lp:~jameinel/bzr/2.0.1-faster-get-deltas-bug374730

Branch merges

Propose for merging

No branches dependent on this one.

Related source package recipes

No recipes using this branch. (?)

Create packaging recipe

Related snap packages

Related bugs

Related blueprints

4683. By John A Meinel on 2009-09-24

We can check if old matches first, which gives us a couple more hits.
Down to 49s. Not the <36s I'd like to be at, though...

4682. By John A Meinel on 2009-09-24

Caching the specific_fileids => expanded form brings us to 0m51s.

4681. By John A Meinel on 2009-09-24

Refactor the tests a bit.

The time is down a little (1m13s), but it seems to be missing one ancestry
in bzr.dev's log of 'tools'.

4680. By John A Meinel on 2009-09-24

The changes_from() implementation fails the include_parents test.
The previous get_deltas_for_revisions() implementation would include
  mv foo bar
if you did
  bzr log bar/file
(It would show you changes to a parent dir, when logging a child.)
It is arguable either way, but I think we want to preserve this behavior.

Also note that this 'changes_from' implementation is taking 1m35s up from the 36s
I was seeing with the simpler implementation.
My guess is that the paths2ids overhead is significant.

4679. By John A Meinel on 2009-09-24

implement CHKInventoryRepository.get_delta_for_revisions

This first implementation just maps file_ids => paths, and then calls
tree.changes_from().
This isn't ideal for a few reasons, but it should be better than what we've
had.

4678. By John A Meinel on 2009-09-24

Split out get_deltas_for_revision into a separate test module.

4677. By John A Meinel on 2009-09-24

Handle the case when the specific_fileids don't exist in this revision.

4676. By John A Meinel on 2009-09-24

I didn't mean to change get_deltas_for_revisions yet.

I didn't implement support for specific_fileids, which means the results
were probably wrong. However, it does show that going through iter_changes()
first, and then filtering second is going to be vastly faster.
(Time was 38s vs 5m+ for 'bzr log tools')
What is surprising is that 'time bzr log -v' which should just be the simple
iter_changes route, was >2min. So something isn't quite right with those
numbers.

4675. By John A Meinel on 2009-09-24

Implement CHKInventory.filter() on the new subset.

With this implementation, we can probably filter() without reading all nodes,
which is the goal of the exercise.

4674. By John A Meinel on 2009-09-24

Catch a corner case that we were missing.
The CHKInventory tests were passing, but failed for test_inv because
we were passing None to _getitems(). That only failed for InternalNodes,
but we were using a structure that didn't have internal nodes.
So now the test is parameterized on a small CHKInventory page size
to force those things out into the open.