9 The diff commands git-diff-index, git-diff-files, and
10 git-diff-tree can be told to manipulate differences they find
11 in unconventional ways before showing diff(1) output. The
12 manipulation is collectively called "diffcore transformation".
13 This short note describes what they are and how to use them to
14 produce diff outputs that are easier to understand than the
18 The chain of operation
19 ----------------------
21 The git-diff-* family works by first comparing two sets of
24 - git-diff-index compares contents of a "tree" object and the
25 working directory (when '--cached' flag is not used) or a
26 "tree" object and the index file (when '--cached' flag is
29 - git-diff-files compares contents of the index file and the
32 - git-diff-tree compares contents of two "tree" objects.
34 In all of these cases, the commands themselves compare
35 corresponding paths in the two sets of files. The result of
36 comparison is passed from these commands to what is internally
37 called "diffcore", in a format similar to what is output when
38 the -p option is not used. E.g.
40 ------------------------------------------------
41 in-place edit :100644 100644 bcd1234... 0123456... M file0
42 create :000000 100644 0000000... 1234567... A file4
43 delete :100644 000000 1234567... 0000000... D file5
44 unmerged :000000 000000 0000000... 0000000... U file6
45 ------------------------------------------------
47 The diffcore mechanism is fed a list of such comparison results
48 (each of which is called "filepair", although at this point each
49 of them talks about a single file), and transforms such a list
50 into another list. There are currently 6 such transformations:
55 - diffcore-merge-broken
59 These are applied in sequence. The set of filepairs git-diff-\*
60 commands find are used as the input to diffcore-pathspec, and
61 the output from diffcore-pathspec is used as the input to the
62 next transformation. The final result is then passed to the
63 output routine and generates either diff-raw format (see Output
64 format sections of the manual for git-diff-\* commands) or
71 The first transformation in the chain is diffcore-pathspec, and
72 is controlled by giving the pathname parameters to the
73 git-diff-* commands on the command line. The pathspec is used
74 to limit the world diff operates in. It removes the filepairs
75 outside the specified set of pathnames.
77 Implementation note. For performance reasons, git-diff-tree
78 uses the pathname parameters on the command line to cull set of
79 filepairs it feeds the diffcore mechanism itself, and does not
80 use diffcore-pathspec, but the end result is the same.
86 The second transformation in the chain is diffcore-break, and is
87 controlled by the -B option to the git-diff-* commands. This is
88 used to detect a filepair that represents "complete rewrite" and
89 break such filepair into two filepairs that represent delete and
90 create. E.g. If the input contained this filepair:
92 ------------------------------------------------
93 :100644 100644 bcd1234... 0123456... M file0
94 ------------------------------------------------
96 and if it detects that the file "file0" is completely rewritten,
99 ------------------------------------------------
100 :100644 000000 bcd1234... 0000000... D file0
101 :000000 100644 0000000... 0123456... A file0
102 ------------------------------------------------
104 For the purpose of breaking a filepair, diffcore-break examines
105 the extent of changes between the contents of the files before
106 and after modification (i.e. the contents that have "bcd1234..."
107 and "0123456..." as their SHA1 content ID, in the above
108 example). The amount of deletion of original contents and
109 insertion of new material are added together, and if it exceeds
110 the "break score", the filepair is broken into two. The break
111 score defaults to 50% of the size of the smaller of the original
112 and the result (i.e. if the edit shrinks the file, the size of
113 the result is used; if the edit lengthens the file, the size of
114 the original is used), and can be customized by giving a number
115 after "-B" option (e.g. "-B75" to tell it to use 75%).
