On Fri, Apr 25, 2008 at 3:53 PM, Gwern Branwen <[EMAIL PROTECTED]> wrote:

> On 2008.04.25 13:42:04 -0700, Jason Dagit <[EMAIL PROTECTED]> scribbled
> 1.9K characters:
> >    On Fri, Apr 25, 2008 at 1:26 PM, Don Stewart <[EMAIL PROTECTED]> wrote:
> >
> >      gwern0:
> >      > Fri Apr 25 16:01:53 EDT 2008  [EMAIL PROTECTED]
> >      >   * fpstring.c: switch a memchr for memrchr
> >      >   See <http://bugs.darcs.net/issue814>; memrchr speeds up
> is_funky quite a bit and thus
> >      helps whatsnew -s. It doesn't seem to break (any more) tests.
> >
> >
> >      memrchr isn't available on BSDs.
> >
> >    Furthermore, I didn't see enough tests in Gwern's bug update to
> convince me that the
> >    difference in run-times is attributable to the use of memrchr.
> >
> >    Jason
>
> I'm not sure how I could've tested it any more: I used a fresh darcs repo,
> made the simple change, compiled and installed, and tested it on a 9.5GB
> file 3 or 4 times, each time getting a roughly 40 or 50 second result from
> 'time'; I then in the same repo tested three or four times using my system's
> darcs binary (or installing from one of my other darcs darcs repos) and get
> roughly 1m30s timings.
>
> A single change, which leads to an expected difference multiple times; what
> did I miss?


I was commenting on this:
http://bugs.darcs.net/msg4338

Which doesn't indicate (or if it does I'm overlooking it) that you ran each
test multiple times, either to get the standard deviation or an intuitive
sense of it.

Looking that over, I see this as the time it takes to run without changes:
(; darcs whats -s; ) 2.03s user 4.43s system 7% cpu 1:27.42 total

I'll just assume then that this timing is representative.

And then I see in your previous email that the times range from ~53 seconds
to ~68 seconds.  That's a pretty good improvement in time although it's for
a 9.5 GB file so on smaller files it may be negligible.  But, yeah I'm all
for this change.  I think I would prefer a change where we don't read more
than 1 page of memory (remembering that we mmap files) looking for the magic
bytes.  If we look at, say 4k, of a file and don't see funky bytes, probably
darcs should give up so that we don't cause the OS to swap everything out.

Jason
_______________________________________________
darcs-users mailing list
[email protected]
http://lists.osuosl.org/mailman/listinfo/darcs-users

Reply via email to