Re: Release 2.23: Ping

[Matthew Gretton-Dann <matthew dot gretton-dann at linaro dot org>]

On 5 September 2012 10:04, Tristan Gingold <gingold@adacore.com> wrote:
>
> On Sep 5, 2012, at 10:54 AM, Matthew Gretton-Dann wrote:
>
>> On 5 September 2012 01:56, Maciej W. Rozycki <macro@codesourcery.com> wrote:
>>> On Tue, 4 Sep 2012, Maciej W. Rozycki wrote:
>>>
>>> Here's the resulting list of regressions:
>>>
>>> arm-aout  +FAIL: Valid v8-a+cryptov1
>>> arm-aout  +FAIL: Valid v8-a+fp
>>> arm-aout  +FAIL: Valid v8-a+simdv3
>>> arm-aout  +FAIL: Valid v8-A barrier (ARM)
>>> arm-aout  +FAIL: Valid v8-A barrier (Thumb)
>>> arm-aout  +FAIL: Valid v8-a
>>> arm-coff  +FAIL: Valid v8-a+cryptov1
>>> arm-coff  +FAIL: Valid v8-a+fp
>>> arm-coff  +FAIL: Valid v8-a+simdv3
>>> arm-coff  +FAIL: Valid v8-A barrier (Thumb)
>>> arm-epoc-pe  +FAIL: Valid v8-a+cryptov1
>>> arm-epoc-pe  +FAIL: Valid v8-a+fp
>>> arm-epoc-pe  +FAIL: Valid v8-a+simdv3
>>> arm-epoc-pe  +FAIL: Valid v8-A barrier (Thumb)
>>> arm-pe  +FAIL: Valid v8-a+cryptov1
>>> arm-pe  +FAIL: Valid v8-a+fp
>>> arm-pe  +FAIL: Valid v8-a+simdv3
>>> arm-pe  +FAIL: Valid v8-A barrier (Thumb)
>>> arm-wince-pe  +FAIL: Valid v8-a+cryptov1
>>> arm-wince-pe  +FAIL: Valid v8-a+fp
>>> arm-wince-pe  +FAIL: Valid v8-a+simdv3
>>> arm-wince-pe  +FAIL: Valid v8-A barrier (Thumb)
>>
>> I'll take a look at these.
>>
>> I don't think they should necessarily hold up the release though as
>> the EABI targets (arm-none-eabi, arm-none-linux-gnueabi) all pass
>> these tests for me.
>
> Fine.
>
> As always, it would be nice to have clean regression output.  Thank you for working to that!
>
> OTOH, these regressions look to be aarch64/armv8 specific, so aren't real regression compared to 2.22 IIUC.

You are correct - these are not new regressions as these are new tests
in 2.23.

Substantial numbers of arm*-*-* tests fail already for non-EABI
targets, for what looks like similar reasons to why this new crop of
tests fail.  Which is that the testcases make implicit assumptions
that are true for EABI targets but aren't necessarily true for other
targets.

Whilst the simple solution would be to mark the tests unsupported for
non-EABI targets this strikes me as being the easy way out, and wrong
as I think these tests should be valid on all arm*-*-* targets.

I will put this on my list of things to look at further, but it is not
going to be high priority - and it certainly shouldn't (in my opinion)
block the binutils 2.23 release.

Thanks,

Matt

-- 
Matthew Gretton-Dann
Linaro Toolchain Working Group
matthew.gretton-dann@linaro.org