|
From: Duane G. <d.g...@ps...> - 2005-03-12 13:54:52
|
Jeremy Fitzhardinge wrote: >Tom Hughes wrote: >> >>I put a patch u plast night, although I'm worried it might break >>with stabs from other compilers. I agree this probably shouldn't >>hold up the release. >> =20 >> >Looks fine to me; I just checked it in. We'll see if it breaks someone = >else. I'm not sure where the 3rd parameter to R came from. The patch has made a difference, but I'm afraid it still crashes. I've = added the full output to the bug entry. This time the error is: @@ unlikely looking definition in unparsed remains ";" If it would help I could make the library it is crashing on available to = download. It is fairly large though (>100MB), and has quite a few = dependencies. Cheers,=20 Duane. |
|
From: Duane G. <d.g...@ps...> - 2005-03-12 19:03:38
|
Tom Hughes wrote:
> The library isn't much use if it has dependencies. The output from
> running "objdump -G" on it might be, but only if we can work which
> stab is causing the crash.
OK, I've dumped the stabs info and recompiled the rc3 valgrind code with =
stabs_debug enabled to figure out where the problem is. It looks like =
the offending entry is this one:
12 GSYM 0 0 00000000 373 logical*2:t22=3D2;
I've had a bit of a look around in the code and managed to get it past =
the problem with the patch included below. Of course I have no idea if =
what I'm doing there is valid, or whether it might break anything else. =
Also, after it gets past that one, it goes on to crash on a completely =
different symbol a bit later. It seems pgf77 3.2 is generating some =
fairly idiosyncractic debugging information.
Just FYI, the next symbol it fails on looks like this:
16 PSYM 0 0 00000008 445 filename:var3;1;A12;13
No time to look into it further at the moment, but I might try and come =
back to it later in the week.
--- coregrind/vg_stabs.c.orig Sat Mar 12 19:02:30 2005
+++ coregrind/vg_stabs.c Sat Mar 12 19:02:08 2005
@@ -668,6 +668,12 @@
case 't': { /* typedef: 't' TYPE */
SymType *td =3D stabtype_parser(si, NULL, &p);
type =3D VG_(st_mktypedef)(def, NULL, td);
+
+ /* Work around problem with output from pgf77 3.2.
+ I've no idea what the spec says about this, or whether it
+ will break anything else. */
+ if (*p =3D=3D ';')
+ p++;
break;
}
|
|
From: Jeremy F. <je...@go...> - 2005-03-12 21:18:10
|
Duane Griffin wrote:
>Just FYI, the next symbol it fails on looks like this:
>
>16 PSYM 0 0 00000008 445 filename:var3;1;A12;13
>
>
I'm not sure I have any idea what that means. A parameter passed by
reference in a register?
>No time to look into it further at the moment, but I might try and come back to it later in the week.
>
>
I think the most useful fix will be to see if there's some way to get
the stabs parser to return cleanly without crashing. I don't think its
a good use of time to chase strange stabs generated by PGI Fortran,
unless you're really excited by it...
J
|
|
From: Tom H. <to...@co...> - 2005-03-12 14:04:12
|
In message <E0D...@pe...>
"Duane Griffin" <d.g...@ps...> wrote:
> Jeremy Fitzhardinge wrote:
> >Tom Hughes wrote:
> >>
> >>I put a patch u plast night, although I'm worried it might break
> >>with stabs from other compilers. I agree this probably shouldn't
> >>hold up the release.
> >>
> >>
> >Looks fine to me; I just checked it in. We'll see if it breaks someone
> >else. I'm not sure where the 3rd parameter to R came from.
>
> The patch has made a difference, but I'm afraid it still crashes. I've
> added the full output to the bug entry. This time the error is:
>
> @@ unlikely looking definition in unparsed remains ";"
Unfortunately this time round there doesn't seem to be enough
information in the crash to diagnose the problem.
> If it would help I could make the library it is crashing on available
> download. It is fairly large though (>100MB), and has quite a few
> to dependencies.
The library isn't much use if it has dependencies. The output from
running "objdump -G" on it might be, but only if we can work which
stab is causing the crash.
Tom
--
Tom Hughes (to...@co...)
http://www.compton.nu/
|
|
From: Jeremy F. <je...@go...> - 2005-03-12 21:09:04
|
Tom Hughes wrote:
>The library isn't much use if it has dependencies. The output from
>running "objdump -G" on it might be, but only if we can work which
>stab is causing the crash.
>
I think if you just mmap the whole thing in, Valgrind will think it is
an object file and parse all the pieces, even though it isn't being used
as a library.
J
|