to build VMS executables on Linux.

GNV got both ar and ld, but that does not help you.

But are you sure you want to do it this way?

My understanding is that you want to do:

source-(compiler on Linux)->.obj->(ld on Linux)->VMS .exe

I would go for:

source-(compiler on Linux)->.obj->(LINK on VMS)->VMS .exe

Or if possible even better:

source-(compiler on Linux)->.s-(GCC AS on VMS)->.obj-(LINK on VMS)->VMS .exe

GCC AS (version 2.8) is working on VMS Alpha.

Arne




--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

In article <108sckg$19hcn$1@paganini.bofh.team>,
Waldek Hebisch <antispam@fricas.org> wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

There is a BFD backend for VMS on Alpha in the main binutils
branch in git, but it is not clear to me how well-tested or
generally maintained it is, nor how compatibile with the
standard VMS utilities it ever was; I suspect it existed
primarily to port GNU software to VMS.

What, exactly, are you trying to do?

- Dan C.

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

I suspect very few try to build VMS executables on Linux.

GNV got both ar and ld, but that does not help you.

Where are the sources? Working sources certainly would help.

But are you sure you want to do it this way?

My understanding is that you want to do:

source-(compiler on Linux)->.obj->(ld on Linux)->VMS .exe

I would go for:

source-(compiler on Linux)->.obj->(LINK on VMS)->VMS .exe

That is plan B. Working binutils would be simpler because
I could use existing automated process.

Actually, ATM I getting errors during gcc build. I can use
crude workaround to go further, but I am affraid that to get
working compiler I need either working binutils or to examine
all build steps involving binutils and do equvalent work in
different way.

Or if possible even better:

source-(compiler on Linux)->.s-(GCC AS on VMS)->.obj-(LINK on VMS)->VMS .exe

GCC AS (version 2.8) is working on VMS Alpha.

Actually, it is possible that I will have to use

source-(compiler on Linux)->.s-(VMS assembler)>.obj...

because one of next goals is to get running compiler wich generate(d)
assembly for VMS assembler and may be incompatible with GNU as. But
this means more changes to build process (= work work => not better).
--
Waldek Hebisch
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 2025-08-29, Dan Cross <cross@spitfire.i.gajendra.net> wrote:

In article <108sckg$19hcn$1@paganini.bofh.team>,
Waldek Hebisch <antispam@fricas.org> wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

There is a BFD backend for VMS on Alpha in the main binutils
branch in git, but it is not clear to me how well-tested or
generally maintained it is, nor how compatibile with the
standard VMS utilities it ever was; I suspect it existed
primarily to port GNU software to VMS.

AdaCore was primarily responsible for working on this as part of
GNAT. After they dropped VMS as a supported target, this code may
have gone stale.

What, exactly, are you trying to do?

I strongly suspect the OP is trying to either run more modern variants
of languages that already exist on VMS or languages that simply do
not yet exist on VMS.

Simon.
--
Simon Clubley, clubley@remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 1:32 PM, Waldek Hebisch wrote:

Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

I suspect very few try to build VMS executables on Linux.

GNV got both ar and ld, but that does not help you.

Where are the sources? Working sources certainly would help.

There are the SourceForge version of GNV and there is the
VSI version of GNV.

The SF version is here:

https://sourceforge.net/p/gnv/ld_tools/ci/default/tree/reference/ld_tools/

Looks like SF GNV ld was last maintained 10 years ago.

But heck that is way later than VMS Alpha!!

Arne

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

Dan Cross <cross@spitfire.i.gajendra.net> wrote:

In article <108sckg$19hcn$1@paganini.bofh.team>,
Waldek Hebisch <antispam@fricas.org> wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

There is a BFD backend for VMS on Alpha in the main binutils
branch in git, but it is not clear to me how well-tested or
generally maintained it is, nor how compatibile with the
standard VMS utilities it ever was; I suspect it existed
primarily to port GNU software to VMS.

What, exactly, are you trying to do?

