Re: [RFA] Fix several problems with the gdb.arch/gdb1291.exp test

[Daniel Jacobowitz <drow@mvista.com>]

On Tue, Feb 10, 2004 at 06:07:37PM -0700, Fred Fish wrote:
> There are several problems with the current gdb1291 test:
> 
> (1) Compilation with -O2 is required to expose the originally reported
> bug.
> 
> (2) A gcc bug (http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14022)
> interferes with properly testing for the reported bug.  The original
> test was apparently modified to use an ASM to avoid having part of it
> optimized away when using -O2.
> 
> (3) The test case has DOS style line separators for no apparent good
> reason.
> 
> (4) The test case doesn't verify that reducing the local variable size
> to less than or equal to 256 bytes makes the problem go away.  Not
> strictly necessary, but useful to have confirmed to ensure that we are
> getting accurate test results.
> 
> This patch fixes the above issues and also causes the failure to be
> logged as:
> 
>   KFAIL: gdb.arch/gdb1291.exp: backtrace with local variable larger than 256 bytes (PRMS: gdb/1291)
> 
> instead of just a regular FAIL.
> 
> -Fred

Sorry, not OK.  There are at least three problems:
 - You have changed what is being tested.  You are now testing the
   dwarf2 unwinder, not the prologue unwinder.
 - You are compiling an empty function with -O2 and expecting the
   result to have some meaning.  GCC will someday, hopefully soon,
   remove the unused stack allocation.  Then we won't be testing
   anything at all.
 - You've added a kfail pattern for PR 1291 that doesn't match the
   description in the PR; the less than 256 bytes case shouldn't
   be a kfail.

I recommend writing the whole test in assembly instead; this is
gdb.arch, after all.  We don't want to be sensitive to GCC bugs.


-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer