Merge branch 'js/import-tars-hardlinks'
[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: $dir/*\$" \
300                 --pretty=format:'START %H%n%s%n%n%b%nEND%n' HEAD |
301         while read a b junk
302         do
303                 debug "$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         git log --grep="^git-subtree-dir: $dir/*\$" \
344                 --pretty=format:'START %H%n%s%n%n%b%nEND%n' $revs |
345         while read a b junk
346         do
347                 case "$a" in
348                 START)
349                         sq="$b"
350                         ;;
351                 git-subtree-mainline:)
352                         main="$b"
353                         ;;
354                 git-subtree-split:)
355                         sub="$(git rev-parse "$b^0")" ||
356                         die "could not rev-parse split hash $b from commit $sq"
357                         ;;
358                 END)
359                         debug "  Main is: '$main'"
360                         if test -z "$main" -a -n "$sub"
361                         then
362                                 # squash commits refer to a subtree
363                                 debug "  Squash: $sq from $sub"
364                                 cache_set "$sq" "$sub"
365                         fi
366                         if test -n "$main" -a -n "$sub"
367                         then
368                                 debug "  Prior: $main -> $sub"
369                                 cache_set $main $sub
370                                 cache_set $sub $sub
371                                 try_remove_previous "$main"
372                                 try_remove_previous "$sub"
373                         fi
374                         main=
375                         sub=
376                         ;;
377                 esac
378         done
379 }
380
381 copy_commit () {
382         # We're going to set some environment vars here, so
383         # do it in a subshell to get rid of them safely later
384         debug copy_commit "{$1}" "{$2}" "{$3}"
385         git log -1 --pretty=format:'%an%n%ae%n%aD%n%cn%n%ce%n%cD%n%B' "$1" |
386         (
387                 read GIT_AUTHOR_NAME
388                 read GIT_AUTHOR_EMAIL
389                 read GIT_AUTHOR_DATE
390                 read GIT_COMMITTER_NAME
391                 read GIT_COMMITTER_EMAIL
392                 read GIT_COMMITTER_DATE
393                 export  GIT_AUTHOR_NAME \
394                         GIT_AUTHOR_EMAIL \
395                         GIT_AUTHOR_DATE \
396                         GIT_COMMITTER_NAME \
397                         GIT_COMMITTER_EMAIL \
398                         GIT_COMMITTER_DATE
399                 (
400                         printf "%s" "$annotate"
401                         cat
402                 ) |
403                 git commit-tree "$2" $3  # reads the rest of stdin
404         ) || die "Can't copy commit $1"
405 }
406
407 add_msg () {
408         dir="$1"
409         latest_old="$2"
410         latest_new="$3"
411         if test -n "$message"
412         then
413                 commit_message="$message"
414         else
415                 commit_message="Add '$dir/' from commit '$latest_new'"
416         fi
417         cat <<-EOF
418                 $commit_message
419
420                 git-subtree-dir: $dir
421                 git-subtree-mainline: $latest_old
422                 git-subtree-split: $latest_new
423         EOF
424 }
425
426 add_squashed_msg () {
427         if test -n "$message"
428         then
429                 echo "$message"
430         else
431                 echo "Merge commit '$1' as '$2'"
432         fi
433 }
434
435 rejoin_msg () {
436         dir="$1"
437         latest_old="$2"
438         latest_new="$3"
439         if test -n "$message"
440         then
441                 commit_message="$message"
442         else
443                 commit_message="Split '$dir/' into commit '$latest_new'"
444         fi
445         cat <<-EOF
446                 $commit_message
447
448                 git-subtree-dir: $dir
449                 git-subtree-mainline: $latest_old
450                 git-subtree-split: $latest_new
451         EOF
452 }
453
454 squash_msg () {
455         dir="$1"
456         oldsub="$2"
457         newsub="$3"
458         newsub_short=$(git rev-parse --short "$newsub")
459
460         if test -n "$oldsub"
461         then
462                 oldsub_short=$(git rev-parse --short "$oldsub")
463                 echo "Squashed '$dir/' changes from $oldsub_short..$newsub_short"
464                 echo
465                 git log --pretty=tformat:'%h %s' "$oldsub..$newsub"
466                 git log --pretty=tformat:'REVERT: %h %s' "$newsub..$oldsub"
467         else
468                 echo "Squashed '$dir/' content from commit $newsub_short"
469         fi
470
471         echo
472         echo "git-subtree-dir: $dir"
473         echo "git-subtree-split: $newsub"
474 }
475
476 toptree_for_commit () {
477         commit="$1"
478         git log -1 --pretty=format:'%T' "$commit" -- || exit $?
