[darcs-users] Google Summer of Code 2009 - ideas, students and mentors?

Gwern Branwen gwern0 at gmail.com
Thu Feb 12 16:25:00 UTC 2009


On Thu, Feb 12, 2009 at 10:55 AM, Petr Rockai <me at mornfall.net> wrote:
> Eric Kow <kowey at darcs.net> writes:
>> * A Windows project: not sure what this would entail, but darcs on Windows
>>   might need a bit of special attention
>>       o 2008-08 : proposed as a priority for darcs 2.2
> Hard to tell. If someone would be interested, they can work out what this
> entails and tell us. :)
>
>> * A single-file database pristine cache, which should be faster and more robust
>>   than our directory-based approach. e.g. a halfs filesystem.
>>       o 2008: darcs 2.0.0 has hashed pristine, which is more robust, but it's not single-file. do we still want this?
> No, unless we want to abolish lazy repos. This is however something that could
> very well be part of my work on hashed-storage, a more tightly-packed hashed
> pristine (and also patches), by using some sort of generic pack generation.
>
>> * Keeping track of which patches affect which files, to speed up query
>>   operations that affect only certain files (e.g. changes, diff, etc, when
>>   given a filename argument).
>>       o 2008-06 status: I don't think we do this yet. Still open?
> This is still open, see my previous mail on this topic.
>
>> * issue1007 Use Data.ByteString.Lazy for better performance
>>       o 2008-08: we now have bytestrings... we just need lazy ones
> Partially covered by hashed-storage, which uses lazy bytestrings everywhere. It
> currently does fairly costly conversion to strict bytestrings on the darcs <->
> hashed-storage boundary, but that can be sorted out later.
>
>> * issue647 A benchmark suite for darcs. Possibly make it version-control-system
>>   independent, so we could have automated and meaningful performance benchmarks
>>   against competitors? Would need to be flexible to simulate real workloads and
>>   automatically generate real repositories.
>>       o 2008: See http://code.google.com/p/maybench - very very early in development. Needs volunteers
> This could become a separate project (in the wider variant, with other-rcs
> comparison). Maybe build on darcs-benchmark? I have no idea about maybench
> though.

A thing that might be worth remembering in this context is that
filestore has a Tests.lhs which uses the filestore commands to
exercise basic functionality and see whether they work & satisfy basic
properties we think they should. Right now, Tests.lhs doesn't serve as
much of a benchmark - darcs finishes the tests faster than git! I
think that may be because the largest files that Tests.lhs generates
are about 5 lines long; if they were a few megabytes of /dev/random
output, then maybe the code could turn into benchmarks as well as
tests?

> --
> Peter Rockai | me()mornfall!net | prockai()redhat!com

-- 
gwern


More information about the darcs-users mailing list