Re: [PATCH v2 2/2] perf jevents: Autogenerate empty-pmu-events.c

[Ian Rogers]

On Wed, Jun 12, 2024 at 8:07 PM Oliver Sang <oliver.sang@xxxxxxxxx> wrote:
> > > > but make still failed. I still saw below in our build log
> > > >
> > > > --- pmu-events/empty-pmu-events.c       2024-06-03 08:41:16.000000000 +0000
> > > > +++ pmu-events/test-empty-pmu-events.c  2024-06-03 13:47:19.522463482 +0000
> > > > @@ -136,7 +136,7 @@
> > > >  { 2623 }, /* M3\000\0001 / M3\000\000\000\000\000\000\000\00000 */
> > > >  { 2078 }, /* cache_miss_cycles\000group1\000dcache_miss_cpi + icache_miss_cycles\000\000\000\000\000\000\000\00000 */
> > > >  { 1947 }, /* dcache_miss_cpi\000\000l1d\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000 */
> > > > -{ 2011 }, /* icache_miss_cycles\000\000l1i\\-loads\\-misses / inst_retiredany\000\000\000\000\000\000\000\00000 */
> > > > +{ 2011 }, /* icache_miss_cycles\000\000l1i\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000 */

Hi Oliver,

I tried to reproduce the problem by installing more locales on my
machine, setting the environment variables, checking my environment
variables, etc. I couldn't reproduce it. Looking more closely at the
diff above and the other diffs in your email something strange appears
to be happening around dots.

From:
 --- pmu-events/empty-pmu-events.c
we know the minus lines are those from the files in this patch series.
From:
+++ pmu-events/test-empty-pmu-events.c
we know the plus lines are those generated by jevents.py.

In the diff output above the "inst_retired.any" doesn't match
"inst_retiredany" (no dot before the word "any"). The repository file
pmu-events/empty-pmu-events.c is missing the dot in the event name but
jevents.py is generating it, hence the diff. But looking at
pmu-events/empty-pmu-events.c in:
https://lore.kernel.org/lkml/20240525013021.436430-2-irogers@xxxxxxxxxx/
The dot is present.

I think what is happening is that when you apply the patches for some
reason the dots are being consumed in
tools/perf/pmu-events/empty-pmu-events.c, the build then informs you
of this by failing. The locales idea was a red herring and this has
something to do with how you apply patches.

Does this make sense? Perhaps you can try testing the patches in an
ordinary client applying the patches using something like "b4 am
20240525013021.436430-1-irogers@xxxxxxxxxx". At the moment I think the
patch series is good and I don't have a way to fix what I think the
problem is, with how you applied the patches.

Thanks,
Ian