Emacs 21.4a segmentation fault on startup (gcc 4.0, Solaris 8)

Emacs 21.4a segmentation fault on startup (gcc 4.0, Solaris 8)

[Andrew Koenig]

I just installed gcc 4.0 on my Solaris 8 machine and rebuilt GNU emacs 21.4; 
the result is a segmentation fault on startup.

The first few lines of the backtrace (abbreviated):

XtInitializeWidgetClass from /usr/lib/libXt.so.4
_XtCreateWidget from /usr/lib/libXt.so.4
XtCreateWidget from /usr/lib/libXt.so.4
Fx_create_frame at xfns.c:3644
Ffuncall at eval.c:2659
Fbyte_code at bytecode.c:716
funcall_lambda at eval.c:2851
....

If I compiler under gcc 2.95, or run it as emacs -nw, it appears to work.

Any suggestions?

Re: Emacs 21.4a segmentation fault on startup (gcc 4.0, Solaris 8)

[Richard Stallman]

This could be a GCC bug, or it could be something in Emacs that is not
well-defined C.  You could look at the compiled code for the call to
XtCreateWidget and see if it looks correct.

You could also try recompiling just xterm.c with the old GCC version
and see if that makes everything work.  If so, you could then try
splitting that file in various ways and compiling half with each
GCC version.  This way maybe you could determine which function
the problem is in.

Re: Emacs 21.4a segmentation fault on startup (gcc 4.0, Solaris 8)

[Kevin Rodgers]

Richard Stallman wrote:
 > This could be a GCC bug, or it could be something in Emacs that is not
 > well-defined C.

The latter possibility could be eliminated from consideration
permanently if gcc -Wall were the default, and the code was tuned up to
eliminate (or at least minimize) gcc's warnings.

-- 
Kevin Rodgers

Re: Emacs 21.4a segmentation fault on startup (gcc 4.0, Solaris 8)

[Eli Zaretskii]

> From: Kevin Rodgers <ihs_4664@yahoo.com>
> Date: Thu, 19 May 2005 08:36:10 -0600
> 
> Richard Stallman wrote:
>  > This could be a GCC bug, or it could be something in Emacs that is not
>  > well-defined C.
> 
> The latter possibility could be eliminated from consideration
> permanently if gcc -Wall were the default, and the code was tuned up to
> eliminate (or at least minimize) gcc's warnings.

We would probably do that if only "gcc -Wall" in its latest versions
did not whine about too many perfectly valid usages of C.

Re: Emacs 21.4a segmentation fault on startup (gcc 4.0, Solaris 8)

[Richard Stallman]

    The latter possibility could be eliminated from consideration
    permanently if gcc -Wall were the default, and the code was tuned up to
    eliminate (or at least minimize) gcc's warnings.

-Wall warns about a lot of things that are not wrong, and causes
more trouble than it is worth.  I agreed to implement -Wall on the
explicit understanding that we would NOT adopt the goal to change our
code so that -Wall would not warn.

If there is a specific kind of warning that would be useful here,
there is probably an -f option to enable that specific kind of
warning.  That might be a good thing to use for Emacs.

Re: Emacs 21.4a segmentation fault on startup (gcc 4.0, Solaris 8)

[Eli Zaretskii]

> From: Richard Stallman <rms@gnu.org>
> Date: Fri, 20 May 2005 17:56:24 -0400
> Cc: bug-gnu-emacs@gnu.org
> 
> -Wall warns about a lot of things that are not wrong, and causes
> more trouble than it is worth.  I agreed to implement -Wall on the
> explicit understanding that we would NOT adopt the goal to change our
> code so that -Wall would not warn.

Except that latest versions of GCC by default enable more and more
warnings that previously belonged to -Wall.  Warnings such as these
for example:

    regex.c: In function `re_search_2':
    regex.c:4231: warning: comparison is always true due to limited range of data type

This warning is a terrible nuisance in a complex macro, for example.

> If there is a specific kind of warning that would be useful here,
> there is probably an -f option to enable that specific kind of
> warning.  That might be a good thing to use for Emacs.

Agreed.