range-diff: populate the man page

The bulk of this patch consists of a heavily butchered version of tbdiff's README written by Thomas Rast and Thomas Gummerer, lifted from https://github.com/trast/tbdiff. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>

Johannes Schindelin committed Aug 13, 2018 at 04:33 UTC ba931edd284f705d0c4b9af1f50f74a784fc6e34
1 file changed +229
Documentation/git-range-diff.txt
+229
@@ -5,6 +5,235 @@ NAME
5 ----
6 git-range-diff - Compare two commit ranges (e.g. two versions of a branch)
7
8 +SYNOPSIS
9 +--------
10 +[verse]
11 +'git range-diff' [--color=[<when>]] [--no-color] [<diff-options>]
12 + [--dual-color] [--creation-factor=<factor>]
13 + ( <range1> <range2> | <rev1>...<rev2> | <base> <rev1> <rev2> )
14 +
15 +DESCRIPTION
16 +-----------
17 +
18 +This command shows the differences between two versions of a patch
19 +series, or more generally, two commit ranges (ignoring merge commits).
20 +
21 +To that end, it first finds pairs of commits from both commit ranges
22 +that correspond with each other. Two commits are said to correspond when
23 +the diff between their patches (i.e. the author information, the commit
24 +message and the commit diff) is reasonably small compared to the
25 +patches' size. See ``Algorithm`` below for details.
26 +
27 +Finally, the list of matching commits is shown in the order of the
28 +second commit range, with unmatched commits being inserted just after
29 +all of their ancestors have been shown.
30 +
31 +
32 +OPTIONS
33 +-------
34 +--dual-color::
35 + When the commit diffs differ, recreate the original diffs'
36 + coloring, and add outer -/+ diff markers with the *background*
37 + being red/green to make it easier to see e.g. when there was a
38 + change in what exact lines were added.
39 +
40 +--creation-factor=<percent>::
41 + Set the creation/deletion cost fudge factor to `<percent>`.
42 + Defaults to 60. Try a larger value if `git range-diff` erroneously
43 + considers a large change a total rewrite (deletion of one commit
44 + and addition of another), and a smaller one in the reverse case.
45 + See the ``Algorithm`` section below for an explanation why this is
46 + needed.
47 +
48 +<range1> <range2>::
49 + Compare the commits specified by the two ranges, where
50 + `<range1>` is considered an older version of `<range2>`.
51 +
52 +<rev1>...<rev2>::
53 + Equivalent to passing `<rev2>..<rev1>` and `<rev1>..<rev2>`.
54 +
55 +<base> <rev1> <rev2>::
56 + Equivalent to passing `<base>..<rev1>` and `<base>..<rev2>`.
57 + Note that `<base>` does not need to be the exact branch point
58 + of the branches. Example: after rebasing a branch `my-topic`,
59 + `git range-diff my-topic@{u} my-topic@{1} my-topic` would
60 + show the differences introduced by the rebase.
61 +
62 +`git range-diff` also accepts the regular diff options (see
63 +linkgit:git-diff[1]), most notably the `--color=[<when>]` and
64 +`--no-color` options. These options are used when generating the "diff
65 +between patches", i.e. to compare the author, commit message and diff of
66 +corresponding old/new commits. There is currently no means to tweak the
67 +diff options passed to `git log` when generating those patches.
68 +
69 +
70 +CONFIGURATION
71 +-------------
72 +This command uses the `diff.color.*` and `pager.range-diff` settings
73 +(the latter is on by default).
74 +See linkgit:git-config[1].
75 +
76 +
77 +EXAMPLES
78 +--------
79 +
80 +When a rebase required merge conflicts to be resolved, compare the changes
81 +introduced by the rebase directly afterwards using:
82 +
83 +------------
84 +$ git range-diff @{u} @{1} @
85 +------------
86 +
87 +
88 +A typical output of `git range-diff` would look like this:
89 +
90 +------------
91 +-: ------- > 1: 0ddba11 Prepare for the inevitable!
92 +1: c0debee = 2: cab005e Add a helpful message at the start
93 +2: f00dbal ! 3: decafe1 Describe a bug
94 + @@ -1,3 +1,3 @@
95 + Author: A U Thor <author@example.com>
96 +
97 + -TODO: Describe a bug
98 + +Describe a bug
99 + @@ -324,5 +324,6
100 + This is expected.
101 +
102 + -+What is unexpected is that it will also crash.
103 + ++Unexpectedly, it also crashes. This is a bug, and the jury is
104 + ++still out there how to fix it best. See ticket #314 for details.
