1Tweaking diff output 2==================== 3June 2005 4 5 6Introduction 7------------ 8 9The diff commands git-diff-index, git-diff-files, and git-diff-tree 10can be told to manipulate differences they find in 11unconventional ways before showing diff(1) output. The manipulation 12is collectively called "diffcore transformation". This short note 13describes what they are and how to use them to produce diff outputs 14that are easier to understand than the conventional kind. 15 16 17The chain of operation 18---------------------- 19 20The git-diff-* family works by first comparing two sets of 21files: 22 23 - git-diff-index compares contents of a "tree" object and the 24 working directory (when '\--cached' flag is not used) or a 25 "tree" object and the index file (when '\--cached' flag is 26 used); 27 28 - git-diff-files compares contents of the index file and the 29 working directory; 30 31 - git-diff-tree compares contents of two "tree" objects; 32 33In all of these cases, the commands themselves compare 34corresponding paths in the two sets of files. The result of 35comparison is passed from these commands to what is internally 36called "diffcore", in a format similar to what is output when 37the -p option is not used. E.g. 38 39------------------------------------------------ 40in-place edit :100644 100644 bcd1234... 0123456... M file0 41create :000000 100644 0000000... 1234567... A file4 42delete :100644 000000 1234567... 0000000... D file5 43unmerged :000000 000000 0000000... 0000000... U file6 44------------------------------------------------ 45 46The diffcore mechanism is fed a list of such comparison results 47(each of which is called "filepair", although at this point each 48of them talks about a single file), and transforms such a list 49into another list. There are currently 6 such transformations: 50 51- diffcore-pathspec 52- diffcore-break 53- diffcore-rename 54- diffcore-merge-broken 55- diffcore-pickaxe 56- diffcore-order 57 58These are applied in sequence. The set of filepairs git-diff-\* 59commands find are used as the input to diffcore-pathspec, and 60the output from diffcore-pathspec is used as the input to the 61next transformation. The final result is then passed to the 62output routine and generates either diff-raw format (see Output 63format sections of the manual for git-diff-\* commands) or 64diff-patch format. 65 66 67diffcore-pathspec: For Ignoring Files Outside Our Consideration 68--------------------------------------------------------------- 69 70The first transformation in the chain is diffcore-pathspec, and 71is controlled by giving the pathname parameters to the 72git-diff-* commands on the command line. The pathspec is used 73to limit the world diff operates in. It removes the filepairs 74outside the specified set of pathnames. E.g. If the input set 75of filepairs included: 76 77------------------------------------------------ 78:100644 100644 bcd1234... 0123456... M junkfile 79------------------------------------------------ 80 81but the command invocation was "git-diff-files myfile", then the 82junkfile entry would be removed from the list because only "myfile" 83is under consideration. 84 85Implementation note. For performance reasons, git-diff-tree 86uses the pathname parameters on the command line to cull set of 87filepairs it feeds the diffcore mechanism itself, and does not 88use diffcore-pathspec, but the end result is the same. 89 90 91diffcore-break: For Splitting Up "Complete Rewrites" 92---------------------------------------------------- 93 94The second transformation in the chain is diffcore-break, and is 95controlled by the -B option to the git-diff-* commands. This is 96used to detect a filepair that represents "complete rewrite" and 97break such filepair into two filepairs that represent delete and 98create. E.g. If the input contained this filepair: 99 100------------------------------------------------ 101:100644 100644 bcd1234... 0123456... M file0 102------------------------------------------------ 103 104and if it detects that the file "file0" is completely rewritten, 105it changes it to: 106 107------------------------------------------------ 108:100644 000000 bcd1234... 0000000... D file0 109:000000 100644 0000000... 0123456... A file0 110------------------------------------------------ 111 112For the purpose of breaking a filepair, diffcore-break examines 113the extent of changes between the contents of the files before 114and after modification (i.e. the contents that have "bcd1234..." 115and "0123456..." as their SHA1 content ID, in the above 116example). The amount of deletion of original contents and 117insertion of new material are added together, and if it exceeds 118the "break score", the filepair is broken into two. The break 119score defaults to 50% of the size of the smaller of the original 120and the result (i.e. if the edit shrinks the file, the size of 121the result is used; if the edit lengthens the file, the size of 122the original is used), and can be customized by giving a number 123after "-B" option (e.g. "-B75" to tell it to use 75%). 124 125 126diffcore-rename: For Detection Renames and Copies 127------------------------------------------------- 128 129This transformation is used to detect renames and copies, and is 130controlled by the -M option (to detect renames) and the -C option 131(to detect copies as well) to the git-diff-* commands. If the 132input contained these filepairs: 133 134------------------------------------------------ 135:100644 000000 0123456... 0000000... D fileX 136:000000 100644 0000000... 0123456... A file0 137------------------------------------------------ 138 139and the contents of the deleted file fileX is similar enough to 140the contents of the created file file0, then rename detection 141merges these filepairs and creates: 142 143------------------------------------------------ 144:100644 100644 0123456... 