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

~bzr-pqm/bzr/bzr.dev » Changes from revision 4679.9.19

Rev	Summary	Authors	Date
4679.9.19	Interning the start and stop group positions saves another 7... Interning the start and stop group positions saves another 7MB peak mem. \o/ I'm hoping some of this savings is the 'dark memory', but that is unclear at this point.	John Arbash Meinel	15 years ago
4679.9.18	Revert the ugly stuff, and do it properly, one step at a tim... Revert the ugly stuff, and do it properly, one step at a time. In _filter_text_keys, we were extracting (file_id, revision_id) from the CHKMap value section. However, we were leaving them as regular strings, and a regular tuple. Using StaticTuple(intern(file_id), intern(revision_id)).intern() results in a 40MB decrease in peak memory during 'stream test'.	John Arbash Meinel	15 years ago
4679.9.17	Checkpoint.... I think I've cut off another 50MB or so peak ... Checkpoint.... I think I've cut off another 50MB or so peak mem. But I've got a bunch of cruft in this patch. So I'm checkpointing so that I can get back to this point, and then reverting a bunch of the changes.	John Arbash Meinel	15 years ago
4679.9.16	More cleanups and clarifications. More cleanups and clarifications.	John Arbash Meinel	15 years ago
4679.9.15	Cleanup some code paths. Make _check_key a helper that can b... Cleanup some code paths. Make _check_key a helper that can be used but isn't currently used.	John Arbash Meinel	15 years ago
4679.9.14	Revert previous change. Revert previous change.	John Arbash Meinel	15 years ago
4679.9.13	Try out interning of some of the strings. Try out interning of some of the strings. Doesn't seem to have a serious impact. If we only have a 2-deep tree (1 internal, 1 leaf), all the internal nodes are 1-byte wide anyway, so they are already unique (single chars). Interning the values is a slight net loss. A bit suprising. At an average of 21 values per node, (and an average of 1 change per node) I would have thought we would have more overlap in values. I'll probably revert this as the meager to nil benefit doesn't outweigh the complexity.	John Arbash Meinel	15 years ago
4679.9.12	Tweak some of the internals of _chk_map_pyx.pyx Tweak some of the internals of _chk_map_pyx.pyx Since we have a fixed type, we can use the StaticTuple_GET_ITEM macro.	John Arbash Meinel	15 years ago
4679.9.11	found one more found one more	John Arbash Meinel	15 years ago
4679.9.10	Change the testing layer so that CHKMap is tested using tupl... Change the testing layer so that CHKMap is tested using tuples. For now, it makes life easier, so go ahead and do it. Potentially we want to have a debug flag that indicates when tuples are being used. Because passing in a tuple wastes memory, if we then convert it to a StaticTuple. The main reason to be type strict is because we want to use StaticTuple as much internally as we can, and it eases some of the extension writing, since we don't have to handle packing multiple object types, etc.	John Arbash Meinel	15 years ago
4679.9.9	Create a Barrier at the CHKMap interface. Create a Barrier at the CHKMap interface. Basically, everything above CHKMap can work in either tuples or StaticTuples. Everything below CHKMap should be working in StaticTuples. I need to fix .iteritems() but apply_delta, map, unmap, from_dict, etc are all handled.	John Arbash Meinel	15 years ago
4679.9.8	Track down a bunch of places in the 'inventory' code that wa... Track down a bunch of places in the 'inventory' code that was passing simple tuples into the CHKMap layer. Note that almost all of these are cases where we started with a file_id and then want to translate that into the CHKMap lookup key. As such, these aren't places where we would expect to already have a tuple. Since the Inventory api is in raw strings and not keys. Anyway, forward progress as this fixes a lot of failures.	John Arbash Meinel	15 years ago
4679.9.7	Bring in bzr.dev 4761 which includes CHKMap and CHKInventory... Bring in bzr.dev 4761 which includes CHKMap and CHKInventory tweaks. It also brings in StaticTuple concatenation, and ability to hold None, etc.	John Arbash Meinel	15 years ago
4679.9.6	Remove the early check, since the test suite passes an Inter... Remove the early check, since the test suite passes an InternalNode rather than a key	John Arbash Meinel	15 years ago
4679.9.5	we don't actually need to use an ST here, it is only a local... we don't actually need to use an ST here, it is only a local check.	John Arbash Meinel	15 years ago
4679.9.4	A bit broken, but getting there. A bit broken, but getting there. Start being much stricter about requiring StaticTuples everywhere. I may go back and loosen this restriction, but getting the code base StaticTuple pure is probably a good idea. The main reason to be 'looser' is so that things don't fail 'in the wild' just because someone calls an api with a tuple rather than a StaticTuple. However, I'd like the internals to be 'pure' if possible. We'll see.	John Arbash Meinel	15 years ago
4679.9.3	Merge bzr.dev 4759, bringing in static_tuple and streaming Merge bzr.dev 4759, bringing in static_tuple and streaming improvements.	John Arbash Meinel	15 years ago
4679.9.2	Clean up the chk test suite a little bit. Clean up the chk test suite a little bit.	John Arbash Meinel	15 years ago
4679.9.1	Merge in the static-tuple-no-use branch, and bring back the ... Merge in the static-tuple-no-use branch, and bring back the chk_map use.	John Arbash Meinel	15 years ago
4679.5.1	Interning the most common 'value' entries saves about 1-2% o... Interning the most common 'value' entries saves about 1-2% of peak memory. Namely, interning the '0' content values. Note that most values have a common prefix now, and an uncommon suffix...	John Arbash Meinel	15 years ago

Older »