[PATCH] Simplify testing symbol sections

David Edelsohn dje.gcc@gmail.com
Mon Nov 16 23:14:52 GMT 2020

On Mon, Nov 16, 2020 at 3:41 PM Jeff Law <law@redhat.com> wrote:
> On 11/14/20 7:08 PM, David Edelsohn wrote:
> > On Sat, Nov 14, 2020 at 8:58 PM Jeff Law <law@redhat.com> wrote:
> >>
> >> On 11/14/20 6:35 PM, David Edelsohn wrote:
> >>>>>>>> Jeffrey Law wrote:
> >>>> I worry a bit about the less common native targets -- aix, hpux and the
> >>>> like.  But testing them is too painful to contemplate these days.  I'm
> >>>> sure those with access to suitable hardware will chime in if something
> >>>> is amiss.
> >>> All of these testcases now fail on AIX with "no section detected".
> >>> One cannot XFAIL the scan, one must skip the entire test because of
> >>> the manner in which scan-assembler-symbol-section works.
> >>>
> >>> And, Jeff, the "too painful to contemplate" snide comment is playing
> >>> the victim.  There are multiple AIX systems in the GNU Compile Farm
> >>> with instructions on how to bootstrap GCC on the system.  Jonathan
> >>> Wakely and others don't have a problem testing patches on AIX.
> >>>
> >>> Other people are able to more thoroughly test patches.  As one of the
> >>> original GCC developers and a member of the GCC SC, this sets a poor
> >>> example of patch development and testing.  This patch should not have
> >>> been pushed and should be reverted until it can gracefully fail on
> >>> non-ELF targets.
> >> I've tried repeatedly through the years to use the compile farm to build
> >> aix without success at some point it's just no longer worth my time.  I
> >> tested far more targets than is required by our policies and procedures
> >> and I made a decision to move forward.
> >>
> >>
> >> Let's just xfail the tests for aix and get on with our lives.
> > As I wrote, XFAIL doesn't work with scanasm.
> >
> > I actually have expanded scan-assembler-symbol-section to recognize
> > AIX CSECT sections and I am updating the tests to check for the
> > appropriate strings generated in AIX XCOFF.  It all just works.
> >
> > It would have been nice if you and Matthew had given me and other less
> > common targets a heads up to test the patch.
> Note this patch series has been pending for a year, that was plenty of
> time for someone to pick it up and move it forward or address issues on
> non-ELF targets.

If one were aware of the patch series and its potential impact on
non-ELF targets.

> Note there's still two more patches in this space that
> were the motivating cases behind the infrastructure that's been
> committed.  The first fixes an LTO vs attribute section problem that
> looks ready to go, but could well break on aix.  The second with
> attribute section and templates that I don't think is ready and I'm not
> considering it for gcc-11 right now.
> What would be helpful would be for someone to make the jenkins tester
> work for aix.  If that was in place I could test the patch by putting it
> into a magic directory on github and firing off an aix build and it'd be
> automatically compared against the prior successful build :-)
> I can't take that on, but I can help with getting the aix box connected
> to the tester and high level issues.  But its' the precise paths and
> sequencing that someone familiar with the ins and outs of aix would have
> to be intimately involved with.

Jenkins does function on AIX.  I will take an action item to create
another LPAR on the AIX systems at OSUOSL for Jenkins and coordinate
with you to connect it to the terrific Jenkins infrastructure for GCC
that you have deployed.

Thanks, David

More information about the Gcc-patches mailing list