0123456... R100 fileX file0 145------------------------------------------------ 146 147When the "-C" option is used, the original contents of modified files, 148and deleted files (and also unmodified files, if the 149"\--find-copies-harder" option is used) are considered as candidates 150of the source files in rename/copy operation. If the input were like 151these filepairs, that talk about a modified file fileY and a newly 152created file file0: 153 154------------------------------------------------ 155:100644 100644 0123456... 1234567... M fileY 156:000000 100644 0000000... bcd3456... A file0 157------------------------------------------------ 158 159the original contents of fileY and the resulting contents of 160file0 are compared, and if they are similar enough, they are 161changed to: 162 163------------------------------------------------ 164:100644 100644 0123456... 1234567... M fileY 165:100644 100644 0123456... bcd3456... C100 fileY file0 166------------------------------------------------ 167 168In both rename and copy detection, the same "extent of changes" 169algorithm used in diffcore-break is used to determine if two 170files are "similar enough", and can be customized to use 171a similarity score different from the default of 50% by giving a 172number after the "-M" or "-C" option (e.g. "-M8" to tell it to use 1738/10 = 80%). 174 175Note. When the "-C" option is used with `\--find-copies-harder` 176option, git-diff-\* commands feed unmodified filepairs to 177diffcore mechanism as well as modified ones. This lets the copy 178detector consider unmodified files as copy source candidates at 179the expense of making it slower. Without `\--find-copies-harder`, 180git-diff-\* commands can detect copies only if the file that was 181copied happened to have been modified in the same changeset. 182 183 184diffcore-merge-broken: For Putting "Complete Rewrites" Back Together 185-------------------------------------------------------------------- 186 187This transformation is used to merge filepairs broken by 188diffcore-break, and not transformed into rename/copy by 189diffcore-rename, back into a single modification. This always 190runs when diffcore-break is used. 191 192For the purpose of merging broken filepairs back, it uses a 193different "extent of changes" computation from the ones used by 194diffcore-break and diffcore-rename. It counts only the deletion 195from the original, and does not count insertion. If you removed 196only 10 lines from a 100-line document, even if you added 910 197new lines to make a new 1000-line document, you did not do a 198complete rewrite. diffcore-break breaks such a case in order to 199help diffcore-rename to consider such filepairs as candidate of 200rename/copy detection, but if filepairs broken that way were not 201matched with other filepairs to create rename/copy, then this 202transformation merges them back into the original 203"modification". 204 205The "extent of changes" parameter can be tweaked from the 206default 80% (that is, unless more than 80% of the original 207material is deleted, the broken pairs are merged back into a 208single modification) by giving a second number to -B option, 209like these: 210 211* -B50/60 (give 50% "break score" to diffcore-break, use 60% 212 for diffcore-merge-broken). 213 214* -B/60 (the same as above, since diffcore-break defaults to 50%). 215 216Note that earlier implementation left a broken pair as a separate 217creation and deletion patches. This was an unnecessary hack and 218the latest implementation always merges all the broken pairs 219back into modifications, but the resulting patch output is 220formatted differently for easier review in case of such 221a complete rewrite by showing the entire contents of old version 222prefixed with '-', followed by the entire contents of new 223version prefixed with '+'. 224 225 226diffcore-pickaxe: For Detecting Addition/Deletion of Specified String 227--------------------------------------------------------------------- 228 229This transformation is used to find filepairs that represent 230changes that touch a specified string, and is controlled by the 231-S option and the `\--pickaxe-all` option to the git-diff-* 232commands. 233 234When diffcore-pickaxe is in use, it checks if there are 235filepairs whose "original" side has the specified string and 236whose "result" side does not. Such a filepair represents "the 237string appeared in this changeset". It also checks for the 238opposite case that loses the specified string. 239 240When `\--pickaxe-all` is not in effect, diffcore-pickaxe leaves 241only such filepairs that touch the specified string in its 242output. When `\--pickaxe-all` is used, diffcore-pickaxe leaves all 243filepairs intact if there is such a filepair, or makes the 244output empty otherwise. The latter behaviour is designed to 245make reviewing of the changes in the context of the whole 246changeset easier. 247 248 249diffcore-order: For Sorting the Output Based on Filenames 250--------------------------------------------------------- 251 252This is used to reorder the filepairs according to the user's 253(or project's) taste, and is controlled by the -O option to the 254git-diff-* commands. 255 256This takes a text file each of whose lines is a shell glob 257pattern. Filepairs that match a glob pattern on an earlier line 258in the file are output before ones that match a later line, and 259filepairs that do not match any glob pattern are output last. 260 261As an example, a typical orderfile for the core git probably 262would look like this: 263 264------------------------------------------------ 265README 266Makefile 267Documentation 268*.h 269*.c 270t 271------------------------------------------------ 272