Well, this is part of bigger rather poorly defined project. ATM
I trying to compile gcc-15.1. I first build binutils, then I do

../gcc-15.1.0/configure --enable-languages=c,c++ --target=alpha-dec-vms --prefix=/sklad/usr
make -j 7 > mlogg 2>&1

There were earlier problems that I resolved. I have now trouble
during linking of libgcc. Some problems went away after switch
from binutils 2.43 to binutils 2.21. But now I get the following
error:

/sklad/usr/alpha-dec-vms/bin/ld: -shared not supported

That is 'ld' from binutils 2.21 says that it does not support
'-shared'. Note that Makefile wants to create 'libgcc_s.exe',
so it knows about VMS file extentions. In other words somebody
at least tried to make it work.

If you want more details I can provide them, but the message
about '-shared' appears also when I attempt to build a
different shared image:

/sklad/usr/alpha-dec-vms/bin/ld -shared -o pop_poll.exe pop_poll.obj

so the problem can be reproduced outside of gcc build.
--
Waldek Hebisch
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 1:32 PM, Waldek Hebisch wrote:

Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

But are you sure you want to do it this way?

My understanding is that you want to do:

source-(compiler on Linux)->.obj->(ld on Linux)->VMS .exe

I would go for:

source-(compiler on Linux)->.obj->(LINK on VMS)->VMS .exe

That is plan B. Working binutils would be simpler because
I could use existing automated process.

But couldn't you have your existing automated process do the
link on VMS via some fancy scripting?

I have little faith in ld producing good VMS .exe.

All the GCC ports to VMS I have seen (GCC 1.4x on VAX,
GCC 2.8.x on Alpha and GNAT 2.8.x on Alpha) have used
VMS linker. I suspect they have their reasons.

Actually, ATM I getting errors during gcc build. I can use
crude workaround to go further, but I am affraid that to get
working compiler I need either working binutils or to examine
all build steps involving binutils and do equvalent work in
different way.

Are you trying to get a recent GCC working on VMS?

In that case there may be some ideas from the GNAT
Itanium port done by the french:

https://github.com/AdaLabs/gnat-vms

Or if possible even better:

source-(compiler on Linux)->.s-(GCC AS on VMS)->.obj-(LINK on VMS)->VMS .exe >>
GCC AS (version 2.8) is working on VMS Alpha.

Actually, it is possible that I will have to use

source-(compiler on Linux)->.s-(VMS assembler)>.obj...

because one of next goals is to get running compiler wich generate(d) assembly for VMS assembler and may be incompatible with GNU as. But
this means more changes to build process (= work work => not better).

If you can generate VMS assembler, then you are all good.

Arne

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 1:39 PM, Simon Clubley wrote:

On 2025-08-29, Dan Cross <cross@spitfire.i.gajendra.net> wrote:

In article <108sckg$19hcn$1@paganini.bofh.team>,
Waldek Hebisch <antispam@fricas.org> wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

There is a BFD backend for VMS on Alpha in the main binutils
branch in git, but it is not clear to me how well-tested or
generally maintained it is, nor how compatibile with the
standard VMS utilities it ever was; I suspect it existed
primarily to port GNU software to VMS.

AdaCore was primarily responsible for working on this as part of
GNAT. After they dropped VMS as a supported target, this code may
have gone stale.

Not just AdaCore GNAT (Ada).

ProGIS also tried to make a business of GCC (C and C++) in the
same time frame (25 years ago).

Arne

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 2025-08-29, Arne Vajhoj <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

I suspect very few try to build VMS executables on Linux.

A decade ago, I got as far as getting a simple C program for VMS Alpha
to compile and link on Linux. Anything else more complicated than that
(ie: other languages) failed and I suspect that either bits were missing
from the public kits or the additional steps required were not obvious.

I discussed this at length in comp.os.vms at the time. Anyone interested
will have to rely on the notes I posted at the time as I have completely forgotten the details of any of this (and have no motivation to get back
up to speed on them because my hobbies these days are very different and because there's no longer a proper hobbyist licence for VMS Alpha).

GNV got both ar and ld, but that does not help you.

But are you sure you want to do it this way?

