Need to get Lisp backtrace when error is in C code

Need to get Lisp backtrace when error is in C code

[Robert Weiner]

[-- Attachment #1: Type: text/plain, Size: 1072 bytes --]

Every so often when using Emacs, I get an error from C code similar to this:
   let: Wrong type argument: markerp, nil

which gives no indication of where the error is coming from.  I have
debug-on-error set to t.  I might not have Emacs built with debugging
symbols at times.

Were I to know the source of the error, I could set debug-on-entry to some
Lisp-exposed C function and get a Lisp stack trace to the point of C
function entry.  But why can't I have something like this automatically?

The C code is raising an error.  The Lisp stack exists.  Why can't the C
error routine force the backtrace with some notion that the error exists
further in C code?  This would at least narrow down the cause of the error
greatly by showing a lot of the path in code taken.

Personally, for errors like this, I would typically include the function
name reference so that the location is obvious and users without debug
symbols in their Emacs could still report a more useful error.

Is this doable?  A good idea?

How do others debug such situations (no C debug symbols)?

Bob

[-- Attachment #2: Type: text/html, Size: 2691 bytes --]

Re: Need to get Lisp backtrace when error is in C code

[Eli Zaretskii]

> From: Robert Weiner <rsw@gnu.org>
> Date: Tue, 3 Oct 2017 12:43:20 -0400
> 
> How do others debug such situations (no C debug symbols)?

Personally, I never strip Emacs.

Re: Need to get Lisp backtrace when error is in C code

[Stefan Monnier]

> Personally, for errors like this, I would typically include the function
> name reference so that the location is obvious and users without debug
> symbols in their Emacs could still report a more useful error.
> Is this doable?  A good idea?

I think it's desirable, yes.  The problem is how to do it.
I guess we could try to replace `Ferror` with a C macro that pushes some
caller-info onto the Lisp backtrace before signaling the error (a bit
like we currently push some pseudo-function onto that stack trace
during garbage collection and redisplay).  It won't handle the cases
where we call `Fsignal` instead, nor the cases where we call some
generic C function which then calls Ferror, so to make it really work
well, it'll require more manual work to place the caller-info at the
"right" spot.


        Stefan

Re: Need to get Lisp backtrace when error is in C code

[Robert Weiner]

[-- Attachment #1: Type: text/plain, Size: 1219 bytes --]

On Tue, Oct 3, 2017 at 1:50 PM, Stefan Monnier <monnier@iro.umontreal.ca>
wrote:

> > Personally, for errors like this, I would typically include the function
> > name reference so that the location is obvious and users without debug
> > symbols in their Emacs could still report a more useful error.
> > Is this doable?  A good idea?
>
> I think it's desirable, yes.  The problem is how to do it.
>

Even just showing the Lisp backtrace by itself with some note that
the error was raised in C code would narrow down the location tremendously
I think.

​​
> I guess we could try to replace `Ferror` with a C macro that pushes some
> ​​
> caller-info onto the Lisp backtrace before signaling the error (a bit
> ​​
> like we currently push some pseudo-function onto that stack trace
> ​​
> during garbage collection and redisplay).  It won't handle the cases
> ​​
> where we call `Fsignal` instead, nor the cases where we call some
> ​​
> generic C function which then calls Ferror, so to make it really work
> ​​
> well, it'll require more manual work to place the caller-info at the
> ​​
> "right" spot.
>

​A full solution would be a big win.

Bob

[-- Attachment #2: Type: text/html, Size: 3196 bytes --]

Re: Need to get Lisp backtrace when error is in C code

[Stefan Monnier]

> Even just showing the Lisp backtrace by itself with some note that the
> error was raised in C code would narrow down the location tremendously
> I think.

The bottom of the backtrace is always "C code".

> ​A full solution would be a big win.

Could be.  But someone will have to try it out before we know for sure.


        Stefan

Re: Need to get Lisp backtrace when error is in C code

[Robert Weiner]

[-- Attachment #1: Type: text/plain, Size: 493 bytes --]

On Tue, Oct 3, 2017 at 3:30 PM, Stefan Monnier <monnier@iro.umontreal.ca>
wrote:

> > Even just showing the Lisp backtrace by itself with some note that the
> > error was raised in C code would narrow down the location tremendously
> > I think.
>
> The bottom of the backtrace is always "C code".
>

> > ​A full solution would be a big win.
>
> Could be.  But someone will have to try it out before we know for sure.
>

​Are you voluntee​ring to make this happen?

Bob

[-- Attachment #2: Type: text/html, Size: 1544 bytes --]

Re: Need to get Lisp backtrace when error is in C code

[Stefan Monnier]

>> Could be.  But someone will have to try it out before we know for sure.
> ​Are you voluntee​ring to make this happen?

No,


        Stefan