Repository's .revisions, __getslice__, __getitem__ and .get_changesets methods changes

[lukaszb]

As a general idea, in git when You iterate the log, it shows You only log for current (e.g. master) branch. This is much different than what mercurial is showing.

My proposal is to add, 'getiterator(branch=None)' function that will generate iterator function for mercurial repository and will iterate only on revision for passed branch parameter. That would be a nice addition to next/prev functions of changeset.

---

Reference: https://bitbucket.org/marcinkuzminski/vcs/issues/44/

[lukaszb]

revisions

• should always contain only revisions' IDs (for hg, git that would be hashes, for svn integers)
• ordered by date (ascending)
• implements list interface
• future: should by more lazy (performance)

get_changesets

Synpisis:

get_changesets(start=None, end=None, start_date=None, end_date=None,
                     branch_name=None, reversed=False)

• all paremeters are optional
• all parameters work as filters for returned generator

getslice

• in practice, should pass parameters to self.revisions and return generator using .get_changesets
• should allow integer parameters too (i.e. to get last ten changesets we use repo[-10:]

getitem

• should use same implementation as getslice (allow int, non positive int or ID as parameter)

[marcinkuzminski]

From what i have read here:http://mercurial.selenic.com/wiki/ChangeSetID short_id(12 hex) should not be considered a valid long term identifier,so i'm proposing to forbid passing short_id as for getting changesets eg. get_changeset, or slicing. It might explode one day...

That should be left just for human readable representation

[marcinkuzminski]

On the other hand, after a talk with mercurial crew I'll make an algorithm that will lookup raw id based in short id since it;s unlikely to have such collision just in one repo.

[lukaszb]

Done at both hg and git backends, all tests passes.