develooper Front page | perl.perl5.porters | Postings from November 2014

Re: some tests are slower than others

Thread Previous
From:
Jarkko Hietaniemi
Date:
November 29, 2014 02:33
Subject:
Re: some tests are slower than others
Message ID:
5479305A.9080809@iki.fi
> Grrr.  Anything to do with stress-testing Unicode....  (regen and
> podcheck are slow for other reasons hard to solve.)

I heard that demerphq had some opinions on the porting tests... :-)

>> cpan cpan/Memoize/t/expmod_t.t 8 186.3 184.2 0.989
>
> 'Not our business.'  (Unless you send a patch upstream.)

Naturally.  But I left it in, not to editorialize too much.

>> core re/pat_psycho.t 8 165.5 71.8 0.434
>
> The name says it all.

Still, if there's a method to the madness, it can be
hoisted as loop invariant, or something.

I'll give away my secret plan: some of these tests
might be useful to skip for some systems, where they
simply burn CPU for no good reason.  In fact, the pat_psycho
already has a bypass of its own, for exactly this purpose.

Another example is the IO-Compress tests.  Is running
compression and uncompression *C* code back and forth really
the most useful thing, for the core testing?  Again, I think
some of them already are skipped.

>> core lib/Benchmark.t 8 158.8 156.9 0.988
>
> What do you expect?

Ditto as for Memoize.

>> core lib/perl5db.t 8 155.5 19.4 0.124
>
> That would be a nice one to fix, but not easy.  It does a lot of
> fresh_perling.
>
>> dist dist/IO/t/io_sock.t 8 148.0 145.9 0.985
>
> No idea.  But anything socket-related is out of my league.  (Are the
> tests blocking on purpose to test timeouts?)

Possibly.  Here differentiating between burning user time
and system time, and the ratio of wall time versus cpu time,
might become useful.   (I've got all those numbers, just had
to collapse the dimensions somehow.)

>> core uni/variables.t 8 147.8 127.5 0.862



Thread Previous


nntp.perl.org: Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at ask@perl.org | Group listing | About