My understanding is that you want to do:

source-(compiler on Linux)->.obj->(ld on Linux)->VMS .exe

I would go for:

source-(compiler on Linux)->.obj->(LINK on VMS)->VMS .exe

I strongly suspect that will not work as the VMS linker may not
have the features needed to link those .o files.

Or if possible even better:

source-(compiler on Linux)->.s-(GCC AS on VMS)->.obj-(LINK on VMS)->VMS .exe

Er, why ? If you can get the BFD library working for VMS Alpha
once again, that's the major problem sorted.

GCC AS (version 2.8) is working on VMS Alpha.

Also, what makes you think that the GNU gas of that era will still
be suitable for today's compilers ?

Simon.
--
Simon Clubley, clubley@remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 2025-08-29, Waldek Hebisch <antispam@fricas.org> wrote:

Arne Vajhoj <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

I suspect very few try to build VMS executables on Linux.

GNV got both ar and ld, but that does not help you.

Where are the sources? Working sources certainly would help.

Aren't they simply wrappers around the native VMS utilities ?

But are you sure you want to do it this way?

My understanding is that you want to do:

source-(compiler on Linux)->.obj->(ld on Linux)->VMS .exe

I would go for:

source-(compiler on Linux)->.obj->(LINK on VMS)->VMS .exe

That is plan B. Working binutils would be simpler because
I could use existing automated process.

Binutils worked at one time because I was able to do the whole
sequence, including generating the final C VMS executable, on Linux.

Actually, ATM I getting errors during gcc build. I can use
crude workaround to go further, but I am affraid that to get
working compiler I need either working binutils or to examine
all build steps involving binutils and do equvalent work in
different way.

That's where I gave up. I either didn't understand the required
VMS-specific build steps or there were bits missing from the public
source kits.

Note that I have a lot of experience building cross compilers on Linux
for various embedded targets (including fixing various problems
I encountered along the way), and I have never had a fraction of the
problems I had when trying to get VMS support working.

Simon.
--
Simon Clubley, clubley@remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 2025-08-29, Arne Vajhoj <arne@vajhoej.dk> wrote:

I have little faith in ld producing good VMS .exe.

That bit worked just fine, but I don't remember if I had to fix up
anything in the BFD librayr first. The problem seems to be that the
current version of the BFD library has stagnated for VMS Alpha and
appears to need bringing up to date.

All the GCC ports to VMS I have seen (GCC 1.4x on VAX,
GCC 2.8.x on Alpha and GNAT 2.8.x on Alpha) have used
VMS linker. I suspect they have their reasons.

AdaCore started with a cross compiler on Linux.

Simon.
--
Simon Clubley, clubley@remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 1:57 PM, Simon Clubley wrote:

On 2025-08-29, Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

But are you sure you want to do it this way?

My understanding is that you want to do:

source-(compiler on Linux)->.obj->(ld on Linux)->VMS .exe

I would go for:

source-(compiler on Linux)->.obj->(LINK on VMS)->VMS .exe

I strongly suspect that will not work as the VMS linker may not
have the features needed to link those .o files.

If the .o files are not valid VMS .obj files, then it will
not work.

But the VMS Alpha target of gcc-as did generate valid
VMS .obj files back in time.

I don't see why anyone should have changed that since.

Or if possible even better:

source-(compiler on Linux)->.s-(GCC AS on VMS)->.obj-(LINK on VMS)->VMS .exe

Er, why ? If you can get the BFD library working for VMS Alpha
once again, that's the major problem sorted.

Getting it working 100% may be a huge effort.

Generating a text file and using the native tools on VMS
seems much less risk and effort to me.

GCC AS (version 2.8) is working on VMS Alpha.

Also, what makes you think that the GNU gas of that era will still
be suitable for today's compilers ?

I would expect Alpha assembler to still be Alpha assembler.

The instructions are obviously the same.

And I find it unlikely that anyone has spend time the last 25
years adding advanced macros and pseudo instructions to the
Alpha assembler and the compiler.

Given the status of Alpha.

Arne

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 2:11 PM, Simon Clubley wrote:

On 2025-08-29, Arne Vajh|+j <arne@vajhoej.dk> wrote:

All the GCC ports to VMS I have seen (GCC 1.4x on VAX,
GCC 2.8.x on Alpha and GNAT 2.8.x on Alpha) have used
VMS linker. I suspect they have their reasons.

