[PATCH] KnitIndex optimizations

[Dmitry Vasiliev]

Brief overview of the changes in the attached patch:

- Extracted the knit index file parsing code from __init__() into _load_data() 
method. For the new code profiling tells about 15% speedup while parse all 
*.kndx from the Bazaar branch.
- Converted loops into list comprehensions where appropriate.
- Localized attributes for faster access inside loops where appropriate.
- Added forgotten import.
- Removed trailing whitespaces.

Open questions:

- KnitVersionedFile._check_versions_present() method now just calls 
self._index.check_versions_present(). The possible problem here is that 
RevisionNotPresent exception now gets filename of the index instead of the base 
filename of the knit. I'm not sure should the exception be caught at the 
KnitVersionedFile level and be rerised with the knit's base filename?
- Maybe _load_data() should use file's iteration protocol instead of the 
readlines() method for parse index file? A file iterator use read-ahead buffer 
so it should be faster than file.readline() and should consume less memory (for 
big files) while be a little bit slower than file.readlines(). For the 
_load_data() method profiling tells about 10% speedup for 'for line in fp' 
instead of 15% for 'for line in fp.readlines()'.

-- 
Dmitry Vasiliev (dima at hlabs.spb.ru)
      http://hlabs.spb.ru

-------------- next part --------------
A non-text attachment was scrubbed...
Name: knit_index.patch
Type: text/x-patch
Size: 15047 bytes
Desc: not available
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20061120/e986609b/attachment.bin