121 This transformation is used to detect renames and copies, and is
122 controlled by the -M option (to detect renames) and the -C option
123 (to detect copies as well) to the git-diff-* commands. If the
124 input contained these filepairs:
126 ------------------------------------------------
127 :100644 000000 0123456... 0000000... D fileX
128 :000000 100644 0000000... 0123456... A file0
129 ------------------------------------------------
131 and the contents of the deleted file fileX is similar enough to
132 the contents of the created file file0, then rename detection
133 merges these filepairs and creates:
135 ------------------------------------------------
136 :100644 100644 0123456... 0123456... R100 fileX file0
137 ------------------------------------------------
139 When the "-C" option is used, the original contents of modified
140 files and contents of unchanged files are considered as
141 candidates of the source files in rename/copy operation, in
142 addition to the deleted files. If the input were like these
143 filepairs, that talk about a modified file fileY and a newly
146 ------------------------------------------------
147 :100644 100644 0123456... 1234567... M fileY
148 :000000 100644 0000000... 0123456... A file0
149 ------------------------------------------------
151 the original contents of fileY and the resulting contents of
152 file0 are compared, and if they are similar enough, they are
155 ------------------------------------------------
156 :100644 100644 0123456... 1234567... M fileY
157 :100644 100644 0123456... 0123456... C100 fileY file0
158 ------------------------------------------------
160 In both rename and copy detection, the same "extent of changes"
161 algorithm used in diffcore-break is used to determine if two
162 files are "similar enough", and can be customized to use
163 similarity score different from the default 50% by giving a
164 number after "-M" or "-C" option (e.g. "-M8" to tell it to use
167 Note. When the "-C" option is used with --find-copies-harder
168 option, git-diff-\* commands feed unmodified filepairs to
169 diffcore mechanism as well as modified ones. This lets the copy
170 detector consider unmodified files as copy source candidates at
171 the expense of making it slower. Without --find-copies-harder,
172 git-diff-\* commands can detect copies only if the file that was
173 copied happened to have been modified in the same changeset.
176 diffcore-merge-broken
177 ---------------------
179 This transformation is used to merge filepairs broken by
180 diffcore-break, and were not transformed into rename/copy by
181 diffcore-rename, back into a single modification. This always
182 runs when diffcore-break is used.
184 For the purpose of merging broken filepairs back, it uses a
185 different "extent of changes" computation from the ones used by
186 diffcore-break and diffcore-rename. It counts only the deletion
187 from the original, and does not count insertion. If you removed
188 only 10 lines from a 100-line document, even if you added 910
189 new lines to make a new 1000-line document, you did not do a
190 complete rewrite. diffcore-break breaks such a case in order to
191 help diffcore-rename to consider such filepairs as candidate of
192 rename/copy detection, but if filepairs broken that way were not
193 matched with other filepairs to create rename/copy, then this
194 transformation merges them back into the original
197 The "extent of changes" parameter can be tweaked from the
198 default 80% (that is, unless more than 80% of the original
199 material is deleted, the broken pairs are merged back into a
200 single modification) by giving a second number to -B option,
203 * -B50/60 (give 50% "break score" to diffcore-break, use 60%
204 for diffcore-merge-broken).
206 * -B/60 (the same as above, since diffcore-break defaults to 50%).
208 Note that earlier implementation left a broken pair as a separate
209 creation and deletion patches. This was unnecessary hack and
210 the latest implementation always merges all the broken pairs
211 back into modifications, but the resulting patch output is
212 formatted differently to still let the reviewing easier for such
213 a complete rewrite by showing the entire contents of old version
214 prefixed with '-', followed by the entire contents of new
215 version prefixed with '+'.
221 This transformation is used to find filepairs that represent
222 changes that touch a specified string, and is controlled by the
223 -S option and the --pickaxe-all option to the git-diff-*
226 When diffcore-pickaxe is in use, it checks if there are
227 filepairs whose "original" side has the specified string and
228 whose "result" side does not. Such a filepair represents "the
229 string appeared in this changeset". It also checks for the
230 opposite case that loses the specified string.
232 When --pickaxe-all is not in effect, diffcore-pickaxe leaves
233 only such filepairs that touches the specified string in its
234 output. When --pickaxe-all is used, diffcore-pickaxe leaves all
235 filepairs intact if there is such a filepair, or makes the
236 output empty otherwise. The latter behaviour is designed to
237 make reviewing of the changes in the context of the whole
244 This is used to reorder the filepairs according to the user's
245 (or project's) taste, and is controlled by the -O option to the
248 This takes a text file each of whose line is a shell glob
249 pattern. Filepairs that match a glob pattern on an earlier line
250 in the file are output before ones that match a later line, and
251 filepairs that do not match any glob pattern are output last.
253 As an example, typical orderfile for the core GIT probably
254 would look like this:
256 ------------------------------------------------
263 ------------------------------------------------