~bzr-pqm/bzr/bzr.dev : changes

~bzr-pqm/bzr/bzr.dev » Changes from revision 4454.3.38

Rev	Summary	Authors	Date
4454.3.38	Start using left-matching-blocks during the actual annotatio... Start using left-matching-blocks during the actual annotation. Down to 11.3s.	John Arbash Meinel	15 years ago
4454.3.37	Add tests tha left-matching-blocks gets populated. Add tests tha left-matching-blocks gets populated.	John Arbash Meinel	15 years ago
4454.3.36	Only cache the content objects that we will reuse. Only cache the content objects that we will reuse. Peak memory is down to 92MB, and annotate time down to 15.9s	John Arbash Meinel	15 years ago
4454.3.35	Figure out why we don't want to copy_base_content=False. Figure out why we don't want to copy_base_content=False.	John Arbash Meinel	15 years ago
4454.3.34	Removing compression_parent Content objs drops us from 238MB... Removing compression_parent Content objs drops us from 238MB down to 146MB. Setting copy_base_content = False seems to cause errors ATM :(. So we need to dig deeper.	John Arbash Meinel	15 years ago
4454.3.33	Change the _expand_record code to pop out old content object... Change the _expand_record code to pop out old content objects. Also, re-use the base parent content object when possible.	John Arbash Meinel	15 years ago
4454.3.32	using this custom extraction code drops us from 30.5s => 17.... using this custom extraction code drops us from 30.5s => 17.6s for annotate NEWS. It would seem that there are some severe inefficiencies for get_record_stream('topological', True)	John Arbash Meinel	15 years ago
4454.3.31	Change the processing lines to now handle fallbacks properly... Change the processing lines to now handle fallbacks properly.	John Arbash Meinel	15 years ago
4454.3.30	add a bit more work to be able to process 'pending_annotatio... add a bit more work to be able to process 'pending_annotations'.	John Arbash Meinel	15 years ago
4454.3.29	Some code comments about what needs to happen. Some code comments about what needs to happen. It also makes it a bit more obvious why the code is hard to wrap your head around. In essense there are 2 queues. Also, we want to intermingle the steps so that the caches can be depopulated as we get the deltas we need, etc.	John Arbash Meinel	15 years ago
4454.3.28	Continue breaking things to build it up cleanly. Continue breaking things to build it up cleanly. At this point we have an _expand_record which is used to turn record streams into text lines. Eventually it will do things like track 'num_compression_children' so that it can re-use content objects, etc. The code already handles removing 'text' entries as part of Annotator. At the moment, nodes get queued up, but we haven't worked out when/how to pull them back out of the queue.	John Arbash Meinel	15 years ago
4454.3.27	Simplify the test__annotator tests by avoiding having a real Simplify the test__annotator tests by avoiding having a real repository.	John Arbash Meinel	15 years ago
4454.3.26	The new _KnitAnnotator based on Annotator seems to pass the ... The new _KnitAnnotator based on Annotator seems to pass the test suite. Now we just need to clean up the guts and get left-matching-blocks working.	John Arbash Meinel	15 years ago
4454.3.25	Use BranchBuilder for duplicate_lines_tree Use BranchBuilder for duplicate_lines_tree	John Arbash Meinel	15 years ago
4454.3.24	update BranchBuilder to support 'committer' update BranchBuilder to support 'committer' Update test_annotate to use BranchBuilder rather than lots of trees and tree.commit.	John Arbash Meinel	15 years ago
4454.3.23	Initial attempt at refactoring _KnitAnnotator to derive from... Initial attempt at refactoring _KnitAnnotator to derive from Annotator. So far it seems possible, by just overriding _get_needed_texts. However, we'll also want to override the get_matching_blocks stuff based on the delta.	John Arbash Meinel	15 years ago
4454.3.22	Need to record the other annotations before we can record th... Need to record the other annotations before we can record this, otherwise popping the extra annotations fails. Factor out an '_annotate_one' helper, as it makes the code cleaner, especially since we added a progress bar.	John Arbash Meinel	15 years ago
4454.3.21	Assert that entries in the annotation cache also get cleaned... Assert that entries in the annotation cache also get cleaned up. This actually drops peak mem by almost 100MB, I wouldn't think the annotations would be that expensive... Anyway, texts themselves are still held by the VF stream, so we only save for other things that we might be caching.	John Arbash Meinel	15 years ago
4454.3.20	Prev patch doesn't help peak mem on Packs because get_record... Prev patch doesn't help peak mem on Packs because get_record_stream() caches all fulltexts. however, the final mem did go from 300MB down to 200MB, still up from what I would expect.	John Arbash Meinel	15 years ago
4454.3.19	Have _record_annotation start to remove texts when they are ... Have _record_annotation start to remove texts when they are no longer needed.	John Arbash Meinel	15 years ago

Older »