vtable.exp fails on ARM...

vtable.exp fails on ARM...

[scottb]

I get the following test failing on a NetWinder
(armv4l-unknown-linux-gnu).

Running /cvstree/binutils/gas/testsuite/gas/vtable/vtable.exp ...
FAIL: vtable entry1

This test is run by the following code from
gas/testsuite/gas/vtable/vtable.exp 

  # The vtable entry results are different on Rel and Rela targets.
  if {[istarget "i*86-*-*"] || [istarget "mips*-*-*"]} then {
    run_dump_test "entry0"
  } else {
    run_dump_test "entry1"
  }

The output from objdump -r is:

  dump.o:     file format elf32-littlearm

  RELOCATION RECORDS FOR [.text]:
  OFFSET   TYPE              VALUE 
  00000010 R_ARM_GNU_VTENTRY  vtbl_a

This matches the output using entry0.

I'm not sure what the exact problem is, but I suspect it stems from our
use
of REL relocs in ARM Linux code generation, and the other ARM targets
using
RELA relocs.

I suspect adding an [istarget "armv*-*-linux*"] to the if clause will 
solve the problem, but thought I would ask since ARM Linux doesn't use
this type of relocation record to my knowledge.

Any comments?

Scott

Re: vtable.exp fails on ARM...

[Catherine Moore]

:I'm not sure what the exact problem is, but I suspect it stems from our
:use
:of REL relocs in ARM Linux code generation, and the other ARM targets
:using
:RELA relocs.
:
:I suspect adding an [istarget "armv*-*-linux*"] to the if clause will 
:solve the problem, but thought I would ask since ARM Linux doesn't use
:this type of relocation record to my knowledge.
:
:Any comments?

   The default for all of the ARM targets is REL relocs at this point.
   The expect script needs to be changed to reflect that.  The relocation
   is used to support the linker option --gc-sections.

   --catherine

Re: vtable.exp fails on ARM...

[scottb]

Catherine Moore wrote:
> 
> :I suspect adding an [istarget "armv*-*-linux*"] to the if clause will
> :solve the problem, but thought I would ask since ARM Linux doesn't use
> :this type of relocation record to my knowledge.
> :
> :Any comments?
> 
>    The default for all of the ARM targets is REL relocs at this point.
>    The expect script needs to be changed to reflect that.  The relocation
>    is used to support the linker option --gc-sections.

I have changed the expect script as mentioned above.  The patch is
attached.  This fixes the failure for me.

Scott

1999-09-01  Scott Bambrough  <scottb@netwinder.org>

	* gas/vtable/vtable.exp: Added [istarget "arm*-*-*"] to run dump 
	test "entry0" instead of "entry1".  ARM targets use REL relocs,
	and should use the test entry0.
vtable.diff

Re: vtable.exp fails on ARM...

[Nick Clifton]

Hi Scott,

: I have changed the expect script as mentioned above.

Great - thanks.  I have also added [istarget "strongarm*-*-*"] to the
test.

Cheers
	Nick