Emacs hung, have attached debugger

Emacs hung, have attached debugger

[Geoff Shannon]

[-- Attachment #1: Type: text/plain, Size: 662 bytes --]

So I've got a hung emacs that I have attached to in a debugger.  I opened a
new emacs to use report-emacs-bug; following the instructions I was going
to include the output of 'bt full'.  Except that it's about 17500 lines
long...

I took a look at the backtrace as well, and of the 3800 calls on the list,
~3700 of them are from either mark_object or mark_vectorlike.

I'm not sure what else to do (I'm not familiar with emacs c code at all),
but I can keep the hung emacs running in the debugger pretty much
indefinitely if that's useful.

I submitted a report to debbugs with the interesting portions of the
backtrace output.

-- 
Geoff

Nothing is ever easy.

[-- Attachment #2: Type: text/html, Size: 885 bytes --]

Re: Emacs hung, have attached debugger

[Glenn Morris]

Geoff Shannon wrote:

> So I've got a hung emacs that I have attached to in a debugger.  I opened a
> new emacs to use report-emacs-bug; following the instructions I was going
> to include the output of 'bt full'.  Except that it's about 17500 lines
> long...

So send it as a compressed attachment.
I wish more people would compress the attachments they send to bug reports.

(I would have been happy to make these comments on your bug report.
No need for a separate discussion on emacs-devel.)

Re: Emacs hung, have attached debugger

[David Kastrup]

Geoff Shannon <geoffpshannon@gmail.com> writes:

> So I've got a hung emacs that I have attached to in a debugger.  I opened a
> new emacs to use report-emacs-bug; following the instructions I was going
> to include the output of 'bt full'.  Except that it's about 17500 lines
> long...
>
> I took a look at the backtrace as well, and of the 3800 calls on the
> list, ~3700 of them are from either mark_object or mark_vectorlike.

There will be likely a lot of repetition here as this sounds like an
infinite recursion.  Just the end and the start of the traceback (enough
to recognize the pattern in the middle) should be enough.  The traceback
will likely be because of stack overflow.

You can use ulimit (see its man page) to limit the available stack space
in advance: that should allow you to get a backtrace that is reasonably
short even without manual intervention.  Just be sure that it's not
bombing out because of a regular operation.

> I submitted a report to debbugs with the interesting portions of the
> backtrace output.

Sounds good.

-- 
David Kastrup

Re: Emacs hung, have attached debugger

[Geoff Shannon]

[-- Attachment #1: Type: text/plain, Size: 1111 bytes --]

I posted compressed versions of the full backtrace to the bug on debbugs.


On Fri, Jun 6, 2014 at 12:33 AM, David Kastrup <dak@gnu.org> wrote:

> There will be likely a lot of repetition here as this sounds like an
> infinite recursion.  Just the end and the start of the traceback (enough
> to recognize the pattern in the middle) should be enough.  The traceback
> will likely be because of stack overflow.
>

Basically the conclusion I came to after examining the backtrace. Before
Glenn
suggested compressing the full content and attaching it, that's what I had
done.

Though of course I didn't think of including the top entries until after
I'd posted the
bottom ones...


> You can use ulimit (see its man page) to limit the available stack space
> in advance: that should allow you to get a backtrace that is reasonably
> short even without manual intervention.  Just be sure that it's not
> bombing out because of a regular operation.
>

Good idea!  Now if only I could reproduce this crash...

Also, sorry if posting about this on emacs-devel was a bit out of place.

-- 
Geoff

Nothing is ever easy.

[-- Attachment #2: Type: text/html, Size: 1846 bytes --]

Re: Emacs hung, have attached debugger

[Eli Zaretskii]

> From: Geoff Shannon <geoffpshannon@gmail.com>
> Date: Fri, 6 Jun 2014 00:14:39 -0700
> 
> So I've got a hung emacs that I have attached to in a debugger.  I opened a
> new emacs to use report-emacs-bug; following the instructions I was going
> to include the output of 'bt full'.  Except that it's about 17500 lines
> long...

I replied to you in the bug tracker.  Please keep further discussion
there, so that it gets recorded with the bug.  Thanks.