[MERGE] #25449

[Aaron Bentley]

Roberto Aguilar wrote:
> On Feb 26, 2009, at 6:55 AM, Aaron Bentley wrote:
>> [...]
>> I think it would be better to move more of cmd_merge.run into Merger, so
>> that compare_basis can be invoked before bundle.read_mergeable_from_url.
>> Even the revised patch is changing things because it's not doing
>> Merger.revision_tree, which gives the Merger a chance to cache it.
> 
> Just getting the merger object is taking quite a while.

There's no intrinsic reason why it should.

> I added a bunch
> of timestamps around the code to see how long things were taking.  On my
> machine, getting the merger object when merging a remote branch via ssh
> took between 3.6 and 4.7 seconds across 5 runs.  With the test where I
> placed it, although it's not caching the result, uncommitted changes
> errors out between 0.54 and 0.65 seconds.  Putting this test in the
> Merger object will delay the error by quite a bit.

The point of putting it in the object would be so that getting the
Merger object would fail, and fail quickly, before doing expensive
things like read_mergeable_from_url.

> I have modified the test so that it passes and checks that
> uncommitted changes error out.  Please let me know if the tests look
> alright.

This looks fine.

bb:approve

Aaron