AdaCore started with a cross compiler on Linux.

But did they use ld on Linux or LINK on VMS?

Arne




--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 2025-08-29, Arne Vajhoj <arne@vajhoej.dk> wrote:

On 8/29/2025 1:57 PM, Simon Clubley wrote:

Also, what makes you think that the GNU gas of that era will still
be suitable for today's compilers ?

I would expect Alpha assembler to still be Alpha assembler.

And are you sure that VMS gas 2.8 will support all of the following:

https://sourceware.org/binutils/docs/as/Pseudo-Ops.html

(Those are just the current target-neutral directives).

The instructions are obviously the same.

And I find it unlikely that anyone has spend time the last 25
years adding advanced macros and pseudo instructions to the
Alpha assembler and the compiler.

That's not where the problem is. You need to think at a higher level
than that. Notice that I said today's compilers which may rely on
underlying features not present in that old VMS Alpha version of gas.

Alpha is but one BFD target. Compilers do not talk to the VMS Alpha
gas directly. The following may help you understand the overall flow:

https://en.wikipedia.org/wiki/Binary_File_Descriptor_library

Simon.
--
Simon Clubley, clubley@remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 2:28 PM, Simon Clubley wrote:

On 2025-08-29, Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 1:57 PM, Simon Clubley wrote:

Also, what makes you think that the GNU gas of that era will still
be suitable for today's compilers ?

I would expect Alpha assembler to still be Alpha assembler.

And are you sure that VMS gas 2.8 will support all of the following:

https://sourceware.org/binutils/docs/as/Pseudo-Ops.html

(Those are just the current target-neutral directives).

The instructions are obviously the same.

And I find it unlikely that anyone has spend time the last 25
years adding advanced macros and pseudo instructions to the
Alpha assembler and the compiler.

That's not where the problem is. You need to think at a higher level
than that. Notice that I said today's compilers which may rely on
underlying features not present in that old VMS Alpha version of gas.

If some of those directives was not supported in GCC 2.8 as
and those directives are actually generated by the compiler
for the VMS Alpha target (because they are generated on all
target or was added to Alpha target later for the benefit
of other OS), then GCC 2.8 as will obviously not work.

It may still be easier to hack the compiler to not use
those than to reverse engineer LINK though.
Arne

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 2025-08-29, Arne Vajhoj <arne@vajhoej.dk> wrote:

On 8/29/2025 2:11 PM, Simon Clubley wrote:

On 2025-08-29, Arne Vajhoj <arne@vajhoej.dk> wrote:

All the GCC ports to VMS I have seen (GCC 1.4x on VAX,
GCC 2.8.x on Alpha and GNAT 2.8.x on Alpha) have used
VMS linker. I suspect they have their reasons.

AdaCore started with a cross compiler on Linux.

But did they use ld on Linux or LINK on VMS?

From what I remember, everything was pure GNU, including the use
of binutils, or at least it was by the time I started becoming
interested in this.

It's possible however that earlier efforts did involve native VMS
tools, but by the time I became interested in building binutils/gcc
from source the standard approach appears to have been via binutils.

Just had a quick look and there are patch emails online
showing AdaCore working on binutils to add and extend VMS support.

What I can't remember is if prior to this, AdaCore had their own
private gcc/binutils branch and if they only started pushing into
the public branches later on. I have some vague memories in this
area, but it's been so long ago, I could simply be wrong.

If anyone still has it installed, does the really old GNAT 3.11/3.12p
use ld or the VMS linker ? I know they used gas, but what about the
final link ?

Simon.
--
Simon Clubley, clubley@remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

In article <108splv$24jsc$2@dont-email.me>,
Simon Clubley <clubley@remove_me.eisner.decus.org-Earth.UFP> wrote:

On 2025-08-29, Arne Vajhoj <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

I suspect very few try to build VMS executables on Linux.

