Re: [6.5.4] perf: bison min. requirements

[Ian Rogers <irogers@google.com>]

On Fri, Sep 29, 2023 at 9:42 AM Anisse Astier <anisse@astier.eu> wrote:
>
> Hi Ian,
>
> On Tue, Sep 26, 2023 at 02:28:50PM -0700, Ian Rogers wrote:
> > On Tue, Sep 26, 2023 at 1:21???AM Pierre Cheynier <p.cheynier@criteo.com> wrote:
> > >
> > > Dear list,
> > > At compiling 6.5.4 it seemed to me that this specific patch [1] added an implicit dependency to bison >= 3.7.4, due to usage of YYNOMEM.
> > > Shouldn't this be part of Documentation/process/changes.rst?
> > >
> > > [1] https://www.spinics.net/lists/linux-perf-users/msg30885.html "perf parse-events: Separate YYABORT and YYNOMEM cases"
> > >
> > > Sorry for the noise, and thanks for inputs if ever I'm totally wrong here.
> >
> > We have a workaround to #define YYNOMEM to YYABORT when Bison is
> > less-than 3.8.1:
> > https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/Build?h=perf-tools-next#n323
> > Perhaps you can help debug why this build logic isn't working for you?
>
> This fix (commit 88cc47e245979 in perf-tools-next) has reached Linus'
> tree, but not yet stable.
>
> But code using a more recent Bison is now backported to v6.5.4 (stable
> commit 0e8501c8a936e0b2ff4ffdff2c4866ba74969189, upstream commit
> 77cdd787fc45e3426b8e0b5038b85c276540dfb4), which is what causes the
> build regression on older bison that Pierre reported.
>
> Someone else noticed and already asked for a backport:
> https://lore.kernel.org/all/CAM9d7cggeTaXR5VBD1BoPr9TLPoE7s9YSS2y0w-PGzTMAGsFWA@mail.gmail.com/

Great! Fwiw, Linux perf is backward compatible with older kernels and
so using the backport version makes little sense. Ideally Linux perf
would be decoupled by package maintainers and allowed to follow Linux
at head.

Thanks,
Ian

> Kind regards,
>
> Anisse