bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Rolf Ade]

With 29.1 as well as with trunk(both tested with -Q) if I call M-x gdb I
get asked:

"Enable querying debuginfod servers for this session? (y or n)"

Even if I answer with "n" I get the message:

"No symbol \"debuginfod\" in current context."

This can't be right. Declining to use a feature should not be answered
with a message that it would not be available anyways (what the message
seem to tell me).

Digging around a bit it turns out there is a variable to customise the
behaviour here: gdb-debuginfod-enable-setting. Its default value is
'ask.

From a user perspective it only make sense to be ask about wanting
something if this something is avaliable.

If it is about the advertising aspect then why not more clear don't ask
questions and instead write a message like: "You cannot use debuginfod
servers. Consider using this marvelous feature".

rolf

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Eli Zaretskii]

> From: Rolf Ade <rolf@pointsman.de>
> Date: Sat, 04 May 2024 04:00:20 +0200
> 
> 
> With 29.1 as well as with trunk(both tested with -Q) if I call M-x gdb I
> get asked:
> 
> "Enable querying debuginfod servers for this session? (y or n)"
> 
> Even if I answer with "n" I get the message:
> 
> "No symbol \"debuginfod\" in current context."

I think this means your GDB is too old and doesn't support debug info
downloading, or was built without that support.  We don't know how to
know in advance that this is not supported, thus the user option that
you can customize to avoid these questions if your GDB cannot use
debuginfod.

What OS are you using GDB on, and what is your GDB version, please?

> If it is about the advertising aspect then why not more clear don't ask
> questions and instead write a message like: "You cannot use debuginfod
> servers. Consider using this marvelous feature".

It isn't about advertising.  The question is real (and important when
the feature is supported).

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Rolf Ade]

Am 04.05.24 um 09:20 schrieb Eli Zaretskii:
>> From: Rolf Ade <rolf@pointsman.de>
>> Date: Sat, 04 May 2024 04:00:20 +0200
>>
>>
>> With 29.1 as well as with trunk(both tested with -Q) if I call M-x gdb I
>> get asked:
>>
>> "Enable querying debuginfod servers for this session? (y or n)"
>>
>> Even if I answer with "n" I get the message:
>>
>> "No symbol \"debuginfod\" in current context."
> 
> I think this means your GDB is too old and doesn't support debug info
> downloading, or was built without that support.  We don't know how to
> know in advance that this is not supported, thus the user option that
> you can customize to avoid these questions if your GDB cannot use
> debuginfod.
> 
> What OS are you using GDB on, and what is your GDB version, please?
> 
>> If it is about the advertising aspect then why not more clear don't ask
>> questions and instead write a message like: "You cannot use debuginfod
>> servers. Consider using this marvelous feature".
> 
> It isn't about advertising.  The question is real (and important when
> the feature is supported).

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Rolf Ade]

Am 04.05.24 um 09:20 schrieb Eli Zaretskii:
>> From: Rolf Ade <rolf@pointsman.de>
>> Date: Sat, 04 May 2024 04:00:20 +0200
>>
>>
>> With 29.1 as well as with trunk(both tested with -Q) if I call M-x gdb I
>> get asked:
>>
>> "Enable querying debuginfod servers for this session? (y or n)"
>>
>> Even if I answer with "n" I get the message:
>>
>> "No symbol \"debuginfod\" in current context."
> 
> I think this means your GDB is too old and doesn't support debug info
> downloading, or was built without that support.  We don't know how to
> know in advance that this is not supported, thus the user option that
> you can customize to avoid these questions if your GDB cannot use
> debuginfod.

I don't know if my gdb support debug info downloading (see exact version info 
below) but if I answer the question with "n" - I don't want to enable this - I 
wonder why I'm (only mildly, of course) bothered with a not very helpful error 
message.

> What OS are you using GDB on, and what is your GDB version, please?

This is basically a GNU/linux debian 11.9 (so, not that old). More specifically, 
uname -a:

Linux pointsman2 5.10.0-28-amd64 #1 SMP Debian 5.10.209-2 (2024-01-31) x86_64 
GNU/Linux

gdb --version (it's the version installed by the debian package management system):

GNU gdb (Debian 10.1-1.7) 10.1.90.20210103-git

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Eli Zaretskii]

> Date: Sat, 4 May 2024 15:32:27 +0200
> From: Rolf Ade <rolf@pointsman.de>
> 
> >> Even if I answer with "n" I get the message:
> >>
> >> "No symbol \"debuginfod\" in current context."
> > 
> > I think this means your GDB is too old and doesn't support debug info
> > downloading, or was built without that support.  We don't know how to
> > know in advance that this is not supported, thus the user option that
> > you can customize to avoid these questions if your GDB cannot use
> > debuginfod.
> 
> I don't know if my gdb support debug info downloading (see exact version info 
> below) but if I answer the question with "n" - I don't want to enable this - I 
> wonder why I'm (only mildly, of course) bothered with a not very helpful error 
> message.

I don't know.  AFAIR, it was not supposed to happen in Emacs 29.1 and
later.