A decade ago, I got as far as getting a simple C program for VMS Alpha
to compile and link on Linux. Anything else more complicated than that
(ie: other languages) failed and I suspect that either bits were missing
from the public kits or the additional steps required were not obvious.

I discussed this at length in comp.os.vms at the time. Anyone interested
will have to rely on the notes I posted at the time as I have completely >forgotten the details of any of this (and have no motivation to get back
up to speed on them because my hobbies these days are very different and >because there's no longer a proper hobbyist licence for VMS Alpha).

GNV got both ar and ld, but that does not help you.

But are you sure you want to do it this way?

My understanding is that you want to do:

source-(compiler on Linux)->.obj->(ld on Linux)->VMS .exe

I would go for:

source-(compiler on Linux)->.obj->(LINK on VMS)->VMS .exe

I strongly suspect that will not work as the VMS linker may not
have the features needed to link those .o files.

What features? Presumably the `.o` files generated are actually
VMS `.OBJ` files. I suppose the compiler could generate generic
ELF or COFF or a.out objects for Alpha, and then rely on GNU LD
to do the right thing and convert them to a properly linked VMS
executable, but that seems a rather roundabout way to go.

Or if possible even better:

source-(compiler on Linux)->.s-(GCC AS on VMS)->.obj-(LINK on VMS)->VMS .exe

Er, why ? If you can get the BFD library working for VMS Alpha
once again, that's the major problem sorted.

GCC AS (version 2.8) is working on VMS Alpha.

Also, what makes you think that the GNU gas of that era will still
be suitable for today's compilers ?

I'd be rather surprised if much had changed in that department.

- Dan C.

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 2:44 PM, Simon Clubley wrote:

If anyone still has it installed, does the really old GNAT 3.11/3.12p
use ld or the VMS linker ? I know they used gas, but what about the
final link ?

I have Gnat 3.12p / GCC 2.8.1 installed.

I think it uses LINK.

<quote>
$ gnat make
List of available qualifiers and options

GNAT MAKE file /qualifiers (includes COMPILE /qualifiers)
gnatmake
...
/LINKER_QUALIFIERS (switches for LINK) -largs switches
...
</quote>

It mentions LINK.

<quote>
$ gnat make hello /verbose

GNATMAKE 3.12p (19990629) Copyright 1995-1999 Free Software Foundation,
Inc.
"hello.ali" being checked ...
-> "hello.ali" missing.
gcc -c hello.adb
gnatbind -x hello.ali
gnatlink hello.ali
GNU:[LIB]CRT0.OBJ
DISK2:[ARNE]B$HELLO.OBJ
DISK2:[]HELLO.OBJ
GNU:[LIB.OPENVMS7_1.2_8_1.DECLIB]LIBDECGNAT.OLB/library GNU:[LIB.OPENVMS7_1.2_8_1.ADALIB]LIBGNAT.EXE/shareable GNU:[LIB]LIBGCC.OLB/library
sys$library:vaxcrtltx.olb/library
GNU:[LIB]LIBGCC.OLB/library
</quote>

That looks very much like VMS LINK option format.

Arne


--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 2025-08-29, Arne Vajhoj <arne@vajhoej.dk> wrote:

On 8/29/2025 2:44 PM, Simon Clubley wrote:

If anyone still has it installed, does the really old GNAT 3.11/3.12p
use ld or the VMS linker ? I know they used gas, but what about the
final link ?

I have Gnat 3.12p / GCC 2.8.1 installed.

I think it uses LINK.

<quote>
$ gnat make
List of available qualifiers and options

GNAT MAKE file /qualifiers (includes COMPILE /qualifiers)
gnatmake
...
/LINKER_QUALIFIERS (switches for LINK) -largs switches
...
</quote>

It mentions LINK.

Interesting thanks. Does the "ld" binary exist in whatever the binutils executables directory is ?

If it does, and you run it as a foreign command with "--help", what
output do you get ?

Simon.
--
Simon Clubley, clubley@remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 3:10 PM, Simon Clubley wrote:

On 2025-08-29, Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 2:44 PM, Simon Clubley wrote:

If anyone still has it installed, does the really old GNAT 3.11/3.12p
use ld or the VMS linker ? I know they used gas, but what about the
final link ?

