Re: [PATCH 0/11] renaming argv_array

From: René Scharfe <l.s.r@web.de>
Date: Aug 12, 2020, 16:23 UTC
Message-ID: <045ea49b-7165-0f45-e670-a24bf648f880@web.de>
In-Reply-To: <20200812151038.GB33189@coredump.intra.peff.net>

Am 12.08.20 um 17:10 schrieb Jeff King:

Show 11 quoted lines

> On Wed, Aug 12, 2020 at 11:06:11AM -0400, Jeff King wrote:
>
>> Debian unstable now ships coccinelle 1.0.8, and it's unbelievably slow
>> compared to 1.0.4. Running "make coccicheck" is currently at 80 minutes
>> of CPU time running each script in parallel, with none of them down.
>> They're also all consuming 6GB of RAM each, so I'm killing them all.
>
> This seems to be related to setting SPATCH_BATCH_SIZE to "0". It used to make
> things go much faster (if you had enough memory), but now seems to just
> consume tons of CPU. Setting it to "1" finishes the whole thing in ~13
> minutes of CPU (~2m wall-clock).

This bit me as well, and I settled with SPATCH_BATCH_SIZE = 10. With MAKEFLAGS += -j3 I get these number, which are quite similar to yours (except I don't dare use more cores due to cooling issues..):

  real	4m12,393s
  user	12m15,447s
  sys	0m10,418s

> So that's at least a path forward, but in general I have been frustrated
> with operational aspects of coccinelle like this.

And I was a bit shocked when Coccinelle's testing package became unmaintained for a while and I had to compile it from source.

And yes, coccicheck is quite heavy. When I merge all .cocci files into one I get:

  real	2m7,164s
  user	2m5,389s
  sys	0m1,572s

Nice.  With spatch -j3 I get basically the same numbers, though.  Hmm.

René