5
The unit for compressed storage in bzr is a *revfile*, whose design
6
was suggested by Matt Mackall.
8
This document describes version 1 of the file, and has some notes on
9
what might be done in version 2.
15
Compressed storage is a tradeoff between several goals:
17
* Reasonably compact storage of long histories.
19
* Robustness and simplicity.
21
* Fast extraction of versions and addition of new versions (preferably
22
without rewriting the whole file, or reading the whole history.)
24
* Fast and precise annotations.
26
* Storage of files of at least a few hundred MB.
28
* Lossless in useful ways: we can extract a series of texts and write
29
them back out without losing any information.
35
revfiles store the history of a single logical file, which is
36
identified in bzr by its file-id. In this sense they are similar to
37
an RCS or CVS ``,v`` file or an SCCS sfile.
39
Each state of the file is called a *text*.
41
Renaming, adding and deleting this file is handled at a higher level
42
by the inventory system, and is outside the scope of the revfile. The
43
revfile name is typically based on the file id which is itself
44
typically based on the name the file had when it was first added. But
45
this is purely cosmetic.
47
For example a file now called ``frob.c`` may have the id
48
``frobber.c-12873`` because it was originally called
49
``frobber.c``. Its texts are kept in the revfile
50
``.bzr/revfiles/frobber.c-12873.revs``.
52
When the file is deleted from the inventory the revfile does not
53
change. It's just not used in reproducing trees from that point
56
The revfile does not record the date when the text was added, a commit
57
message, properties, or any other metadata. That is handled in the
58
higher-level revision history.
60
Inventories and other metadata files that vary from one version to the
61
next can themselves be stored in revfiles.
63
revfiles store files as simple byte streams, with no consideration of
64
translating character sets, line endings, or keywords. Those are also
65
handled at a higher level. However, the revfile may make use of
66
knowledge that a file is line-based in generating a diff.
68
(The Python builtin difflib is too slow when generating a purely
69
byte-by-byte delta so we always make a line-by-line diff; when this
70
is fixed it may be feasible to use line-by-line diffs for all
73
Files whose text does not change from one revision to the next are
74
stored as just a single text in the revfile. This can happen even if
75
the file was renamed or other properties were changed in the
78
The revfile is held on disk as two files: an *index* and a *data*
79
file. The index file is short and always read completely into memory;
80
the data file is much longer and only the relevant bits of it,
81
identified by the index file, need to be read.
83
This design is similar to that of Netscape `mail summary files`_, in
84
that there is a small index which can always be read into memory and
85
that quickly identifies where to look in the main file. They differ
86
in many other ways though, most particularly that the index is not
87
just a cache but holds precious data in its own right.
89
.. _`mail summary files`: http://www.jwz.org/doc/mailsum.html
91
This is meant to scale to hold 100,000 revisions of a single file, by
92
which time the index file will be ~4.8MB and a bit big to read
95
Some of the reserved fields could be used to implement a (semi?)
96
balanced tree indexed by SHA1 so we can much more efficiently find the
97
index associated with a particular hash. For 100,000 revs we would be
98
able to find it in about 17 random reads, which is not too bad. On
99
the other hand that would compromise the append-only indexing, and
100
100,000 revs is a fairly extreme case.
102
This performs pretty well except when trying to calculate deltas of
103
really large files. For that the main thing would be to plug in
104
something faster than difflib, which is after all pure Python.
105
Another approach is to just store the gzipped full text of big files,
106
though perhaps that's too perverse?
112
In the current version, texts are identified by their SHA-1.
118
Because the basis of a delta does not need to be the text's logical
119
predecessor, we can adjust the deltas to avoid ever needing to apply
120
too many deltas to reproduce a particular file.
126
The revfile module can be invoked as a program to give low-level
127
access for data recovery, debugging, etc.
135
* revfiles use unsigned 32-bit integers both in diffs and the index.
136
This should be more than enough for any reasonable source file but
137
perhaps not enough for large binaries that are frequently committed.
139
Perhaps for those files there should be an option to continue to use
140
the text-store. There is unlikely to be any benefit in holding
141
deltas between them, and deltas will anyhow be hard to calculate.
143
* The append-only design does not allow for destroying committed data,
144
as when confidential information is accidentally added. That could
145
be fixed by creating the fixed repository as a separate branch, into
146
which only the preserved revisions are exported.
148
* Should we also store full-texts as a transitional step?