> > What OS are you using GDB on, and what is your GDB version, please?
> 
> This is basically a GNU/linux debian 11.9 (so, not that old). More specifically, 
> uname -a:
> 
> Linux pointsman2 5.10.0-28-amd64 #1 SMP Debian 5.10.209-2 (2024-01-31) x86_64 
> GNU/Linux
> 
> gdb --version (it's the version installed by the debian package management system):
> 
> GNU gdb (Debian 10.1-1.7) 10.1.90.20210103-git

Strange.  GDB 10.1 was the version where debuginfod support was added.
What does the shell command below show?

  $ gdb --config | fgrep debuginfod

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Eli Zaretskii]

> Cc: 70764@debbugs.gnu.org
> Date: Sat, 04 May 2024 17:11:52 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> 
> > Date: Sat, 4 May 2024 15:32:27 +0200
> > From: Rolf Ade <rolf@pointsman.de>
> > 
> > I don't know if my gdb support debug info downloading (see exact version info 
> > below) but if I answer the question with "n" - I don't want to enable this - I 
> > wonder why I'm (only mildly, of course) bothered with a not very helpful error 
> > message.
> 
> I don't know.  AFAIR, it was not supposed to happen in Emacs 29.1 and
> later.

FWIW, I just tried "M-x gdb" in Emacs 29.1, and didn't see any "No
symbol" messages after answering "n" to the debuginfod question.

My GDB on that system is version 12.1, though, and it was compiled
with debuginfod support.  So maybe that's the reason.

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Rolf Ade]

Am 04.05.24 um 16:11 schrieb Eli Zaretskii:
>> Date: Sat, 4 May 2024 15:32:27 +0200
>> From: Rolf Ade <rolf@pointsman.de>
>>
>>>> Even if I answer with "n" I get the message:
>>>>
>>>> "No symbol \"debuginfod\" in current context."
>>>
[...]
>>> What OS are you using GDB on, and what is your GDB version, please?
>>
>> This is basically a GNU/linux debian 11.9 (so, not that old). More specifically,
>> uname -a:
>>
>> Linux pointsman2 5.10.0-28-amd64 #1 SMP Debian 5.10.209-2 (2024-01-31) x86_64
>> GNU/Linux
>>
>> gdb --version (it's the version installed by the debian package management system):
>>
>> GNU gdb (Debian 10.1-1.7) 10.1.90.20210103-git
> 
> Strange.  GDB 10.1 was the version where debuginfod support was added.
> What does the shell command below show?
> 
>    $ gdb --config | fgrep debuginfod

This returns

              --with-debuginfod

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Rolf Ade]

Am 04.05.24 um 16:23 schrieb Eli Zaretskii:
>> Cc: 70764@debbugs.gnu.org
>> Date: Sat, 04 May 2024 17:11:52 +0300
>> From: Eli Zaretskii <eliz@gnu.org>
>>
>>> Date: Sat, 4 May 2024 15:32:27 +0200
>>> From: Rolf Ade <rolf@pointsman.de>
>>>
>>> I don't know if my gdb support debug info downloading (see exact version info
>>> below) but if I answer the question with "n" - I don't want to enable this - I
>>> wonder why I'm (only mildly, of course) bothered with a not very helpful error
>>> message.
>>
>> I don't know.  AFAIR, it was not supposed to happen in Emacs 29.1 and
>> later.
> 
> FWIW, I just tried "M-x gdb" in Emacs 29.1, and didn't see any "No
> symbol" messages after answering "n" to the debuginfod question.

As reported in the original bug report I tried this also with current trunk (and 
-Q) and the same gdb (10.1) and saw the same message: "No symbol \"debuginfod\" 
in current context."

> My GDB on that system is version 12.1, though, and it was compiled
> with debuginfod support.  So maybe that's the reason.

The still officially supported debian 11 (Bullseye) comes with gdb 10.1 I use, see:

https://packages.debian.org/bullseye/gdb

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Eli Zaretskii]

> Date: Sat, 4 May 2024 17:42:20 +0200
> Cc: 70764@debbugs.gnu.org
> From: Rolf Ade <rolf@pointsman.de>
> 
> Am 04.05.24 um 16:11 schrieb Eli Zaretskii:
> >> Date: Sat, 4 May 2024 15:32:27 +0200
> >> From: Rolf Ade <rolf@pointsman.de>
> >>
> >>>> Even if I answer with "n" I get the message:
> >>>>
> >>>> "No symbol \"debuginfod\" in current context."
> >>>
> [...]
> >>> What OS are you using GDB on, and what is your GDB version, please?
> >>
> >> This is basically a GNU/linux debian 11.9 (so, not that old). More specifically,
> >> uname -a:
> >>
> >> Linux pointsman2 5.10.0-28-amd64 #1 SMP Debian 5.10.209-2 (2024-01-31) x86_64
> >> GNU/Linux
> >>
> >> gdb --version (it's the version installed by the debian package management system):
> >>
> >> GNU gdb (Debian 10.1-1.7) 10.1.90.20210103-git
> > 
> > Strange.  GDB 10.1 was the version where debuginfod support was added.
> > What does the shell command below show?
> > 
> >    $ gdb --config | fgrep debuginfod
> 
> This returns
> 
>               --with-debuginfod

Then I'm really stumped.

