subtree: make --ignore-joins pay attention to adds
[git] / contrib / subtree / git-subtree.sh
1 #!/bin/sh
2 #
3 # git-subtree.sh: split/join git repositories in subdirectories of this one
4 #
5 # Copyright (C) 2009 Avery Pennarun <apenwarr@gmail.com>
6 #
7 if test $# -eq 0
8 then
9         set -- -h
10 fi
11 OPTS_SPEC="\
12 git subtree add   --prefix=<prefix> <commit>
13 git subtree add   --prefix=<prefix> <repository> <ref>
14 git subtree merge --prefix=<prefix> <commit>
15 git subtree pull  --prefix=<prefix> <repository> <ref>
16 git subtree push  --prefix=<prefix> <repository> <ref>
17 git subtree split --prefix=<prefix> <commit...>
18 --
19 h,help        show the help
20 q             quiet
21 d             show debug messages
22 P,prefix=     the name of the subdir to split out
23 m,message=    use the given message as the commit message for the merge commit
24  options for 'split'
25 annotate=     add a prefix to commit message of new commits
26 b,branch=     create a new branch from the split subtree
27 ignore-joins  ignore prior --rejoin commits
28 onto=         try connecting new tree to an existing one
29 rejoin        merge the new branch back into HEAD
30  options for 'add', 'merge', and 'pull'
31 squash        merge subtree changes as a single commit
32 "
33 eval "$(echo "$OPTS_SPEC" | git rev-parse --parseopt -- "$@" || echo exit $?)"
34
35 PATH=$PATH:$(git --exec-path)
36 . git-sh-setup
37
38 require_work_tree
39
40 quiet=
41 branch=
42 debug=
43 command=
44 onto=
45 rejoin=
46 ignore_joins=
47 annotate=
48 squash=
49 message=
50 prefix=
51
52 debug () {
53         if test -n "$debug"
54         then
55                 printf "%s\n" "$*" >&2
56         fi
57 }
58
59 say () {
60         if test -z "$quiet"
61         then
62                 printf "%s\n" "$*" >&2
63         fi
64 }
65
66 progress () {
67         if test -z "$quiet"
68         then
69                 printf "%s\r" "$*" >&2
70         fi
71 }
72
73 assert () {
74         if ! "$@"
75         then
76                 die "assertion failed: " "$@"
77         fi
78 }
79
80
81 while test $# -gt 0
82 do
83         opt="$1"
84         shift
85
86         case "$opt" in
87         -q)
88                 quiet=1
89                 ;;
90         -d)
91                 debug=1
92                 ;;
93         --annotate)
94                 annotate="$1"
95                 shift
96                 ;;
97         --no-annotate)
98                 annotate=
99                 ;;
100         -b)
101                 branch="$1"
102                 shift
103                 ;;
104         -P)
105                 prefix="${1%/}"
106                 shift
107                 ;;
108         -m)
109                 message="$1"
110                 shift
111                 ;;
112         --no-prefix)
113                 prefix=
114                 ;;
115         --onto)
116                 onto="$1"
117                 shift
118                 ;;
119         --no-onto)
120                 onto=
121                 ;;
122         --rejoin)
123                 rejoin=1
124                 ;;
125         --no-rejoin)
126                 rejoin=
127                 ;;
128         --ignore-joins)
129                 ignore_joins=1
130                 ;;
131         --no-ignore-joins)
132                 ignore_joins=
133                 ;;
134         --squash)
135                 squash=1
136                 ;;
137         --no-squash)
138                 squash=
139                 ;;
140         --)
141                 break
142                 ;;
143         *)
144                 die "Unexpected option: $opt"
145                 ;;
146         esac
147 done
148
149 command="$1"
150 shift
151
152 case "$command" in
153 add|merge|pull)
154         default=
155         ;;
156 split|push)
157         default="--default HEAD"
158         ;;
159 *)
160         die "Unknown command '$command'"
161         ;;
162 esac
163
164 if test -z "$prefix"
165 then
166         die "You must provide the --prefix option."
167 fi
168
169 case "$command" in
170 add)
171         test -e "$prefix" &&
172                 die "prefix '$prefix' already exists."
