[darcs-users] filecache and making darcs faster

[Simon Marlow]

On 28/08/2009 15:45, Jason Dagit wrote:
> On Fri, Aug 28, 2009 at 7:11 AM, Simon Marlow<marlowsd at gmail.com>  wrote:
>> On 26/08/2009 14:00, Eric Kow wrote:
>>
>>> Is Darcs slower on Windows and MacOS X than
>>> on Linux?
>>
>> I don't know if this is a new one, but 'darcs whatsnew' with hashed
>> repositories on Windows seems to have a performance problem:
>
> I don't know if it's a new one either, but it's new to me and interesting.
>
>>
>> $ darcs show repo
>>           Type: darcs
>>         Format: hashed
>>           Root: d:/builds/ghc-testing
>>       Pristine: HashedPristine
>>          Cache: thisrepo:d:/builds/ghc-testing, cache:C:\Documents and
>> Settings\simonmar\Application Data\darcs\cache
>> boringfile Pref: .darcs-boring
>> Default Remote: x:/ghc-HEAD
>>    Num Patches: 21106
>> $ time darcs w -s
>> No changes!
>>
>> real    0m3.126s
>> user    0m0.000s
>> sys     0m0.015s
>> $ darcs --version
>> 2.3.0 (release)
>>
>> Compared to Linux, where the time is ~0.25s.  These are all local
>> filesystems.
>
> I see that the vast majority of that time is reported as "real".  Does
> that give us any leads to trouble shoot this?  My vague understanding
> of the time command is that when there is a lot of "real" time that
> often means the program spent most of its time in system calls or
> doing IO.
>
>> (I'd like to use hashed repos on Windows, mainly to avoid the
>> case-insensitive-filesystem issue with non-hashed repos).
>
> I don't know how much hassle this would be for you, but anytime you
> can send us detailed profiling reports or specific workflows that are
> too slow that helps us.  Knowing that whatsnew --summary is too slow
> on the ghc repo is a good example.  But also, if you sent us the
> darcs.prof file that helps too.

If it's easy to build a profiled darcs, then I'll happily do it (cabal 
install darcs --enable-profiling?).

> And of course, Petr's work will eventually speed this up quite a bit.
> I don't recall if whatsnew was improved yet in HEAD, but it's on the
> roadmap as I understand it.

This is darcs 2.3.0, which has Petr's whatsnew improvements in it, which 
is why the result is so surprising!

Cheers,	
	Simon