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

~bzr-pqm/bzr/bzr.dev » Changes from revision 5365.5.29

Rev	Summary	Authors	Date
5365.5.29	Handle test_source and extensions. Also define an 'extern' p... Handle test_source and extensions. Also define an 'extern' protocol, to allow the test suite to recognize that returning an object of that type is a Python object.	John Arbash Meinel	14 years ago
5365.5.28	Lots of compatibility changes for python2.4 and mingw32 comp... Lots of compatibility changes for python2.4 and mingw32 compilers. gcc has strtoll available, and does the wrong thing if you use _strtoi64 (it implicitly defines it to return an 'int' which isn't a 64-bit integer.) Further, python2.4 doesn't support the %lu syntax, only having %ld and %d. So we go back to casting everything into real python objects and then stringifying them, but only for python 2.4.	John Arbash Meinel	14 years ago
5365.5.27	switch 'x += 1' to 'x = x + 1' to deal with brain-damaged ol... switch 'x += 1' to 'x = x + 1' to deal with brain-damaged old versions of pyrex.	John Arbash Meinel	14 years ago
5365.5.26	Update What's New and NEWS Update What's New and NEWS	John Arbash Meinel	14 years ago
5365.5.25	Merge bzr.dev 5387 in prep for NEWS Merge bzr.dev 5387 in prep for NEWS	John Arbash Meinel	14 years ago
5365.5.24	Change the _test names to _get and _py so that it doesn't pr... Change the _test names to _get and _py so that it doesn't provoke bad smells :)	John Arbash Meinel	14 years ago
5365.5.23	A __sizeof__ check that ensure we are getting what we are lo... A __sizeof__ check that ensure we are getting what we are looking for. It also checks that it grows appropriately for how many new entries are added.	John Arbash Meinel	14 years ago
5365.5.22	Pyrexification. Rip out all the stuff that only Cython or ne... Pyrexification. Rip out all the stuff that only Cython or new versions of pyrex understand.	John Arbash Meinel	14 years ago
5365.5.21	Remove counters, start the cleanup Remove counters, start the cleanup	John Arbash Meinel	14 years ago
5365.5.20	Add some tests that check the leaf factory is correct. Add some tests that check the leaf factory is correct. I'm not really happy about how this is getting exposed. I'll ask in review.	John Arbash Meinel	14 years ago
5365.5.19	Lots of instrumentation. Add a last-lookup cache. Lots of instrumentation. Add a last-lookup cache. The core loop in the btree code is: if x in node: val = node[x] As such, we avoid all the processing if we just cache when we had a __contains__ hit. For ls -R we actually get a 100% __getitem__ hit rate.	John Arbash Meinel	14 years ago
5365.5.18	Expose the new leaf node factory across the stack. Expose the new leaf node factory across the stack. This means creating something we can get to at the btree level and using it for the gcchk code.	John Arbash Meinel	14 years ago
5365.5.17	rename 'entries' to 'records', some comment cleanups. rename 'entries' to 'records', some comment cleanups.	John Arbash Meinel	14 years ago
5365.5.16	Trim the class a little bit. Trim the class a little bit. All we need is the common_shift, common_bits and common_mask aren't used. This puts as at about 42.? bytes/record. We have 40 actual, and then <3 bytes/record of other overhead.	John Arbash Meinel	14 years ago
5365.5.15	Handle the case where there are more than 255 entries. Handle the case where there are more than 255 entries. This is very unlikely in real-world, but we want to preserve correctness rather than failing. There doesn't seem to be a performance impact (still 20us), which is fairly expected. It is a single unlikely if check in the lookup logic.	John Arbash Meinel	14 years ago
5365.5.14	We should be exposing StaticTuple_FromSequence in its api de... We should be exposing StaticTuple_FromSequence in its api definition	John Arbash Meinel	14 years ago
5365.5.13	Populate the offsets array. Populate the offsets array. This cuts down the number of bisections dramatically, basically by pre-caching the first step. On real-world data it drops the steps from 587 to 156. Or from 4.9/key to 1.3/key. This drops the time to lookup from 23.7us to 20.3us. Note that (k in dict) is 12.2us. I do wish we were just a bit closer to that. However, with _LeafNode inherited from dict, I get 26us, so maybe there is something in the interpreter that does a PyDict_CheckExact call, and there isn't much we can do about it.	John Arbash Meinel	14 years ago
5365.5.12	Make _LeafNode inherit from dict (is-a rather than have-a) Make _LeafNode inherit from dict (is-a rather than have-a) It avoids having another object layer, and indirection via attribute and forwarding functions. TIMEIT shows: 14.6us raw dict 26.6us subclass 57.9us wrapper class	John Arbash Meinel	14 years ago
5365.5.11	get into the nitty gritty for the _key_to_sha1 function. get into the nitty gritty for the _key_to_sha1 function. It doesn't seem to help to op-out of the actual unhexlify call, so at least that doesn't seem to be the performance overhead. We get down to around 19.9us for _key_to_sha1 across 120 keys, which is 0.166us per key.	John Arbash Meinel	14 years ago
5365.5.10	Do some timing with different objects. Do some timing with different objects. It turns out that pure dict is 13.7us lookup time, dict-as-attr is 20.7us, PyObject __getitem__ to dict-as-attr is 54us.	John Arbash Meinel	14 years ago

Older »