173         ;;
174 *)
175         test -e "$prefix" ||
176                 die "'$prefix' does not exist; use 'git subtree add'"
177         ;;
178 esac
179
180 dir="$(dirname "$prefix/.")"
181
182 if test "$command" != "pull" &&
183                 test "$command" != "add" &&
184                 test "$command" != "push"
185 then
186         revs=$(git rev-parse $default --revs-only "$@") || exit $?
187         dirs=$(git rev-parse --no-revs --no-flags "$@") || exit $?
188         if test -n "$dirs"
189         then
190                 die "Error: Use --prefix instead of bare filenames."
191         fi
192 fi
193
194 debug "command: {$command}"
195 debug "quiet: {$quiet}"
196 debug "revs: {$revs}"
197 debug "dir: {$dir}"
198 debug "opts: {$*}"
199 debug
200
201 cache_setup () {
202         cachedir="$GIT_DIR/subtree-cache/$$"
203         rm -rf "$cachedir" ||
204                 die "Can't delete old cachedir: $cachedir"
205         mkdir -p "$cachedir" ||
206                 die "Can't create new cachedir: $cachedir"
207         mkdir -p "$cachedir/notree" ||
208                 die "Can't create new cachedir: $cachedir/notree"
209         debug "Using cachedir: $cachedir" >&2
210 }
211
212 cache_get () {
213         for oldrev in "$@"
214         do
215                 if test -r "$cachedir/$oldrev"
216                 then
217                         read newrev <"$cachedir/$oldrev"
218                         echo $newrev
219                 fi
220         done
221 }
222
223 cache_miss () {
224         for oldrev in "$@"
225         do
226                 if ! test -r "$cachedir/$oldrev"
227                 then
228                         echo $oldrev
229                 fi
230         done
231 }
232
233 check_parents () {
234         missed=$(cache_miss "$@")
235         for miss in $missed
236         do
237                 if ! test -r "$cachedir/notree/$miss"
238                 then
239                         debug "  incorrect order: $miss"
240                 fi
241         done
242 }
243
244 set_notree () {
245         echo "1" > "$cachedir/notree/$1"
246 }
247
248 cache_set () {
249         oldrev="$1"
250         newrev="$2"
251         if test "$oldrev" != "latest_old" &&
252                 test "$oldrev" != "latest_new" &&
253                 test -e "$cachedir/$oldrev"
254         then
255                 die "cache for $oldrev already exists!"
256         fi
257         echo "$newrev" >"$cachedir/$oldrev"
258 }
259
260 rev_exists () {
261         if git rev-parse "$1" >/dev/null 2>&1
262         then
263                 return 0
264         else
265                 return 1
266         fi
267 }
268
269 rev_is_descendant_of_branch () {
270         newrev="$1"
271         branch="$2"
272         branch_hash=$(git rev-parse "$branch")
273         match=$(git rev-list -1 "$branch_hash" "^$newrev")
274
275         if test -z "$match"
276         then
277                 return 0
278         else
279                 return 1
280         fi
281 }
282
283 # if a commit doesn't have a parent, this might not work.  But we only want
284 # to remove the parent from the rev-list, and since it doesn't exist, it won't
285 # be there anyway, so do nothing in that case.
286 try_remove_previous () {
287         if rev_exists "$1^"
288         then
289                 echo "^$1^"
290         fi
291 }
292
293 find_latest_squash () {
294         debug "Looking for latest squash ($dir)..."
295         dir="$1"
296         sq=
297         main=
298         sub=
299         git log --grep="^git-subtree-dir: $dir/*\$" \
300                 --no-show-signature --pretty=format:'START %H%n%s%n%n%b%nEND%n' HEAD |
301         while read a b junk
302         do
303                 debug "$a $b $junk"
304                 debug "{{$sq/$main/$sub}}"
305                 case "$a" in
306                 START)
307                         sq="$b"
308                         ;;
309                 git-subtree-mainline:)
310                         main="$b"
311                         ;;
312                 git-subtree-split:)
313                         sub="$(git rev-parse "$b^0")" ||
314                         die "could not rev-parse split hash $b from commit $sq"
315                         ;;
316                 END)
317                         if test -n "$sub"
318                         then
319                                 if test -n "$main"
320                                 then
321                                         # a rejoin commit?
