[PATCH RFC] obsolete: mark unreachable extinct changesets as hidden

[Patrick Mézard]

Le 16/07/12 13:00, Idan Kamara a écrit :
> On Fri, Jul 13, 2012 at 11:37 PM, Pierre-Yves David <
> pierre-yves.david at ens-lyon.org> wrote:
>> # HG changeset patch
>> # User Pierre-Yves.David at ens-lyon.org
>> # Date 1342192006 -7200
>> # Node ID a08309196cf89676b4f611b7384d1256dc1d7f82
>> # Parent  de13ccd4ef8d1ed1abc052a57a88b757c17a0594
>> obsolete: mark unreachable extinct changesets as hidden
>>
>> The repo.changelog.hiddenrev set is update with all extinct() changeset
> which
>> ain't descendants of either:
>>
>> - the current working copy,
>> - a bookmark,
>> - a tag.
>>
>> It's not really possible to make a lazy filling this set. So we fill it
>> unconditionally. This will have a noticeable performance impact on simple
>> operation. I expect the future changelog filtering to allows such lazy
>> computation.
>>
>> This changeset focus on a simple implementation of that implement the
> right
>> behavior. A better implementation will be necessary before it's made
> available
>> to every user.
>>
>> diff -r de13ccd4ef8d -r a08309196cf8 mercurial/localrepo.py
>> --- a/mercurial/localrepo.py    Fri Jul 13 14:38:49 2012 +0200
>> +++ b/mercurial/localrepo.py    Fri Jul 13 17:06:46 2012 +0200
>> @@ -207,6 +207,29 @@
>>              p = os.environ['HG_PENDING']
>>              if p.startswith(self.root):
>>                  c.readpending('00changelog.i.a')
>> +        if self.obsstore:
>> +            ### hide extinct changeset that are not accessible by any
> mean
>> +            # XXX Making this for each changelog creation force complex
> graph
>> +            # XXX computation this is a bad idea and we should have a
> more lazy
>> +            # XXX approach here. But this is ok for a first quick and
> dirty
>> +            # XXX version as used in several other place regarding
> obsstore.
>> +            # XXX Repo without any obsolete marker won't be impacted
>> +            hiddenquery = 'extinct() - ::(. + bookmark() + tagged())'
>> +            # The repo need a changelog to run revset. But don't have any
>> +            # because this method is meant to provide one. However the
>> +            # changelog is already fully parsed and ready to use. We
> temporally
>> +            # assign it as an attribute to be able to run revset.
>  Later, more
>> +            # lazy implementation will probably be called after this
> step and
>> +            # won't suffer from this issue
>> +            self.__dict__['changelog'] = c
>> +            # Note that the hiddenrevs needs invalidations when
>> +            # - a new changesets is added (possible unstable above
> extinct)
>> +            # - a new obsolete marker is added (possible new extinct
> changeset)
>> +            c.hiddenrevs.update(self.revs(hiddenquery))
>> +            # remove inserted changelog from class attribut to let
> storecache
>> +            # do it's jobs
>> +            if self.__dict__['changelog'] is c:
>> +                self.__dict__.pop('changelog')
> 
> This is pretty nasty even for a hack.
> 
> Since nothing is going to use obsolete in core this version, can't this be
> done outside in an extension (in a more sane way through reposetup)?

Wouldn't moving hiddenrevs from changelog to localrepo solve the data dependency issue and help with lazy initialization?

--
Patrick Mézard