I have Gnat 3.12p / GCC 2.8.1 installed.

I think it uses LINK.

<quote>
$ gnat make
List of available qualifiers and options

GNAT MAKE file /qualifiers (includes COMPILE /qualifiers)
gnatmake
...
/LINKER_QUALIFIERS (switches for LINK) -largs switches
...
</quote>

It mentions LINK.

Interesting thanks. Does the "ld" binary exist in whatever the binutils executables directory is ?

If it does, and you run it as a foreign command with "--help", what
output do you get ?

Nothing with ld.

$ dir gnu:[bin]

Directory GNU:[BIN]

AS.EXE;10 ASISTANT.EXE;1 GCC.EXE;1 GNAT.EXE;1 GNATBIND.EXE;1 GNATBL.EXE;1 GNATCHOP.EXE;2 GNATELIM.EXE;1 GNATFIND.EXE;1 GNATKR.EXE;1 GNATLBR.EXE;1 GNATLINK.EXE;1 GNATLS.EXE;1 GNATMAKE.EXE;1 GNATMEM.EXE;1 GNATPREP.EXE;1 GNATPSTA.EXE;1 GNATPSYS.EXE;1 GNATSTUB.EXE;1 GNATXREF.EXE;1 MAKE.EXE;260 OPENVMS7_1-GCC.EXE;1 SET_EXE.COM;11

$ dir gnu:[bin]*ld*.*
%DIRECT-W-NOFILES, no files found

Arne


--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 3:14 PM, Arne Vajh|+j wrote:

On 8/29/2025 3:10 PM, Simon Clubley wrote:

On 2025-08-29, Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 2:44 PM, Simon Clubley wrote:

If anyone still has it installed, does the really old GNAT 3.11/3.12p
use ld or the VMS linker ? I know they used gas, but what about the
final link ?

I have Gnat 3.12p / GCC 2.8.1 installed.

I think it uses LINK.

<quote>
$ gnat make
List of available qualifiers and options

GNAT MAKE file /qualifiers (includes COMPILE /qualifiers)
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a gnatmake
-a-a-a-a ...
-a-a-a-a /LINKER_QUALIFIERS (switches for LINK)-a-a-a-a-a-a-a-a-a-a -largs switches
-a-a-a-a ...
</quote>

It mentions LINK.

Interesting thanks. Does the "ld" binary exist in whatever the binutils
executables directory is ?

If it does, and you run it as a foreign command with "--help", what
output do you get ?

Nothing with ld.

$ dir gnu:[bin]

Directory GNU:[BIN]

AS.EXE;10-a-a-a-a-a-a-a-a-a-a ASISTANT.EXE;1-a-a-a-a-a GCC.EXE;1-a-a-a-a-a-a-a-a-a-a GNAT.EXE;1
GNATBIND.EXE;1-a-a-a-a-a GNATBL.EXE;1-a-a-a-a-a-a-a GNATCHOP.EXE;2-a-a-a-a-a GNATELIM.EXE;1
GNATFIND.EXE;1-a-a-a-a-a GNATKR.EXE;1-a-a-a-a-a-a-a GNATLBR.EXE;1-a-a-a-a-a-a GNATLINK.EXE;1
GNATLS.EXE;1-a-a-a-a-a-a-a GNATMAKE.EXE;1-a-a-a-a-a GNATMEM.EXE;1-a-a-a-a-a-a GNATPREP.EXE;1
GNATPSTA.EXE;1-a-a-a-a-a GNATPSYS.EXE;1-a-a-a-a-a GNATSTUB.EXE;1-a-a-a-a-a GNATXREF.EXE;1
MAKE.EXE;260-a-a-a-a-a-a-a OPENVMS7_1-GCC.EXE;1-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a SET_EXE.COM;11

$ dir gnu:[bin]*ld*.*
%DIRECT-W-NOFILES, no files found

$ link=="XXXX"
$ link
%DCL-W-IVVERB, unrecognized command verb - check validity and spelling
\XXXX\
$ gnat make hello
gcc -c hello.adb
gnatbind -x hello.ali
gnatlink hello.ali
%DCL-W-IVVERB, unrecognized command verb - check validity and spelling
\XXXX\
gnatmake: *** link failed.