322                                         # Pretend its sub was a squash.
323                                         sq="$sub"
324                                 fi
325                                 debug "Squash found: $sq $sub"
326                                 echo "$sq" "$sub"
327                                 break
328                         fi
329                         sq=
330                         main=
331                         sub=
332                         ;;
333                 esac
334         done
335 }
336
337 find_existing_splits () {
338         debug "Looking for prior splits..."
339         dir="$1"
340         revs="$2"
341         main=
342         sub=
343         local grep_format="^git-subtree-dir: $dir/*\$"
344         if test -n "$ignore_joins"
345         then
346                 grep_format="^Add '$dir/' from commit '"
347         fi
348         git log --grep="$grep_format" \
349                 --no-show-signature --pretty=format:'START %H%n%s%n%n%b%nEND%n' $revs |
350         while read a b junk
351         do
352                 case "$a" in
353                 START)
354                         sq="$b"
355                         ;;
356                 git-subtree-mainline:)
357                         main="$b"
358                         ;;
359                 git-subtree-split:)
360                         sub="$(git rev-parse "$b^0")" ||
361                         die "could not rev-parse split hash $b from commit $sq"
362                         ;;
363                 END)
364                         debug "  Main is: '$main'"
365                         if test -z "$main" -a -n "$sub"
366                         then
367                                 # squash commits refer to a subtree
368                                 debug "  Squash: $sq from $sub"
369                                 cache_set "$sq" "$sub"
370                         fi
371                         if test -n "$main" -a -n "$sub"
372                         then
373                                 debug "  Prior: $main -> $sub"
374                                 cache_set $main $sub
375                                 cache_set $sub $sub
376                                 try_remove_previous "$main"
377                                 try_remove_previous "$sub"
378                         fi
379                         main=
380                         sub=
381                         ;;
382                 esac
383         done
384 }
385
386 copy_commit () {
387         # We're going to set some environment vars here, so
388         # do it in a subshell to get rid of them safely later
389         debug copy_commit "{$1}" "{$2}" "{$3}"
390         git log -1 --no-show-signature --pretty=format:'%an%n%ae%n%aD%n%cn%n%ce%n%cD%n%B' "$1" |
391         (
392                 read GIT_AUTHOR_NAME
393                 read GIT_AUTHOR_EMAIL
394                 read GIT_AUTHOR_DATE
395                 read GIT_COMMITTER_NAME
396                 read GIT_COMMITTER_EMAIL
397                 read GIT_COMMITTER_DATE
398                 export  GIT_AUTHOR_NAME \
399                         GIT_AUTHOR_EMAIL \
400                         GIT_AUTHOR_DATE \
401                         GIT_COMMITTER_NAME \
402                         GIT_COMMITTER_EMAIL \
403                         GIT_COMMITTER_DATE
404                 (
405                         printf "%s" "$annotate"
406                         cat
407                 ) |
408                 git commit-tree "$2" $3  # reads the rest of stdin
409         ) || die "Can't copy commit $1"
410 }
411
412 add_msg () {
413         dir="$1"
414         latest_old="$2"
415         latest_new="$3"
416         if test -n "$message"
417         then
418                 commit_message="$message"
419         else
420                 commit_message="Add '$dir/' from commit '$latest_new'"
421         fi
422         cat <<-EOF
423                 $commit_message
424
425                 git-subtree-dir: $dir
426                 git-subtree-mainline: $latest_old
427                 git-subtree-split: $latest_new
428         EOF
429 }
430
431 add_squashed_msg () {
432         if test -n "$message"
433         then
434                 echo "$message"
435         else
436                 echo "Merge commit '$1' as '$2'"
437         fi
438 }
439
440 rejoin_msg () {
441         dir="$1"
442         latest_old="$2"
443         latest_new="$3"
444         if test -n "$message"
445         then
446                 commit_message="$message"
447         else
448                 commit_message="Split '$dir/' into commit '$latest_new'"
449         fi
450         cat <<-EOF
451                 $commit_message
452
453                 git-subtree-dir: $dir
454                 git-subtree-mainline: $latest_old
455                 git-subtree-split: $latest_new
456         EOF
457 }
458
459 squash_msg () {
460         dir="$1"
461         oldsub="$2"
462         newsub="$3"
463         newsub_short=$(git rev-parse --short "$newsub")
464
465         if test -n "$oldsub"
466         then
467                 oldsub_short=$(git rev-parse --short "$oldsub")
468                 echo "Squashed '$dir/' changes from $oldsub_short..$newsub_short"
469                 echo
470                 git log --no-show-signature --pretty=tformat:'%h %s' "$oldsub..$newsub"
471                 git log --no-show-signature --pretty=tformat:'REVERT: %h %s' "$newsub..$oldsub"
472         else
473                 echo "Squashed '$dir/' content from commit $newsub_short"
474         fi
475
476         echo
477         echo "git-subtree-dir: $dir"
478         echo "git-subtree-split: $newsub"
479 }
480
481 toptree_for_commit () {
482         commit="$1"
483         git rev-parse --verify "$commit^{tree}" || exit $?
