Philippe Verdy
2004-11-25 17:00:46 UTC
The previous message contains the complete version of this file.
This is a CVS patch version (roughly the same size)...
A few things I'd like to see is to allow this class to reuse its working
byte[] data buffer, instead of allocating a new one each time patches are
updated (notably for the generation of UP-to-UP patches).
Note that another optimization it performs also include a faster check of
bad patch sizes (it is checked early only once instead of for each patch
value within a patch fragment).
This class is still the one that uses the most the CPU when running as an
UltraPeer in the background. I wonder if using a BitSet is a good idea (we
could use the working byte[] buffer directly, if we decide later to activate
the generation of 1-bit patches; however we'll need to support for some time
the generation of 4-bit patches with all legacy limewires and other
servents.)
For a next I wonder which solution will be good: activating the 2-bit
format, or directly the 1-bit format?
(This code supports both on reception).
This is a CVS patch version (roughly the same size)...
A few things I'd like to see is to allow this class to reuse its working
byte[] data buffer, instead of allocating a new one each time patches are
updated (notably for the generation of UP-to-UP patches).
Note that another optimization it performs also include a faster check of
bad patch sizes (it is checked early only once instead of for each patch
value within a patch fragment).
This class is still the one that uses the most the CPU when running as an
UltraPeer in the background. I wonder if using a BitSet is a good idea (we
could use the working byte[] buffer directly, if we decide later to activate
the generation of 1-bit patches; however we'll need to support for some time
the generation of 4-bit patches with all legacy limewires and other
servents.)
For a next I wonder which solution will be good: activating the 2-bit
format, or directly the 1-bit format?
(This code supports both on reception).