Re: [PATCH 4/4] coccicheck: add indexing enhancement options

[Julia Lawall]

On Mon, 13 Jun 2016, Luis R. Rodriguez wrote:

> On Mon, Jun 13, 2016 at 09:50:15PM +0200, Julia Lawall wrote:
> > 
> > 
> > On Mon, 13 Jun 2016, Luis R. Rodriguez wrote:
> > 
> > > On Fri, Jun 10, 2016 at 11:21:28PM +0200, Julia Lawall wrote:
> > > > 
> > > > 
> > > > On Fri, 10 Jun 2016, Luis R. Rodriguez wrote:
> > > > 
> > > > > On Fri, Jun 10, 2016 at 11:02:38PM +0200, Julia Lawall wrote:
> > > > > > 
> > > > > > 
> > > > > > On Fri, 10 Jun 2016, Luis R. Rodriguez wrote:
> > > > > > 
> > > > > > > Enable indexing optimizations heuristics. Coccinelle has
> > > > > > > support to make use of its own enhanced "grep" mechanisms
> > > > > > > instead of using regular grep for searching code 'coccigrep',
> > > > > > > in practice though this seems to not perform better than
> > > > > > > regular grep however its expected to help with some use cases
> > > > > > > so we use that if you have no other indexing options in place
> > > > > > > available.
> > > > > > > 
> > > > > > > Since git has its own index, support for using 'git grep' has been
> > > > > > > added to Coccinelle, that should on average perform better than
> > > > > > > using the internal cocci grep, and regular grep. Lastly, Coccinelle
> > > > > > > has had support for glimpseindex for a long while, however the
> > > > > > > tool was previously closed source, its now open sourced, and
> > > > > > > provides the best performance, so support that if we can detect
> > > > > > > you have a glimpse index.
> > > > > > > 
> > > > > > > These tests have been run on an 8 core system:
> > > > > > > 
> > > > > > > Before:
> > > > > > > 
> > > > > > > $ export COCCI=scripts/coccinelle/free/kfree.cocci
> > > > > > > $ time make coccicheck MODE=report
> > > > > > > 
> > > > > > > Before this patch with no indexing or anything:
> > > > > > > 
> > > > > > > real    16m22.435s
> > > > > > > user    128m30.060s
> > > > > > > sys     0m2.712s
> > > > > > > 
> > > > > > > Using coccigrep (after this patch if you have no .git):
> > > > > > > 
> > > > > > > real    16m27.650s
> > > > > > > user    128m47.904s
> > > > > > > sys     0m2.176s
> > > > > > > 
> > > > > > > If you have .git and therefore use gitgrep:
> > > > > > > 
> > > > > > > real    16m21.220s
> > > > > > > user    129m30.940s
> > > > > > > sys     0m2.060s
> > > > > > > 
> > > > > > > And if you have a .glimpse_index:
> > > > > > > 
> > > > > > > real    16m14.794s
> > > > > > > user    128m42.356s
> > > > > > > sys     0m1.880s
> > > > > > 
> > > > > > I don't see any convincing differences in these times.
> > > > > > 
> > > > > > I believe that Coccinelle's internal grep is always used, even with no 
> > > > > > option.
> > > > > 
> > > > > Ah that would explain it. This uses coccinelle 1.0.5, is the default
> > > > > there to use --use-coccigrep if no other index is specified ?
> > > > 
> > > > It has been the default for a long time.
> > > > 
> > > > > > I'm puzzled why glimpse gives no benefit.
> > > > > 
> > > > > Well, slightly better.
> > > > 
> > > > No, it should be much better.  You would have to look at the standard 
> > > > error to see if you are getting any benefit.  There should be very few 
> > > > occurrences of Skipping if you are really using glimpse.  In any case, if 
> > > > you asked for glimpse and it was not able to provide it, there should be 
> > > > warning messages at the top of stderr.
> > > 
> > > I'll redirect stderr to stdout by default when parmap support is used then.
> > 
> > Usually I put them in different files.
> 
> We can do that as well but I would only want to deal with parmap support 
> case. Any preference? How about .coccicheck.stderr.$PID where PID would 
> be the PID of the shell script?

I don't understand the connection with parmap.

Originally our use of parmap made output files based on pids.  Maybe this 
is the default for parmap.  I found this completely unusable.  I guess one 
could look at the dates to see which file is the most recent one, but it 
seems tedious.  If you are putting the standard output in x.out, then put 
the standard error in x.err.

julia