484 }
485
486 subtree_for_commit () {
487         commit="$1"
488         dir="$2"
489         git ls-tree "$commit" -- "$dir" |
490         while read mode type tree name
491         do
492                 assert test "$name" = "$dir"
493                 assert test "$type" = "tree" -o "$type" = "commit"
494                 test "$type" = "commit" && continue  # ignore submodules
495                 echo $tree
496                 break
497         done
498 }
499
500 tree_changed () {
501         tree=$1
502         shift
503         if test $# -ne 1
504         then
505                 return 0   # weird parents, consider it changed
506         else
507                 ptree=$(toptree_for_commit $1)
508                 if test "$ptree" != "$tree"
509                 then
510                         return 0   # changed
511                 else
512                         return 1   # not changed
513                 fi
514         fi
515 }
516
517 new_squash_commit () {
518         old="$1"
519         oldsub="$2"
520         newsub="$3"
521         tree=$(toptree_for_commit $newsub) || exit $?
522         if test -n "$old"
523         then
524                 squash_msg "$dir" "$oldsub" "$newsub" |
525                 git commit-tree "$tree" -p "$old" || exit $?
526         else
527                 squash_msg "$dir" "" "$newsub" |
528                 git commit-tree "$tree" || exit $?
529         fi
530 }
531
532 copy_or_skip () {
533         rev="$1"
534         tree="$2"
535         newparents="$3"
536         assert test -n "$tree"
537
538         identical=
539         nonidentical=
540         p=
541         gotparents=
542         for parent in $newparents
543         do
544                 ptree=$(toptree_for_commit $parent) || exit $?
545                 test -z "$ptree" && continue
546                 if test "$ptree" = "$tree"
547                 then
548                         # an identical parent could be used in place of this rev.
549                         identical="$parent"
550                 else
551                         nonidentical="$parent"
552                 fi
553
554                 # sometimes both old parents map to the same newparent;
555                 # eliminate duplicates
556                 is_new=1
557                 for gp in $gotparents
558                 do
559                         if test "$gp" = "$parent"
560                         then
561                                 is_new=
562                                 break
563                         fi
564                 done
565                 if test -n "$is_new"
566                 then
567                         gotparents="$gotparents $parent"
568                         p="$p -p $parent"
569                 fi
570         done
571
572         copycommit=
573         if test -n "$identical" && test -n "$nonidentical"
574         then
575                 extras=$(git rev-list --count $identical..$nonidentical)
576                 if test "$extras" -ne 0
577                 then
578                         # we need to preserve history along the other branch
579                         copycommit=1
580                 fi
581         fi
582         if test -n "$identical" && test -z "$copycommit"
583         then
584                 echo $identical
585         else
586                 copy_commit "$rev" "$tree" "$p" || exit $?
587         fi
588 }
589
590 ensure_clean () {
591         if ! git diff-index HEAD --exit-code --quiet 2>&1
592         then
593                 die "Working tree has modifications.  Cannot add."
594         fi
595         if ! git diff-index --cached HEAD --exit-code --quiet 2>&1
596         then
597                 die "Index has modifications.  Cannot add."