105 +
106 + Contact
107 +3: bedead < -: ------- TO-UNDO
108 +------------
109 +
110 +In this example, there are 3 old and 3 new commits, where the developer
111 +removed the 3rd, added a new one before the first two, and modified the
112 +commit message of the 2nd commit as well its diff.
113 +
114 +When the output goes to a terminal, it is color-coded by default, just
115 +like regular `git diff`'s output. In addition, the first line (adding a
116 +commit) is green, the last line (deleting a commit) is red, the second
117 +line (with a perfect match) is yellow like the commit header of `git
118 +show`'s output, and the third line colors the old commit red, the new
119 +one green and the rest like `git show`'s commit header.
120 +
121 +The color-coded diff is actually a bit hard to read, though, as it
122 +colors the entire lines red or green. The line that added "What is
123 +unexpected" in the old commit, for example, is completely red, even if
124 +the intent of the old commit was to add something.
125 +
126 +To help with that, use the `--dual-color` mode. In this mode, the diff
127 +of diffs will retain the original diff colors, and prefix the lines with
128 +-/+ markers that have their *background* red or green, to make it more
129 +obvious that they describe how the diff itself changed.
130 +
131 +
132 +Algorithm
133 +---------
134 +
135 +The general idea is this: we generate a cost matrix between the commits
136 +in both commit ranges, then solve the least-cost assignment.
137 +
138 +The cost matrix is populated thusly: for each pair of commits, both
139 +diffs are generated and the "diff of diffs" is generated, with 3 context
140 +lines, then the number of lines in that diff is used as cost.
141 +
142 +To avoid false positives (e.g. when a patch has been removed, and an
143 +unrelated patch has been added between two iterations of the same patch
144 +series), the cost matrix is extended to allow for that, by adding
145 +fixed-cost entries for wholesale deletes/adds.
146 +
147 +Example: Let commits `1--2` be the first iteration of a patch series and
148 +`A--C` the second iteration. Let's assume that `A` is a cherry-pick of
149 +`2,` and `C` is a cherry-pick of `1` but with a small modification (say,
150 +a fixed typo). Visualize the commits as a bipartite graph:
151 +
152 +------------
153 + 1 A
154 +
155 + 2 B
156 +
157 + C
158 +------------
159 +
160 +We are looking for a "best" explanation of the new series in terms of
161 +the old one. We can represent an "explanation" as an edge in the graph:
162 +
163 +
164 +------------
165 + 1 A
166 + /
167 + 2 --------' B
168 +
169 + C
170 +------------
171 +
172 +This explanation comes for "free" because there was no change. Similarly
173 +`C` could be explained using `1`, but that comes at some cost c>0
174 +because of the modification:
175 +
176 +------------
177 + 1 ----. A
178 + | /
179 + 2 ----+---' B
180 + |
181 + `----- C
182 + c>0
183 +------------
184 +
185 +In mathematical terms, what we are looking for is some sort of a minimum
186 +cost bipartite matching; `1` is matched to `C` at some cost, etc. The
187 +underlying graph is in fact a complete bipartite graph; the cost we
188 +associate with every edge is the size of the diff between the two
189 +commits' patches. To explain also new commits, we introduce dummy nodes
190 +on both sides:
191 +
192 +------------
193 + 1 ----. A
194 + | /
195 + 2 ----+---' B
196 + |
197 + o `----- C
198 + c>0
199 + o o
200 +
201 + o o
202 +------------
203 +
204 +The cost of an edge `o--C` is the size of `C`'s diff, modified by a
205 +fudge factor that should be smaller than 100%. The cost of an edge
206 +`o--o` is free. The fudge factor is necessary because even if `1` and
207 +`C` have nothing in common, they may still share a few empty lines and
208 +such, possibly making the assignment `1--C`, `o--o` slightly cheaper
209 +than `1--o`, `o--C` even if `1` and `C` have nothing in common. With the
210 +fudge factor we require a much larger common part to consider patches as
211 +corresponding.
212 +
213 +The overall time needed to compute this algorithm is the time needed to
214 +compute n+m commit diffs and then n*m diffs of patches, plus the time
215 +needed to compute the least-cost assigment between n and m diffs. Git
216 +uses an implementation of the Jonker-Volgenant algorithm to solve the
217 +assignment problem, which has cubic runtime complexity. The matching
218 +found in this case will look like this:
219 +
220 +------------
221 + 1 ----. A
222 + | /
223 + 2 ----+---' B
224 + .--+-----'
225 + o -' `----- C
226 + c>0
227 + o ---------- o
228 +
229 + o ---------- o
230 +------------
231 +
232 +
233 +SEE ALSO
234 +--------
235 +linkgit:git-log[1]
236 +
237 GIT
238 ---
239 Part of the linkgit:git[1] suite