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

~bzr-pqm/bzr/bzr.dev » Changes from revision 3735.36.11

Rev	Summary	Authors	Date
3735.36.11	Shave a little bit of time by using itervalues() rather than... Shave a little bit of time by using itervalues() rather than casting through refs()	John Arbash Meinel	15 years ago
3735.36.10	type(node) is InternalNode is supposedly better than isinsta... type(node) is InternalNode is supposedly better than isinstance(node, InternalNode)	John Arbash Meinel	15 years ago
3735.36.9	Simple fix to avoid using small.difference_update(large) Simple fix to avoid using small.difference_update(large) It seems the obvious thing to do, but Python's implementation scales poorly. small = small.difference(large) scales much better [O(small) rather than O(large)].	John Arbash Meinel	15 years ago
3735.36.8	Merge bzr.dev 4208. Merge bzr.dev 4208. This brings in some more smart-server improvements, as well as the iter_files_bytes as chunked, and multi-file and directory logging.	John Arbash Meinel	15 years ago
3735.36.7	Get rid of inline and const, to reduce warnings and errors. Get rid of inline and const, to reduce warnings and errors. It seems compilers don't like it if you pass &(unsigned char ) to (const unsigned char ), and pyrex doesn't let you define 'const unsigned char', (it doesn't like const at all), so for now, just remove it, because it doesn't hurt anything. Removing inline because MSVC doesn't understand it, and causes compile failures. It wasn't really important anyway.	John Arbash Meinel	15 years ago
3735.36.6	get inventory filtering working for CHKInventory get inventory filtering working for CHKInventory	Ian Clatworthy	15 years ago
3735.36.5	iter_interesting_nodes really is a culprit causing us to hol... iter_interesting_nodes really is a culprit causing us to hold on to a record after iterating to the next record. We should try to fix that code first.	John Arbash Meinel	15 years ago
3735.36.4	Fix the GPL and copyright statements in the pyrex files Fix the GPL and copyright statements in the pyrex files	John Arbash Meinel	15 years ago
3735.36.3	Add the new address for FSF to the new files. Add the new address for FSF to the new files.	John Arbash Meinel	15 years ago
3735.36.2	Merge bzr.dev 4198, expect to have a bit of fallout, but wil... Merge bzr.dev 4198, expect to have a bit of fallout, but will resolve in followup commits.	John Arbash Meinel	15 years ago
3735.36.1	Fix InternalNode._iter_nodes to split the key filter based o... Fix InternalNode._iter_nodes to split the key filter based on matches. Update LeafNode.iteritems() to use a dict lookup when the key width is correct. Shaves a reasonable amount of time off of stuff that wants to look at large portions of the inventory.	John Arbash Meinel	15 years ago
3735.2.165	fix iter_just_entries() in CHKInventories fix iter_just_entries() in CHKInventories	Ian Clatworthy	15 years ago
3735.2.164	Fix a critical bug that caused problems with the index entri... Fix a critical bug that caused problems with the index entries. We were grabbing the 'start_point' but not using it, and we weren't resetting basis_end correctly. So we ended up recording the end of the previous record as the start of the next.	John Arbash Meinel	15 years ago
3735.2.163	Merge bzr.dev 4187, and revert the change to fix refcycle is... Merge bzr.dev 4187, and revert the change to fix refcycle issues. I apparently didn't run the smart fetch tests. Which show that we access inv+chk pages as a fulltext, and then insert the stream, which expects to get the block as a compressed block. :(. Need to rethink how to do it, possibly with weakrefs. This also brings in CommitBuilder.record_iter_changes() and the updates to btree_index and backing indices.	John Arbash Meinel	15 years ago
3735.2.162	Change GroupCompressor.compress() to return the start_point. Change GroupCompressor.compress() to return the start_point. Also, mark empty content with start=end=0. This also gives us a good starting point to handle duplicate entries (if we find that makes a difference.) From experimentation, using 0,0 for empty entries actually makes a big difference in the text index. Mostly because about 1/2 of all entries have no content, (all of the directory records, for example), so it allows the compression to shrink the index a bit.	John Arbash Meinel	15 years ago
3735.2.161	Change _LazyGroupContentManager.get_record_stream() to inval... Change _LazyGroupContentManager.get_record_stream() to invalidate the nodes. This means that records are genuinely only valid until you go to the next one. This breaks the reference cycle with the manager, and drops peak memory by a decent amount. (as otherwise we have to wait until gc.collect() runs). Also, change get_bytes_as() so that it caches the bytes (though this cache is also removed when we go to the next object). This is mostly useful for CHK pages, which get processed at several layers. The get_record_stream() code uses it to determine what to send next, and the insert_record_stream() needs the raw bytes to put into the target repo. This saves an extract() on every chk page during 'pack'. apply_delta is rather fast, but there is no reason to waste calls.	John Arbash Meinel	15 years ago
3735.2.160	Fix a trivial typo Fix a trivial typo	John Arbash Meinel	15 years ago
3735.2.159	Remove an isinstance(..., tuple) assertion. Remove an isinstance(..., tuple) assertion. According to lsprof it was actually a bit expensive, and didn't help much anyway.	John Arbash Meinel	15 years ago
3735.2.158	Remove support for passing None for end in GroupCompressBloc... Remove support for passing None for end in GroupCompressBlock.extract. I decided the removal of the extra int in wire-bytes and indices was not a worthy trade-off versus the ability to _prepare_for_extract and cheaply filter bytes during fetch. And it makes the code simpler/easier to maintain. Also, add support for having a 'empty content' record, which has start=end=0. Support costs very little, and simplifies things. And now GroupCompressBlock.extract() just returns the bytes. It doesn't try to sha the content, nor does it return a GCBEntry. We weren't using it anyway. And it can save ~50 seconds of sha-ing all the content during 'bzr pack' of a launchpad branch.	John Arbash Meinel	15 years ago
3735.2.157	Merge the updates to the groupcompress DeltaIndex. Merge the updates to the groupcompress DeltaIndex.	John Arbash Meinel	15 years ago

Older »