Arne

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

In article <108sp9v$1a4ae$2@paganini.bofh.team>,
Waldek Hebisch <antispam@fricas.org> wrote:

Dan Cross <cross@spitfire.i.gajendra.net> wrote:

In article <108sckg$19hcn$1@paganini.bofh.team>,
Waldek Hebisch <antispam@fricas.org> wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

There is a BFD backend for VMS on Alpha in the main binutils
branch in git, but it is not clear to me how well-tested or
generally maintained it is, nor how compatibile with the
standard VMS utilities it ever was; I suspect it existed
primarily to port GNU software to VMS.

What, exactly, are you trying to do?

Well, this is part of bigger rather poorly defined project. ATM
I trying to compile gcc-15.1.

That's a pretty big project in itself. I expect, as others have
noted and I mentioned earlier, that the port to VMS has fallen
into disrepair. Getting it back up to snuff will likely be
hard.

I first build binutils, then I do

../gcc-15.1.0/configure --enable-languages=c,c++ --target=alpha-dec-vms --prefix=/sklad/usr
make -j 7 > mlogg 2>&1

There were earlier problems that I resolved. I have now trouble
during linking of libgcc. Some problems went away after switch
from binutils 2.43 to binutils 2.21. But now I get the following
error:

/sklad/usr/alpha-dec-vms/bin/ld: -shared not supported

That is 'ld' from binutils 2.21 says that it does not support
'-shared'. Note that Makefile wants to create 'libgcc_s.exe',
so it knows about VMS file extentions. In other words somebody
at least tried to make it work.

It definitely ran on VMS in the past. I don't think anyone has
actively tried to remove VMS support. So it's very likely you
are getting the vestigal remnants on that prior porting effort,
along with a lot of what one might refer to as "soft" bugs; that
is, things that are not outright _wrong_ in the code, but which
exhibit incorrect behavior on VMS because no one has been
actively maintaining the GCC VMS port.

If you want more details I can provide them, but the message
about '-shared' appears also when I attempt to build a
different shared image:

/sklad/usr/alpha-dec-vms/bin/ld -shared -o pop_poll.exe pop_poll.obj

so the problem can be reproduced outside of gcc build.

I mean, in a way the meaning of the message is obvious: that
flag is not supported on the version of binutils you built for
VMS. Merely getting confirmation of that is unlikely to yield
much in the way of insightful new information, as it says it
right on the tin.

The more pressing question is *why* is the flag unsupported?
Figure that out and you'll likely unlock the entire thing.

- Dan C.

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 1:50 PM, Waldek Hebisch wrote:

There were earlier problems that I resolved. I have now trouble
during linking of libgcc. Some problems went away after switch
from binutils 2.43 to binutils 2.21. But now I get the following
error:

/sklad/usr/alpha-dec-vms/bin/ld: -shared not supported

That is 'ld' from binutils 2.21 says that it does not support
'-shared'. Note that Makefile wants to create 'libgcc_s.exe',
so it knows about VMS file extentions. In other words somebody
at least tried to make it work.

If you want more details I can provide them, but the message
about '-shared' appears also when I attempt to build a
different shared image:

/sklad/usr/alpha-dec-vms/bin/ld -shared -o pop_poll.exe pop_poll.obj

so the problem can be reproduced outside of gcc build.

https://sourceware.org/git/?p=binutils-gdb.git;a=blob;f=bfd/vms-alpha.c;h=c4b248494ee624f7b1d6acbf45065469cda7c682;hb=HEAD

says:

<quote>
/* vms.c -- BFD back-end for EVAX (openVMS/Alpha) files.
Copyright (C) 1996-2025 Free Software Foundation, Inc.

Initial version written by Klaus Kaempf (kkaempf@rmi.de)
Major rewrite by Adacore.

This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
MA 02110-1301, USA. */

/* TODO:
o overlayed sections
o PIC
o Generation of shared image
o Relocation optimizations
o EISD for the stack
o Vectors isect
o 64 bits sections
o Entry point
o LIB$INITIALIZE
o protected sections (for messages)
...
*/
</quote>

It is on the TODO list.

