1
diff Performance Analysis
2
=========================
9
Reuse of historical comparisons
10
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
11
A significant part of the work done by diff is sequence matching. This
12
scales O(n^2) with the number of lines in the file. Therefore, it
13
is worthwile to avoid content comparisons as much as possible.
15
Our current knit format contains content comparisons, and this data can
16
be converted into lists of matching blocks. Other future formats such as
17
mpdiff may also support such conversion. So it is possible to reuse past
20
It is also possible to combine sequential comparisons. So given a comparison
21
of "foo" to "bar", and "bar" to "baz", it is possible to derive a comparison of
24
Reuse of historical comparisons will scale with the number of uncommon
25
build-parents between the two historical revisions. This will typically be
26
proportional to the amount of change that the file has undergone. Therefore,
27
in the common case, reuse of historical comparisons will scale with the
30
The downside of such reuse is that it ties the comparison to the historical
31
data. But given the performance improvement, it seems to be worth
32
consideration. Fresh comparisons can be performed if the user requests them.
34
It may also be possible to accelerate comparisons by including annotation data,
35
thus increasing the number of unique lines.
37
Historical Tree Against Historical Tree
38
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
39
This operation should be strictly proportional to the amount of change, because
40
a comparison has already been done at commit time. Achieving that performance
41
requires the committed data to be properly structured, so that the comparison
42
can be extracted and combined with other comparisons. This comparision
43
extraction should be possible at the inventory and file-content levels.
47
1. Extract and combine inventory comparisons
48
2. Extract and combine text comparisions for modified texts
50
Basis Against Historical Tree
51
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
52
This is another case of Historical Tree Against Historical Tree.
56
This is another case of Historical Tree Against Historical Tree.
58
Working Tree Against Basis
59
~~~~~~~~~~~~~~~~~~~~~~~~~~
60
This must scale with the number of versioned files, unless the user indicates
61
that only certain files should be compared.
63
Performance can be further improved by caching comparisons to avoid repeating
64
them. Caching could potentially be performed by ``diff`` and perhaps by
65
``merge``. Merge is aware of the relationship of a text merge's result to
66
the THIS value, and the THIS value is generally the basis value. So the
67
comparison is latent, but present. The only issue is extracting it.
69
The cache could be indexed by sha1sum pairs. It could also be indexed by
70
file-id, to facilitate removal of stale data.
74
1. Scan working tree for modified files
75
2. Retrieve cached comparisons
76
3. Perform comparisons on files with no cached comparisons
77
4. Cache comparisons for files with no cached comparisons
79
Working Tree Against Historical Tree
80
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
81
This can be structured as a comparison of working tree against basis tree,
82
followed by basis tree against historical tree. Therefore, it combines the
83
performance characteristics of "Working Tree Against Basis" with "Basis Against
86
Working Tree Against Working Tree
87
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
88
This can be structured as two comparisons against basis, and one comparison
89
of basis against basis. Its performance is therefore similar to Working Tree
90
Against Historical Tree.
97
- Tree.get_comparision(file_id, tree)
99
This probably entails:
101
- WorkingTree.store_comparison(file_id, revision_id, sha1, comparison)
102
- WorkingTree.get_comparison(file_id, revision_id, sha1)
103
- Repository.get_comparision(file_id, revision_id, revision_id)
104
- merge_comparisions(comparison, comparision)
106
Storage considerations
107
----------------------
108
It must be cheap (e.g. scale with number of intermediate revisions) to perform
109
comparison of two historical texts. It must be cheap to perform comparison of
110
the inventories of two historical trees.
added
removed
doc/developers/diff.txt