598         fi
599 }
600
601 ensure_valid_ref_format () {
602         git check-ref-format "refs/heads/$1" ||
603                 die "'$1' does not look like a ref"
604 }
605
606 process_split_commit () {
607         local rev="$1"
608         local parents="$2"
609         revcount=$(($revcount + 1))
610         progress "$revcount/$revmax ($createcount)"
611         debug "Processing commit: $rev"
612         exists=$(cache_get "$rev")
613         if test -n "$exists"
614         then
615                 debug "  prior: $exists"
616                 return
617         fi
618         createcount=$(($createcount + 1))
619         debug "  parents: $parents"
620         newparents=$(cache_get $parents)
621         debug "  newparents: $newparents"
622
623         tree=$(subtree_for_commit "$rev" "$dir")
624         debug "  tree is: $tree"
625
626         check_parents $parents
627
628         # ugly.  is there no better way to tell if this is a subtree
629         # vs. a mainline commit?  Does it matter?
630         if test -z "$tree"
631         then
632                 set_notree "$rev"
633                 if test -n "$newparents"
634                 then
635                         cache_set "$rev" "$rev"
636                 fi
637                 return
638         fi
639
640         newrev=$(copy_or_skip "$rev" "$tree" "$newparents") || exit $?
641         debug "  newrev is: $newrev"
642         cache_set "$rev" "$newrev"
643         cache_set latest_new "$newrev"
644         cache_set latest_old "$rev"
645 }
646
647 cmd_add () {
648         if test -e "$dir"
649         then
650                 die "'$dir' already exists.  Cannot add."
651         fi
652
653         ensure_clean
654
655         if test $# -eq 1
656         then
657                 git rev-parse -q --verify "$1^{commit}" >/dev/null ||
658                         die "'$1' does not refer to a commit"
659
660                 cmd_add_commit "$@"
661
662         elif test $# -eq 2
663         then
664                 # Technically we could accept a refspec here but we're
665                 # just going to turn around and add FETCH_HEAD under the
666                 # specified directory.  Allowing a refspec might be
667                 # misleading because we won't do anything with any other
668                 # branches fetched via the refspec.
669                 ensure_valid_ref_format "$2"
670
671                 cmd_add_repository "$@"
672         else
673                 say "error: parameters were '$@'"
674                 die "Provide either a commit or a repository and commit."
675         fi
676 }
677
678 cmd_add_repository () {
679         echo "git fetch" "$@"
680         repository=$1
681         refspec=$2
682         git fetch "$@" || exit $?
683         revs=FETCH_HEAD
684         set -- $revs
685         cmd_add_commit "$@"
686 }
687
688 cmd_add_commit () {
689         revs=$(git rev-parse $default --revs-only "$@") || exit $?
690         set -- $revs
691         rev="$1"
692
693         debug "Adding $dir as '$rev'..."
694         git read-tree --prefix="$dir" $rev || exit $?
695         git checkout -- "$dir" || exit $?
696         tree=$(git write-tree) || exit $?
697
698         headrev=$(git rev-parse HEAD) || exit $?
699         if test -n "$headrev" && test "$headrev" != "$rev"
700         then
701                 headp="-p $headrev"
702         else
703                 headp=
704         fi
705
706         if test -n "$squash"
707         then
708                 rev=$(new_squash_commit "" "" "$rev") || exit $?
709                 commit=$(add_squashed_msg "$rev" "$dir" |
710                         git commit-tree "$tree" $headp -p "$rev") || exit $?
711         else
712                 revp=$(peel_committish "$rev") &&
713                 commit=$(add_msg "$dir" $headrev "$rev" |
714                         git commit-tree "$tree" $headp -p "$revp") || exit $?
715         fi
716         git reset "$commit" || exit $?
717
718         say "Added dir '$dir'"
719 }
720
721 cmd_split () {
722         debug "Splitting $dir..."
723         cache_setup || exit $?
724
725         if test -n "$onto"
726         then
727                 debug "Reading history for --onto=$onto..."
