[RFC] deprecation of Tree.path_content_summary()

[Martin Pool]

2009/8/25 John Arbash Meinel <john at arbash-meinel.com>:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Ian Clatworthy wrote:
>> John Arbash Meinel wrote:
>>
>>> Given that Martin wants to release a 2.0 and content filters are quite
>>> broken *today*, we need a fix. I personally don't care what it is (it
>>> could be to disable content filters completely, but that seems a bit
>>> overboard).
>>
>> We think the particular problem you found doesn't appear in 2a, "just"
>> 1.14 formats, right?
>
> I believe --2a doesn't ever enable the particular code path. I'm not
> positive about it.
>
> I can say that 1.14-rich-root on Windows is unusable. And given that it
> *exists* to provide Windows people line-ending conversions....
>
>>
>> Content filters still need some important bug fixes over and above this
>> one. I'm planning to make that a priority next week. This week has been
>> lost trying to get data migration, upgrades and GUIs in good shape for
>> the karmic feature freeze.

I wrote a lot (some may say too much, but I like journaling state)
into the bug <https://bugs.edge.launchpad.net/bzr/+bug/415508>.  After
talking to John a bit, this fix may need some more work to not make
things slower on non-iter_changes commits when there are no filters.

I think we might be able to clean this code up by introducing a
separate virtual WorkingTree for the canonical form, so that filters
are applied when going between that and the real working tree.  It
might give a cleaner separation of concerns and cleaner indication
what data can be expected.  It might be too large for 2.0 though.

-- 
Martin <http://launchpad.net/~mbp/>