From Newsgroup: comp.lang.tcl

I was not expecting this to throw an error. Docs says it should return 0
or 1. How would you do this check?

% package req Thread
2.8.5

% thread::exists ""
invalid thread handle ""

% thread::exists abc
invalid thread handle "abc"


But then this works:

% thread::exists tid1
0

Weird. It looks like the check expects the argument to start with "tid"
and a number.

--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

Am 08.02.2026 um 20:05 schrieb saito:

I was not expecting this to throw an error. Docs says it should return 0
or 1. How would you do this check?

% package req Thread
2.8.5

% thread::exists ""
invalid thread handle ""

% thread::exists abc
invalid thread handle "abc"

But then this works:

% thread::exists tid1
0

Weird. It looks like the check expects the argument to start with "tid"
and a number.

Yes, please file a bug report:

https://core.tcl-lang.org/thread/ticket

Thanks,
Harald
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On 2/8/26 20:26, Harald Oehlmann wrote:

Am 08.02.2026 um 20:05 schrieb saito:

I was not expecting this to throw an error. Docs says it should return
0 or 1. How would you do this check?

% package req Thread
2.8.5

% thread::exists ""
invalid thread handle ""

% thread::exists abc
invalid thread handle "abc"


But then this works:

% thread::exists tid1
0

Weird. It looks like the check expects the argument to start with
"tid" and a number.

Yes, please file a bug report:

https://core.tcl-lang.org/thread/ticket

Honestly am I puzzled. Why should this non-problem deserve a bug report?
The fact to construct a thread identifier out of blue strings seems to
me to be a somehow somewhat malicious act. So how can there be any
expectations on arbitrary generated thread identifiers being error'd?
Or answered as non existing?

In other words: the errors thrown on impossible thread identifiers is
the correct answer for the query on thread existence.

Finally, IMO please do not create a ticket.

BR,
Christian
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On 2/8/26 22:40, undroidwish wrote:

Finally, IMO please do not create a ticket.

... and simply live with the fact that some strings are
impossible thread identifiers, and some not, but don't
represent existing threads.

BR,
Christian

--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On 2/8/2026 4:53 PM, Robert Heller wrote:

At Sun, 8 Feb 2026 22:40:39 +0100 undroidwish <undroidwish@googlemail.com> wrote:

Finally, IMO please do not create a ticket.

The OP is NOT expecting thread::exists to behave any differently than it does,
just that thread::exists behavior be completely and accurately documented. Appearently, the documentation is mute as to what thread::exists does when given a string that is not a proper thread id, with the (erronious) *presumption* that it would return 0 (which is in fact NOT what it does). This
is a *documentation* bug. And the suggested bug report relates to the *documentation*, not the actual code.

It started like this but it looks like thread::exists does not
necessarily function 100%. Documentation can fix this easily by
requiring a certain format for the id's. But then you will be left to
add extra code on your own which I think is the purpose of thread::exists.


--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On 2/8/2026 2:32 PM, saito wrote:

If it is important, then it should be documented. You think it is "tid" followed by numbers? Not so. Why is "tidabc" ok here but not "tidxyz"?

% thread::exists tidabc
0

% thread::exists tidxyz
invalid thread handle "tidxyz"

The code that checks is,

static int
ThreadGetId(interp, handleObj, thrIdPtr)
Tcl_Interp *interp;
Tcl_Obj *handleObj;
Tcl_ThreadId *thrIdPtr;
{
const char *thrHandle = Tcl_GetString(handleObj);

if (sscanf(thrHandle, THREAD_HNDLPREFIX"%p", thrIdPtr) == 1) {
return TCL_OK;
}

Tcl_AppendResult(interp, "invalid thread handle \"",
thrHandle, "\"", NULL);
return TCL_ERROR;
}


So, it depends on how sscanf treats %p, and in most cases this has to be a valid pointer, usually a hex number, and abc is valid hex, while xyz is not.

-e

--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

Hi Chrstian, all,

I am very happy about the discussion.

Ihe intention to file a bug report is to have this discussion in a
ticket and not here. So the discussion may be found later.

About "error or not".
Most "exists" functions in TCL don't throw errors on invalid arguments.
For example,
dict exits {} a b
has thrown an error in an earlier version of TCL.
This was changed to not throw an error.

Thanks for all,
Harald
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On 2/9/2026 2:22 AM, Harald Oehlmann wrote:

Hi Chrstian, all,

I am very happy about the discussion.

Ihe intention to file a bug report is to have this discussion in a
ticket and not here. So the discussion may be found later.

That sounds good. I will try to file the report sometime this week when
I get the chance.


--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

Am 09.02.26 um 08:22 schrieb Harald Oehlmann:

Hi Chrstian, all,

I am very happy about the discussion.

Ihe intention to file a bug report is to have this discussion in a
ticket and not here. So the discussion may be found later.

About "error or not".
Most "exists" functions in TCL don't throw errors on invalid arguments.
For example,
dict exits {} a b
has thrown an error in an earlier version of TCL.
This was changed to not throw an error.

I tend to agree with the other Christian. If Tcl had types, then the
argument to thread::exists would be a Thread identifier and there would
not be a way to pass some artbitrary object to it. The purpose of this function is to see, if a thread identifier - something you got back from thread::create - corresponds to a running thread. If you pass it
something which is not a thread identifier, than there is a bug in your program.

