4
At the moment bzr just stores the full text of every file state ever.
5
(Files unmodified from one revision are stored only once.)
7
This is simple to write, and adequate for even reasonably large trees
8
with many versions. Disk is very cheap.
10
Eventually we might like something more compressed, but this is
11
neither an interesting nor urgent problem. (Not "interesting" in the
12
sense that doing it is just a matter of coding; there is no
13
theoretical problem or risk.)
15
There are various possibilities:
17
* Store history of each file in RCS, relying on RCS to do line-by-line
18
delta compression. (Does not handle binaries very well.)
20
OpenCMS paper has a horror story about using RCS for file storage.
22
* Store full copies of each file in a container with gzip compression,
23
which should fairly efficiently eliminate unchanged areas. This
24
works on binaries, and gives compression of file text as a side
27
(Note that ``.zip`` will *not* do for this, because every file is
28
compressed independently.)
30
The OpenCMS paper notes that RCS storage is only 20% more efficient
31
than gzip'd storage of individual file versions.
33
* Store history of each file in SCCS; allows quick retrieval of any
34
previous state and may give more efficient storage than RCS. Allows
35
for divergent branches within a single file.
37
* Store xdeltas between consecutive file states.
39
* Store xdeltas according to a spanning delta algorithm; this probably
40
requires files are stored with some kind of sequence number so that
41
we can predict related version names.
43
* Store in something like XDFS.
45
* Any of the above, but with the final storage in some kind of
46
database: psql, sqlite, mysql, whatever is convenient.
48
It should be something that is safe across system crashes, which
49
rules out tdb at the moment.
53
These properties are seen as desirable in darcs and arch:
55
* Passive HTTP downloads: without requiring any server-side
56
intelligence, a client can get updates from one version to the next
57
by requesting a set of self-contained files.
59
The number of files necessary to do this must not be unfeasibly
60
large, and the size of each of those files should be proportionate
61
to the amount of actual change. In other words the data downloaded
62
should be of comparable size to the actual edits between the trees.
65
* Write-once storage: once a file is written to the repository, it is