728                 git rev-list $onto |
729                 while read rev
730                 do
731                         # the 'onto' history is already just the subdir, so
732                         # any parent we find there can be used verbatim
733                         debug "  cache: $rev"
734                         cache_set "$rev" "$rev"
735                 done
736         fi
737
738         unrevs="$(find_existing_splits "$dir" "$revs")"
739
740         # We can't restrict rev-list to only $dir here, because some of our
741         # parents have the $dir contents the root, and those won't match.
742         # (and rev-list --follow doesn't seem to solve this)
743         grl='git rev-list --topo-order --reverse --parents $revs $unrevs'
744         revmax=$(eval "$grl" | wc -l)
745         revcount=0
746         createcount=0
747         eval "$grl" |
748         while read rev parents
749         do
750                 process_split_commit "$rev" "$parents"
751         done || exit $?
752
753         latest_new=$(cache_get latest_new)
754         if test -z "$latest_new"
755         then
756                 die "No new revisions were found"
757         fi
758
759         if test -n "$rejoin"
760         then
761                 debug "Merging split branch into HEAD..."
762                 latest_old=$(cache_get latest_old)
763                 git merge -s ours \
764                         --allow-unrelated-histories \
765                         -m "$(rejoin_msg "$dir" "$latest_old" "$latest_new")" \
766                         "$latest_new" >&2 || exit $?
767         fi
768         if test -n "$branch"
769         then
770                 if rev_exists "refs/heads/$branch"
771                 then
772                         if ! rev_is_descendant_of_branch "$latest_new" "$branch"
773                         then
774                                 die "Branch '$branch' is not an ancestor of commit '$latest_new'."
775                         fi
776                         action='Updated'
777                 else
778                         action='Created'
779                 fi
780                 git update-ref -m 'subtree split' \
781                         "refs/heads/$branch" "$latest_new" || exit $?
782                 say "$action branch '$branch'"
783         fi
784         echo "$latest_new"
785         exit 0
786 }
787
788 cmd_merge () {
789         revs=$(git rev-parse $default --revs-only "$@") || exit $?
790         ensure_clean
791
792         set -- $revs
793         if test $# -ne 1
794         then
795                 die "You must provide exactly one revision.  Got: '$revs'"
796         fi
797         rev="$1"
798
799         if test -n "$squash"
800         then
801                 first_split="$(find_latest_squash "$dir")"
802                 if test -z "$first_split"
803                 then
804                         die "Can't squash-merge: '$dir' was never added."
805                 fi
806                 set $first_split
807                 old=$1
808                 sub=$2
809                 if test "$sub" = "$rev"
810                 then
811                         say "Subtree is already at commit $rev."
812                         exit 0
813                 fi
814                 new=$(new_squash_commit "$old" "$sub" "$rev") || exit $?
815                 debug "New squash commit: $new"
816                 rev="$new"
817         fi
818
819         version=$(git version)
820         if test "$version" \< "git version 1.7"
821         then
822                 if test -n "$message"
823                 then
824                         git merge -s subtree --message="$message" "$rev"
825                 else
826                         git merge -s subtree "$rev"
827                 fi
828         else
829                 if test -n "$message"
830                 then
831                         git merge -Xsubtree="$prefix" \
832                                 --message="$message" "$rev"
833                 else
834                         git merge -Xsubtree="$prefix" $rev
835                 fi
836         fi
837 }
838
839 cmd_pull () {
840         if test $# -ne 2
841         then
842                 die "You must provide <repository> <ref>"
843         fi
844         ensure_clean
845         ensure_valid_ref_format "$2"
846         git fetch "$@" || exit $?
847         revs=FETCH_HEAD
848         set -- $revs
849         cmd_merge "$@"
850 }
851
852 cmd_push () {
853         if test $# -ne 2
854         then
855                 die "You must provide <repository> <ref>"
856         fi
857         ensure_valid_ref_format "$2"
858         if test -e "$dir"
859         then
860                 repository=$1
861                 refspec=$2
862                 echo "git push using: " "$repository" "$refspec"
863                 localrev=$(git subtree split --prefix="$prefix") || die
864                 git push "$repository" "$localrev":"refs/heads/$refspec"
865         else
866                 die "'$dir' must already exist. Try 'git subtree add'."
867         fi
868 }
869
870 "cmd_$command" "$@"