Fix for unexpected failure in ld-cris/warn3

Fix for unexpected failure in ld-cris/warn3

[Nick Clifton]

Hi Guys,

  I am applying the patch below to remove an unexpected warning from
  the linker testsuite for the cris-elf target.  It appears that the
  ld-cris/warn[34].d files were intended to be different versions of
  the same test, one for ELF based ports and one for AOUT.
  Unfortunately the code discriminate between them was not working and
  this patch fixes that.

Cheers
  Nick

ld/testsuite/ChangeLog
2005-03-01  Nick Clifton  <nickc@redhat.com>

	* ld-cris/warn3.d: Only run this version of the test for cris-aout
	targets.
	* ld-cris/warn4.d: Only run this version of the test for cris-elf
	targets.

Index: ld/testsuite/ld-cris/warn3.d
===================================================================
RCS file: /cvs/src/src/ld/testsuite/ld-cris/warn3.d,v
retrieving revision 1.1
diff -c -3 -p -r1.1 warn3.d
*** ld/testsuite/ld-cris/warn3.d	7 Feb 2005 03:16:43 -0000	1.1
--- ld/testsuite/ld-cris/warn3.d	1 Mar 2005 16:10:50 -0000
***************
*** 1,7 ****
  #source: start1.s
  #source: globsym1ref.s
  #source: globsymw2.s
! #target: cris-*-*elf* cris-*-*aout*
  #as: --em=crisaout
  #ld: -mcrisaout
  #warning: warning: isatty is not implemented and will always fail$
--- 1,7 ----
  #source: start1.s
  #source: globsym1ref.s
  #source: globsymw2.s
! #target: cris-*-*aout*
  #as: --em=crisaout
  #ld: -mcrisaout
  #warning: warning: isatty is not implemented and will always fail$
Index: ld/testsuite/ld-cris/warn4.d
===================================================================
RCS file: /cvs/src/src/ld/testsuite/ld-cris/warn4.d,v
retrieving revision 1.1
diff -c -3 -p -r1.1 warn4.d
*** ld/testsuite/ld-cris/warn4.d	7 Feb 2005 03:16:43 -0000	1.1
--- ld/testsuite/ld-cris/warn4.d	1 Mar 2005 16:10:50 -0000
***************
*** 1,7 ****
  #source: start1.s
  #source: globsym1ref.s
  #source: globsymw2.s
! #target: cris-*-*elf* cris-*-*aout*
  #as: --em=criself
  #ld: -mcriself
  #warning: warning: isatty is not implemented and will always fail$
--- 1,7 ----
  #source: start1.s
  #source: globsym1ref.s
  #source: globsymw2.s
! #target: cris-*-*elf*
  #as: --em=criself
  #ld: -mcriself
  #warning: warning: isatty is not implemented and will always fail$

Re: Fix for unexpected failure in ld-cris/warn3

[Hans-Peter Nilsson]

On Tue, 1 Mar 2005, Nick Clifton wrote:

> Hi Guys,
>
>   I am applying the patch below to remove an unexpected warning from
>   the linker testsuite for the cris-elf target.

I don't see warnings in my regression tests.  Please specify.

>  It appears that the
>   ld-cris/warn[34].d files were intended to be different versions of
>   the same test, one for ELF based ports and one for AOUT.
>   Unfortunately the code discriminate between them was not working and
>   this patch fixes that.

No, it seems to just paper over whatever problem there might be.
The as and ld options should take care of the differences; both
a.out and ELF formats are supported by both toolchains; defaults
are different.

Please revert and elaborate.

brgds, H-P

Re: Fix for unexpected failure in ld-cris/warn3

[Nick Clifton]

Hi Hans-Peter,

> I don't see warnings in my regression tests.  Please specify.

Certainly.  With a toolchain configured as --target=cris-elf and built 
on an x86_64 host running Linux I saw these messages in the ld.log when 
running the linker tests:

   succeeded with: <>, expected: <warning: isatty is not implemented and 
will always fail$>
   FAIL: ld-cris/warn3

> No, it seems to just paper over whatever problem there might be.
> The as and ld options should take care of the differences; both
> a.out and ELF formats are supported by both toolchains; defaults
> are different.
> 
> Please revert and elaborate.

OK - I have reverted my patch.

 From looking at the warn3.d and warn4.d files I had thought that one 
was intended to be run for the cris-elf target and another for the 
cris-aout target.  Obviously I was wrong.

Incidentally I do not normally test the cris-aout target but whilst 
investigating this problem I thought that it would be a good idea to 
check.  It seems to generate quite a lot of unexpected failures in the 
linker's ld-cris testsuite, most of which are due to this error message:

   Error: --no-underscore is invalid with a.out format

Cheers
   Nick

Re: Fix for unexpected failure in ld-cris/warn3

[Hans-Peter Nilsson]

Hi Nick

On Tue, 1 Mar 2005, Nick Clifton wrote:
> With a toolchain configured as --target=cris-elf and built
> on an x86_64 host running Linux I saw these messages in the ld.log when
> running the linker tests:
>
>    succeeded with: <>, expected: <warning: isatty is not implemented and
> will always fail$>
>    FAIL: ld-cris/warn3

Thanks.  I have not seen this in autotester runs on
i686-pc-linux-gnu cross to cris-axis-elf.  Maybe it's a
64-bit-host thingy; the test for the warning failing there.
I'll try and get hold of one and test there.