If you start GDB from the shell prompt, like this:

  $ gdb ./emacs

and then, when GDB shows its "(gdb)" prompt, type

  (gdb) set debuginfod enabled on

do you also see that "No symbol" error message?

And also please try this:

 $ gdb -i=mi ./emacs
 ...
 -gdb-set debuginfod enabled on

and see if you get any error message that way.

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Rolf Ade]

Am 04.05.24 um 18:12 schrieb Eli Zaretskii:
>> Date: Sat, 4 May 2024 17:42:20 +0200
>> Cc: 70764@debbugs.gnu.org
>> From: Rolf Ade <rolf@pointsman.de>
>>
>> Am 04.05.24 um 16:11 schrieb Eli Zaretskii:
>>>> Date: Sat, 4 May 2024 15:32:27 +0200
>>>> From: Rolf Ade <rolf@pointsman.de>
[...]
>>>
>>> Strange.  GDB 10.1 was the version where debuginfod support was added.
>>> What does the shell command below show?
>>>
>>>     $ gdb --config | fgrep debuginfod
>>
>> This returns
>>
>>                --with-debuginfod
> 
> Then I'm really stumped.
> 
> If you start GDB from the shell prompt, like this:
> 
>    $ gdb ./emacs
> 
> and then, when GDB shows its "(gdb)" prompt, type
> 
>    (gdb) set debuginfod enabled on
> 
> do you also see that "No symbol" error message?

Yes:

(gdb) set debuginfod enabled on
No symbol "debuginfod" in current context.
(gdb)


> And also please try this:
> 
>   $ gdb -i=mi ./emacs
>   ...
>   -gdb-set debuginfod enabled on
> 
> and see if you get any error message that way.

Yes:

(gdb)
-gdb-set debuginfod enabled on
^error,msg="No symbol \"debuginfod\" in current context."
(gdb)

So, this all seems to lead to that the gdb shipped with Bullseye isn't correctly 
build (despite the --with-debuginfod in the gdb --config output). I see this on 
two different boxes with debian 11.9; this is not specific to a single machine. 
But Bullseye still isn't that uncommon; so others should also see this, I suppose?

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Eli Zaretskii]

> Date: Sat, 4 May 2024 18:26:41 +0200
> Cc: 70764@debbugs.gnu.org
> From: Rolf Ade <rolf@pointsman.de>
> 
> > If you start GDB from the shell prompt, like this:
> > 
> >    $ gdb ./emacs
> > 
> > and then, when GDB shows its "(gdb)" prompt, type
> > 
> >    (gdb) set debuginfod enabled on
> > 
> > do you also see that "No symbol" error message?
> 
> Yes:
> 
> (gdb) set debuginfod enabled on
> No symbol "debuginfod" in current context.
> (gdb)
> 
> 
> > And also please try this:
> > 
> >   $ gdb -i=mi ./emacs
> >   ...
> >   -gdb-set debuginfod enabled on
> > 
> > and see if you get any error message that way.
> 
> Yes:
> 
> (gdb)
> -gdb-set debuginfod enabled on
> ^error,msg="No symbol \"debuginfod\" in current context."
> (gdb)
> 
> So, this all seems to lead to that the gdb shipped with Bullseye isn't correctly 
> build (despite the --with-debuginfod in the gdb --config output).

Most probably, yes.

> But Bullseye still isn't that uncommon; so others should also see this, I suppose?

Unfortunately, the only better alternative to stop these error
messages is for you to customize gdb-debuginfod-enable-setting to the
nil value.  That is the only way we know of that allows to prevent the
error messages by avoiding to even try the problematic setting.  No
other way we tried at the time to dynamically detect the support
worked in a satisfactory manner, basically because trying an
unsupported command is bound to trigger some error message.

bug#70764: 29.1; M-x gdb: don't ask me about debuginfod server if debuginfod is not in context

[Eli Zaretskii]

> Cc: 70764@debbugs.gnu.org
> Date: Sat, 04 May 2024 20:07:31 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> 
> > (gdb)
> > -gdb-set debuginfod enabled on
> > ^error,msg="No symbol \"debuginfod\" in current context."
> > (gdb)
> > 
> > So, this all seems to lead to that the gdb shipped with Bullseye isn't correctly 
> > build (despite the --with-debuginfod in the gdb --config output).
> 
> Most probably, yes.
> 
> > But Bullseye still isn't that uncommon; so others should also see this, I suppose?
> 
> Unfortunately, the only better alternative to stop these error
> messages is for you to customize gdb-debuginfod-enable-setting to the
> nil value.  That is the only way we know of that allows to prevent the
> error messages by avoiding to even try the problematic setting.  No
> other way we tried at the time to dynamically detect the support
> worked in a satisfactory manner, basically because trying an
> unsupported command is bound to trigger some error message.

No further comments within 2 weeks, so I'm now closing this bug.