Christian
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

At Mon, 9 Feb 2026 20:44:13 +0100 Christian Gollwitzer <auriocus@gmx.de> wrote:

Am 09.02.26 um 08:22 schrieb Harald Oehlmann:

Hi Chrstian, all,

I am very happy about the discussion.

Ihe intention to file a bug report is to have this discussion in a
ticket and not here. So the discussion may be found later.

About "error or not".
Most "exists" functions in TCL don't throw errors on invalid arguments.
For example,
dict exits {} a b
has thrown an error in an earlier version of TCL.
This was changed to not throw an error.

I tend to agree with the other Christian. If Tcl had types, then the
argument to thread::exists would be a Thread identifier and there would
not be a way to pass some artbitrary object to it. The purpose of this function is to see, if a thread identifier - something you got back from thread::create - corresponds to a running thread. If you pass it
something which is not a thread identifier, than there is a bug in your program.

The problem is that this is NOT documented. The documentation should document this behaviour.

Christian

--
Robert Heller -- Cell: 413-658-7953 GV: 978-633-5364
Deepwoods Software -- Custom Software Services
http://www.deepsoft.com/ -- Linux Administration Services
heller@deepsoft.com -- Webhosting Services
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On 2/9/2026 2:44 PM, Christian Gollwitzer wrote:

Am 09.02.26 um 08:22 schrieb Harald Oehlmann:

If you pass it
something which is not a thread identifier, than there is a bug in your program.

In that case, I am afraid we may need a new command to check if
something is a valid thread identifier, not just a real/active one. I
get it, internally it makes sense that the id's have a nice logic and
format to it. I am of the opinion that this detail can remain internal.

My original question remains: "How would you do this check?" when you
receive a thread id from an evil user or colleague or simply by mistake?

--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On Mon, 9 Feb 2026 20:44:13 +0100
Christian Gollwitzer <auriocus@gmx.de> wrote:

[...]

I tend to agree with the other Christian. If Tcl had types, then the argument to thread::exists would be a Thread identifier and there would
not be a way to pass some artbitrary object to it. The purpose of this function is to see, if a thread identifier - something you got back from thread::create - corresponds to a running thread. If you pass it
something which is not a thread identifier, than there is a bug in your program.

Well, I agree with Harald here. The semantics of "exists" subcommands is
to return true if the thing exists, or false otherwise, without considering *why* it doesn't exists; it doesn't, so return 0. There are plenty of
examples in Tcl (and Tk):

# namespace foo::bar doesn't exists
% info exists foo::bar::baz
0
% namespace exists foo::bar::baz
0
% namespace ensemble exists foo::bar::baz
0
# "foo" is not a valid dictionary
% dict exists foo a
0
# "Nosuchwindow" is not a valid window name
% package require Tk; winfo exists Nosuchwindow
0

Having to test if the handle conforms to certain naming convention and
only then test whether it exists or not is, IMO, a bug. Not to mention
that values such as "tidabcXYZ AND SOME NONSENSE", which is an invalid
thread identifier, will pass the test.

Regards
--
Emiliano <emil.gavilan@gmail.com>
--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

On 2/9/26 23:47, saito wrote:

...
My original question remains: "How would you do this check?" when you receive a thread id from an evil user or colleague or simply by mistake?

Due to the nature of threads their sheer existence is bound to their
containing process. Are there real use cases where a thread identifier
is given to the process' outer world? Yes, I know, some POSIX IPC means
can be process shared, but those are mutexen and the like, threads not
so much. Ahem, well there might be the need to kill a thread, but then
again, the Tcl thread identifier doesn't make sense either, since the
native thread id is needed which most likely does not equal the Tcl
thread identifier.

BR,
Christian

--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

Am 10.02.26 um 23:06 schrieb undroidwish:

On 2/9/26 23:47, saito wrote:

...
My original question remains: "How would you do this check?" when you
receive a thread id from an evil user or colleague or simply by mistake?

Due to the nature of threads their sheer existence is bound to their containing process. Are there real use cases where a thread identifier
is given to the process' outer world?

Exactly that was my reasoning. If you pass anything as a thread
identifier to any of the thread:: functions, and it is not what you got
from thread::create, then you have a bug in your program. The thread ID
is an opaque handle and should be treated like a pointer in C. You don't
write a pointer to a file and read it back later, your program will
crash. I could only see the need for a sentinel value, like the empty
string, so that you can initialize a variable.

The Other Christian


--- Synchronet 3.21b-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

* Christian Gollwitzer <auriocus@gmx.de>
| Am 10.02.26 um 23:06 schrieb undroidwish:
| > On 2/9/26 23:47, saito wrote:
| >
| >> ...
| >> My original question remains: "How would you do this check?" when
| >> you receive a thread id from an evil user or colleague or simply by
| >> mistake?
| > Due to the nature of threads their sheer existence is bound to their
| > containing process. Are there real use cases where a thread identifier
| > is given to the process' outer world?

| Exactly that was my reasoning. If you pass anything as a thread
| identifier to any of the thread:: functions, and it is not what you
| got from thread::create, then you have a bug in your program.

It should be documented, I think that much is undisputed.

If anyone needs a threadid-safe version (pun intended :-) of thread::exists,

proc thread_exists_safe {id} {
return [expr {![catch {thread::exists $id} exists] && $exists}]
}

is quickly coded (even without asking ChatGPT)...

R'
--- Synchronet 3.21b-Linux NewsLink 1.2