gdb/gdbtk/generic/gdbtk-stack.c - fail due to strict aliasing

gdb/gdbtk/generic/gdbtk-stack.c - fail due to strict aliasing

[Alon Bar-Lev]

Hello,

Reported as binutils bug#6986 [1]

arguments = *(int*) &clientData;

Gets strict aliasing warning that escalate to error and fails build.

Attached a simple temp fix against head.

But the apparently the original has some issue too.

Regards,
Alon Bar-Lev.

------- Additional Comment #2 From Andreas Schwab 2008-10-25 22:53   ------- 
Neither gdb nor gdbtk are part of binutils.  And the patch is as broken as the 
original code, it still fails on big-endian 64-bit architectures.  Please 
report that to the insight mailinglist.

[1] http://sourceware.org/bugzilla/show_bug.cgi?id=6986

---

Index: gdb/gdbtk/generic/gdbtk-stack.c
===================================================================
RCS file: /cvs/src/src/gdb/gdbtk/generic/gdbtk-stack.c,v
retrieving revision 1.29
diff -u -B -r1.29 gdbtk-stack.c
--- gdb/gdbtk/generic/gdbtk-stack.c	2 Sep 2008 17:29:23 -0000	1.29
+++ gdb/gdbtk/generic/gdbtk-stack.c	25 Oct 2008 22:28:12 -0000
@@ -285,7 +285,10 @@
       return TCL_ERROR;
     }
 
-  arguments = *(int*) &clientData;
+{
+  void *p = &clientData;
+  arguments = *(int*) p;
+}
 
   /* Initialize the result pointer to an empty list. */
 

Re: gdb/gdbtk/generic/gdbtk-stack.c - fail due to strict aliasing

[Martin Rakhmanov]

This allows to compile currently available snapshot 
(ftp://sourceware.org/pub/insight/snapshots/current/insight-weekly-6.8.50-20081021-tar.bz2) 
but again does not fix problem with running insight on Fedora 9: the 
snapshot ships with old tcl/tk versions and uses them.

Thanks
Martin

Alon Bar-Lev wrote:
> Hello,
> 
> Reported as binutils bug#6986 [1]
> 
> arguments = *(int*) &clientData;
> 
> Gets strict aliasing warning that escalate to error and fails build.
> 
> Attached a simple temp fix against head.
> 
> But the apparently the original has some issue too.
> 
> Regards,
> Alon Bar-Lev.
> 
> ------- Additional Comment #2 From Andreas Schwab 2008-10-25 22:53   ------- 
> Neither gdb nor gdbtk are part of binutils.  And the patch is as broken as the 
> original code, it still fails on big-endian 64-bit architectures.  Please 
> report that to the insight mailinglist.
> 
> [1] http://sourceware.org/bugzilla/show_bug.cgi?id=6986
> 
> ---
> 
> Index: gdb/gdbtk/generic/gdbtk-stack.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/gdbtk/generic/gdbtk-stack.c,v
> retrieving revision 1.29
> diff -u -B -r1.29 gdbtk-stack.c
> --- gdb/gdbtk/generic/gdbtk-stack.c	2 Sep 2008 17:29:23 -0000	1.29
> +++ gdb/gdbtk/generic/gdbtk-stack.c	25 Oct 2008 22:28:12 -0000
> @@ -285,7 +285,10 @@
>        return TCL_ERROR;
>      }
>  
> -  arguments = *(int*) &clientData;
> +{
> +  void *p = &clientData;
> +  arguments = *(int*) p;
> +}
>  
>    /* Initialize the result pointer to an empty list. */
>  
> 

Re: gdb/gdbtk/generic/gdbtk-stack.c - fail due to strict aliasing

[Keith Seitz]

Alon Bar-Lev wrote:

> Reported as binutils bug#6986 [1]
> 
> arguments = *(int*) &clientData;

Thank you for the patch: it's not quite correct, but then the existing 
code is completely wrong. That got really botched somehow: clientData is 
an int, not an int*. I don't know how I messed that up. O:-)

I've committed a proper fix. Or at least I think I have, since gcc 4.4.0 
no longer complains about this. Let me know if this still presents 
problems for you.

I'm remaking the snapshot to include this patch.

Keith

Re: gdb/gdbtk/generic/gdbtk-stack.c - fail due to strict aliasing

[Gene Smith]

Keith Seitz wrote:
> Alon Bar-Lev wrote:
> 
>> Reported as binutils bug#6986 [1]
>>
>> arguments = *(int*) &clientData;
> 
> Thank you for the patch: it's not quite correct, but then the existing 
> code is completely wrong. That got really botched somehow: clientData is 
> an int, not an int*. I don't know how I messed that up. O:-)
> 
> I've committed a proper fix. Or at least I think I have, since gcc 4.4.0 
> no longer complains about this. Let me know if this still presents 
> problems for you.
> 
> I'm remaking the snapshot to include this patch.
> 
> Keith
> 

1027 SS won't build as pointed out other msg. When I manually apply this 
patch to 1021 SS it builds OK.