1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
|
Todo for groupcompress:
* layers - gc reader/writer and vf layers should be separate
* alternate copy/insert encoding?
* other stuff from design
* Look at splitting the 'labels' section into a header/index for all
compressed entries, separate from the actual content. This could potentially
improve compression efficiency.
* Network stream form. Possibly just the gc group. Possibly a cluster of gc
groups with bzip2 over them. Possibly a gc group but with fulltexts that the
target has stripped out and replaced with an 'external' record.
* Consider a more 'binary' layout for things like sha1 sums, or length
indicators.
* Investigate sha1 overhead. Currently when doing 'bzr pack' we sha once on
extraction, and then one more time on insertion. Removing both calls saves
6s out of 32s for 'bzr pack' (and 8s=>5.5s for 'repository-details')
|