dump (temacs) fails, core in alloc.c

dump (temacs) fails, core in alloc.c

[Jan Djärv]

Hello.

Can anybody in the know fix this? temacs -batch -l loadup dump
gets a segmentation violation at alloc.c:

Program received signal SIGSEGV, Segmentation fault.
0x00000000005c2b23 in compact_small_strings () at alloc.c:2264
2264	  for (b = tb->next; b; b = next)
(gdb) p b
$1 = (struct sblock *) 0x0
(gdb) p tb
$2 = (struct sblock *) 0x0
(gdb) p oldest_sblock
$3 = (struct sblock *) 0x0

	Jan D.

Re: dump (temacs) fails, core in alloc.c

[Jan Djärv]

Never mind, a bootstrap was required.

	Jan D.

Jan Djärv skrev:
> Hello.
> 
> Can anybody in the know fix this? temacs -batch -l loadup dump
> gets a segmentation violation at alloc.c:
> 
> Program received signal SIGSEGV, Segmentation fault.
> 0x00000000005c2b23 in compact_small_strings () at alloc.c:2264
> 2264      for (b = tb->next; b; b = next)
> (gdb) p b
> $1 = (struct sblock *) 0x0
> (gdb) p tb
> $2 = (struct sblock *) 0x0
> (gdb) p oldest_sblock
> $3 = (struct sblock *) 0x0
> 
>     Jan D.
> 
> 

Re: dump (temacs) fails, core in alloc.c

[Dan Nicolaescu]

Jan Djärv <jan.h.d@swipnet.se> writes:

  > Never mind, a bootstrap was required.

lisp.h is not consistenly listed as a dependency in Makefile (not sure why).
So it's probably Stefan's lisp.h change that probably tripped you...

  > 	Jan D.
  > 
  > Jan Djärv skrev:
  > > Hello.
  > >
  > > Can anybody in the know fix this? temacs -batch -l loadup dump
  > > gets a segmentation violation at alloc.c:
  > >
  > > Program received signal SIGSEGV, Segmentation fault.
  > > 0x00000000005c2b23 in compact_small_strings () at alloc.c:2264
  > > 2264      for (b = tb->next; b; b = next)
  > > (gdb) p b
  > > $1 = (struct sblock *) 0x0
  > > (gdb) p tb
  > > $2 = (struct sblock *) 0x0
  > > (gdb) p oldest_sblock
  > > $3 = (struct sblock *) 0x0
  > >
  > >     Jan D.
  > >
  > >

Re: dump (temacs) fails, core in alloc.c

[Eli Zaretskii]

> Date: Mon, 9 Nov 2009 11:12:05 -0800 (PST)
> From: Dan Nicolaescu <dann@ics.uci.edu>
> Cc: Emacs-Devel <emacs-devel@gnu.org>
> 
> lisp.h is not consistenly listed as a dependency in Makefile (not sure why).

Richard used to be opposed to that, to avoid recompiling everything
whenever some small part or even comment in lisp.h changes.  Not sure
if this is still a valid concern, what with today's compilation speed.

> So it's probably Stefan's lisp.h change that probably tripped you...

I always manually delete all *.o files when there's a significant
change in lisp.h.

Re: dump (temacs) fails, core in alloc.c

[Dan Nicolaescu]

Eli Zaretskii <eliz@gnu.org> writes:

  > > Date: Mon, 9 Nov 2009 11:12:05 -0800 (PST)
  > > From: Dan Nicolaescu <dann@ics.uci.edu>
  > > Cc: Emacs-Devel <emacs-devel@gnu.org>
  > > 
  > > lisp.h is not consistenly listed as a dependency in Makefile (not sure why).
  > 
  > Richard used to be opposed to that, to avoid recompiling everything
  > whenever some small part or even comment in lisp.h changes.  Not sure
  > if this is still a valid concern, what with today's compilation speed.
  > 
  > > So it's probably Stefan's lisp.h change that probably tripped you...
  > 
  > I always manually delete all *.o files when there's a significant
  > change in lisp.h.

Tools are much more reliable than humans, compilation speed for emacs is
not a big concern anymore, so IMHO lisp.h should be added as a
dependency.

Re: dump (temacs) fails, core in alloc.c

[Stefan Monnier]

> Tools are much more reliable than humans, compilation speed for emacs is
> not a big concern anymore, so IMHO lisp.h should be added as a
> dependency.

Fine by me,


        Stefan "who uses a `make depend' scheme and hence gets those
                extra dependencies anyway"

Re: dump (temacs) fails, core in alloc.c

[Dan Nicolaescu]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:

  > > Tools are much more reliable than humans, compilation speed for emacs is
  > > not a big concern anymore, so IMHO lisp.h should be added as a
  > > dependency.
  > 
  > Fine by me,

If nobody beats me to it (but hopefully somebody will), I'll do it after
I clear my current patch backlog (63 files changed...)

  >         Stefan "who uses a `make depend' scheme and hence gets those
  >                 extra dependencies anyway"

That's TRTD, but ...
Will Stefan's "magic tree that implements all kinds of interesting
stuff" appear in BZR once we make the switch?

Re: dump (temacs) fails, core in alloc.c

[Stefan Monnier]

>> Stefan "who uses a `make depend' scheme and hence gets those
>> extra dependencies anyway"
> That's TRTD, but ...
> Will Stefan's "magic tree that implements all kinds of interesting
> stuff" appear in BZR once we make the switch?

That tree is not secret.  If people want it, I can make it accessible
publically, but really: there's a reason why those patches aren't in the
trunk yet.  Sometimes that reason is just that I haven't gotten to it
yet, but for more important changes (e.g., "make depend", or
multiple-inheritance in keymaps) the reason is that the implementation
I have is not good enough for the trunk.


        Stefan

Re: dump (temacs) fails, core in alloc.c

[Jan D.]

On 2009-11-09 23:02, Stefan Monnier wrote:
>> Tools are much more reliable than humans, compilation speed for emacs is
>> not a big concern anymore, so IMHO lisp.h should be added as a
>> dependency.
>
> Fine by me,
>
>
>          Stefan "who uses a `make depend' scheme and hence gets those
>                  extra dependencies anyway"

How about detecting if we are using gcc and generate dependencies on the 
fly while compiling the first time?  I see many packages that does so. 
Would such a change be OK?

	Jan D.

Re: dump (temacs) fails, core in alloc.c

[Stefan Monnier]

> How about detecting if we are using gcc and generate dependencies on the fly
> while compiling the first time?  I see many packages that does so. Would
> such a change be OK?

That would be good, yes.  As long as it doesn't break builds on non-gcc
and non-gmake systems.


        Stefan

Gcc-generated dependencise. (Was: Re: dump (temacs) fails, core in alloc.c)

[Jan Djärv]

Stefan Monnier skrev:
>> How about detecting if we are using gcc and generate dependencies on the fly
>> while compiling the first time?  I see many packages that does so. Would
>> such a change be OK?
> 
> That would be good, yes.  As long as it doesn't break builds on non-gcc
> and non-gmake systems.
> 

I have checked in dependency generation on the fly for src.  Configure checks 
for gcc -MMD -MF and GNU Make and if not explicitly disabled with 
--disable-autodepend, dependencies are then generated when files are compiled.

	Jan D.

Re: dump (temacs) fails, core in alloc.c

[Richard Stallman]

In the past, recompiling everything due to a change in lisp.h was
extremely painful.  Removing that dependency was really important.

What to do today is up to the current maintainers.