Which of course does not help you much.

Arne

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

On 8/29/2025 1:57 PM, Simon Clubley wrote:

On 2025-08-29, Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

I suspect very few try to build VMS executables on Linux.

A decade ago, I got as far as getting a simple C program for VMS Alpha
to compile and link on Linux. Anything else more complicated than that
(ie: other languages) failed and I suspect that either bits were missing
from the public kits or the additional steps required were not obvious.

I discussed this at length in comp.os.vms at the time. Anyone interested
will have to rely on the notes I posted at the time as I have completely forgotten the details of any of this (and have no motivation to get back
up to speed on them because my hobbies these days are very different and because there's no longer a proper hobbyist licence for VMS Alpha).

I did the search:

https://groups.google.com/g/comp.os.vms/c/KN9yqM7_8PU/m/T4xwyTz5A3oJ

https://groups.google.com/g/comp.os.vms/c/S7-VwQmmOZU/m/LXQ2TrbCBAAJ

https://groups.google.com/g/comp.os.vms/c/oSNzXLKeu4o/m/8Tq5dIFaDAAJ

Arne

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 1:32 PM, Waldek Hebisch wrote:

Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

I suspect very few try to build VMS executables on Linux.

GNV got both ar and ld, but that does not help you.

Where are the sources? Working sources certainly would help.

There are the SourceForge version of GNV and there is the
VSI version of GNV.

The SF version is here:

https://sourceforge.net/p/gnv/ld_tools/ci/default/tree/reference/ld_tools/

Thanks. AFAICS main work is done by VMS binaries, so it does not
help in my immediate problem, but it contains interesting information.
--
Waldek Hebisch
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.os.vms

Simon Clubley <clubley@remove_me.eisner.decus.org-earth.ufp> wrote:

On 2025-08-29, Waldek Hebisch <antispam@fricas.org> wrote:

Arne Vajh|+j <arne@vajhoej.dk> wrote:

On 8/29/2025 10:14 AM, Waldek Hebisch wrote:

I wonder if anybody is using GNU binutils targeting VMS. I am
trying to set up toolchain for cross-compilation (from Linux) to
Alpha VMS. Trying reasonably recent (2.43 and 2.39) binutils I
get errors suggesting serious brokennes, for example when I use
'ar' to create a library and later extract a member I get changed
file and 'objdump' says that extracted file is malformed.

I am now trying binutils 2.21 and apparently they do not have the
problem above. But I see some weirdness and apparently '-shared'
option (to create shared image) is not supported.

I suspect very few try to build VMS executables on Linux.

GNV got both ar and ld, but that does not help you.

Where are the sources? Working sources certainly would help.

Aren't they simply wrappers around the native VMS utilities ?

Unfortunately yes.

But are you sure you want to do it this way?

My understanding is that you want to do:

source-(compiler on Linux)->.obj->(ld on Linux)->VMS .exe

I would go for:

source-(compiler on Linux)->.obj->(LINK on VMS)->VMS .exe

That is plan B. Working binutils would be simpler because
I could use existing automated process.

Binutils worked at one time because I was able to do the whole
sequence, including generating the final C VMS executable, on Linux.

Thanks, good to know this.

Actually, ATM I getting errors during gcc build. I can use
crude workaround to go further, but I am affraid that to get
working compiler I need either working binutils or to examine
all build steps involving binutils and do equvalent work in
different way.

That's where I gave up. I either didn't understand the required
VMS-specific build steps or there were bits missing from the public
source kits.

There are missing pieces, build needs VMS header files and libraries.
If those pieces are incomplete, then configuration may go wrong
and lead to trobles later.

Currently I have fairly incomplete set of header files, so I expect
trouble due to this. OTOH libgcc seem to compile fine, so C
compiler probably will work. I worked around trouble with
shared images by configuring gcc with '--disable-shared',
suboptimal but must do for now.
--
Waldek Hebisch
--- Synchronet 3.21a-Linux NewsLink 1.2