Re: [PATCH 0/11] renaming argv_array

From: Jeff King <peff@peff.net>
Date: Aug 12, 2020, 19:57 UTC
Message-ID: <20200812195703.GA1340519@coredump.intra.peff.net>
In-Reply-To: <13d0b147-3c62-5d73-630e-2305c08c3ab1@web.de>

On Wed, Aug 12, 2020 at 08:18:33PM +0200, René Scharfe wrote:

Show 9 quoted lines

> Am 12.08.20 um 19:08 schrieb Jeff King:
> > Yeah, I've had various issues with the packaging. For a long time they
> > had 1.0.7 in experimental, but with no python support. I wonder if it's
> > worth starting to use python scriptlets in our coccinelle rules, as
> > described in 4d168e742a (coccinelle: use <...> for function exclusion,
> > 2018-08-28). They're faster and IMHO easier to understand.
> 
> The idea to use Python as a faster alternative to anything makes me
> a bit uneasy.  That can't be right. ;-)

Well, when you're comparing it to some exponential algorithm with the other technique, even a scripted language can do well. :)

I definitely remember getting big speedups back at the time of that commit. But applying the patch below, building object_id.cocci.patch actually gets a few seconds slower. So maybe something changed between the various coccinelle versions. At any rate, it doesn't seem worth pursuing further.

Show 9 quoted lines

> > Of course I tried it out and got some inscrutable errors:
> >
> >   SPATCH contrib/coccinelle/object_id.cocci
> >   init_defs_builtins: /usr/bin/../lib/coccinelle/standard.h
> >   Python path configuration:
> >     PYTHONHOME = '/lib/x86_64-linux-gnu/..'
> 
> This looks bogus.  Can you try to unset this environment variable?
> Or set it to "/usr"?

It's not set in my environment. However, I was able to solve it by fiddling with my system python packages (I'm not sure of the exact cause; it was a case of corrected-while-testing).

-Peff