>  From looking at the warn3.d and warn4.d files I had thought that one
> was intended to be run for the cris-elf target and another for the
> cris-aout target.  Obviously I was wrong.

They're indeed meant for a.out and ELF, but as the assembler and
linker support both formats, both tests run on both cris-*-elf
and cris-*-aout.  Maybe a bit confusing if you're not used to
GAS being multiformat. :-)

> Incidentally I do not normally test the cris-aout target but whilst
> investigating this problem I thought that it would be a good idea to
> check.  It seems to generate quite a lot of unexpected failures in the
> linker's ld-cris testsuite, most of which are due to this error message:
>
>    Error: --no-underscore is invalid with a.out format

Thanks.  I do not normally test cris-aout but will do so and fix
stragglers.

brgds, H-P

Re: Fix for unexpected failure in ld-cris/warn3

[Hans-Peter Nilsson]

On Tue, 1 Mar 2005, Nick Clifton wrote:
> > I don't see warnings in my regression tests.  Please specify.
> Certainly.  With a toolchain configured as --target=cris-elf and built
> on an x86_64 host running Linux I saw these messages in the ld.log when
> running the linker tests:
>
>    succeeded with: <>, expected: <warning: isatty is not implemented and
> will always fail$>
>    FAIL: ld-cris/warn3

I tried to repeat this but still do not see the failure on
x86_64-unknown-linux-gnu (RH Ent AS 3.0,
spe189.testdrive.hp.com).  No failures in the ld-cris/cris.exp
section.  I do see failures that look like 32/64-bit issues in
the gas testsuite FWIW.

I'm suspecting local patches in your tree (completely without
cause).

I don't know how to proceed here except to leave it, as it works
for me.  Can you repeat the test?

brgds, H-P

Re: Fix for unexpected failure in ld-cris/warn3

[Nick Clifton]

Hi Hans-Peter,

> I tried to repeat this but still do not see the failure on
> x86_64-unknown-linux-gnu (RH Ent AS 3.0,
> spe189.testdrive.hp.com). 

Strange...  I am running pretty the same environment on my host:

   Linux workshop 2.4.21-27.0.2.EL #1 Wed Jan 12 23:35:46 EST 2005 i686 
athlon i386 GNU/Linux

> No failures in the ld-cris/cris.exp
> section.  I do see failures that look like 32/64-bit issues in
> the gas testsuite FWIW.
> 
> I'm suspecting local patches in your tree (completely without
> cause).

Actually with good cause - I do have several of those, but I have just 
backed them out, reconfigured and rebuilt a cris-elf toolchain from 
scratch and the failure still happens...

> I don't know how to proceed here except to leave it, as it works
> for me.  Can you repeat the test?

Yes and the failure is still there.

It occurred to me that the discrepancy might be due to the fact that I 
am using a different version of gcc.  I am using the GNUPro version that 
RedHat ships to various paying customers which is based on gcc 3.4 
rather than the gcc 3.2.3 version which is the system default for RHEL3. 
  Unfortunately I tried reverting to the system compiler and the failure 
still happened.

Where is the stabs warning message generated anyway ?  ie which 
functions in bfd and/or ld are responsible for detecting the stab 
message and printing it out ?

Cheers
   Nick

Re: Fix for unexpected failure in ld-cris/warn3

[Hans-Peter Nilsson]

On Tue, 8 Mar 2005, Nick Clifton wrote:
> > I don't know how to proceed here except to leave it, as it works
> > for me.  Can you repeat the test?
>
> Yes and the failure is still there.
>
> It occurred to me that the discrepancy might be due to the fact that I
> am using a different version of gcc.

That could be it.

>  I am using the GNUPro version that
> RedHat ships to various paying customers which is based on gcc 3.4
> rather than the gcc 3.2.3 version which is the system default for RHEL3.
>   Unfortunately I tried reverting to the system compiler and the failure
> still happened.

Sorry, but are you *sure* that the other compiler was actually
used?

I also tried an ia64-unknown-linux-gnu (spe177.testdrive.hp.com)
no problems there either.  Because I don't see the failure on
*any* system I've tested I can only conclude that there's
something wrong with the system you tested on, or that
something's brittle enough that some other difference mattered
(like, a weak qsort ordering function in binutils coupled with
the brittleness of the warning construct.  Or some uninitialized
variable problem.)

> Where is the stabs warning message generated anyway ?  ie which
> functions in bfd and/or ld are responsible for detecting the stab
> message and printing it out ?

Um, I forgot.  See the patches I sent some time ago, maybe. :-)
IIRC the a.out stuff in bfd just passes on the symbol with
BSF_WARNING and lets the generic stuff generate the warning.

brgds, H-P

Re: Fix for unexpected failure in ld-cris/warn3

[Nick Clifton]

Hi Hans-Peter,

>> I am using the GNUPro version that
>>RedHat ships to various paying customers which is based on gcc 3.4
>>rather than the gcc 3.2.3 version which is the system default for RHEL3.
>>  Unfortunately I tried reverting to the system compiler and the failure
>>still happened.

> Sorry, but are you *sure* that the other compiler was actually
> used?

Well no, not 100%.  But I have just cleared out my ccache, removed all 
other compilers from my search path and rebuild a cris-elf toolchain 
from scratch and the failure still happens. :-(

I am not going to worry about it though.  If you are happy with the test 
results then I am not going to fuss about one small, hard to reproduce 
failure.

Cheers
   Nick