479 }
480
481 subtree_for_commit () {
482         commit="$1"
483         dir="$2"
484         git ls-tree "$commit" -- "$dir" |
485         while read mode type tree name
486         do
487                 assert test "$name" = "$dir"
488                 assert test "$type" = "tree" -o "$type" = "commit"
489                 test "$type" = "commit" && continue  # ignore submodules
490                 echo $tree
491                 break
492         done
493 }
494
495 tree_changed () {
496         tree=$1
497         shift
498         if test $# -ne 1
499         then
500                 return 0   # weird parents, consider it changed
501         else
502                 ptree=$(toptree_for_commit $1)
503                 if test "$ptree" != "$tree"
504                 then
505                         return 0   # changed
506                 else
507                         return 1   # not changed
508                 fi
509         fi
510 }
511
512 new_squash_commit () {
513         old="$1"
514         oldsub="$2"
515         newsub="$3"
516         tree=$(toptree_for_commit $newsub) || exit $?
517         if test -n "$old"
518         then
519                 squash_msg "$dir" "$oldsub" "$newsub" |
520                 git commit-tree "$tree" -p "$old" || exit $?
521         else
522                 squash_msg "$dir" "" "$newsub" |
523                 git commit-tree "$tree" || exit $?
524         fi
525 }
526
527 copy_or_skip () {
528         rev="$1"
529         tree="$2"
530         newparents="$3"
531         assert test -n "$tree"
532
533         identical=
534         nonidentical=
535         p=
536         gotparents=
537         for parent in $newparents
538         do
539                 ptree=$(toptree_for_commit $parent) || exit $?
540                 test -z "$ptree" && continue
541                 if test "$ptree" = "$tree"
542                 then
543                         # an identical parent could be used in place of this rev.
544                         identical="$parent"
545                 else
546                         nonidentical="$parent"
547                 fi
548
549                 # sometimes both old parents map to the same newparent;
550                 # eliminate duplicates
551                 is_new=1
552                 for gp in $gotparents
553                 do
554                         if test "$gp" = "$parent"
555                         then
556                                 is_new=
557                                 break
558                         fi
559                 done
560                 if test -n "$is_new"
561                 then
562                         gotparents="$gotparents $parent"
563                         p="$p -p $parent"
564                 fi
565         done
566
567         copycommit=
568         if test -n "$identical" && test -n "$nonidentical"
569         then
570                 extras=$(git rev-list --count $identical..$nonidentical)
571                 if test "$extras" -ne 0
572                 then
573                         # we need to preserve history along the other branch
574                         copycommit=1
575                 fi
576         fi
577         if test -n "$identical" && test -z "$copycommit"
578         then
579                 echo $identical
580         else
581                 copy_commit "$rev" "$tree" "$p" || exit $?
582         fi
583 }
584
585 ensure_clean () {
586         if ! git diff-index HEAD --exit-code --quiet 2>&1
587         then
588                 die "Working tree has modifications.  Cannot add."
589         fi
590         if ! git diff-index --cached HEAD --exit-code --quiet 2>&1
591         then
592                 die "Index has modifications.  Cannot add."
593         fi
594 }
595
596 ensure_valid_ref_format () {
597         git check-ref-format "refs/heads/$1" ||
598                 die "'$1' does not look like a ref"
599 }
600
601 cmd_add () {
602         if test -e "$dir"
603         then
604                 die "'$dir' already exists.  Cannot add."
605         fi
606
607         ensure_clean
608
609         if test $# -eq 1
610         then
611                 git rev-parse -q --verify "$1^{commit}" >/dev/null ||
612                         die "'$1' does not refer to a commit"
613
614                 cmd_add_commit "$@"
615
616         elif test $# -eq 2
617         then
618                 # Technically we could accept a refspec here but we're
619                 # just going to turn around and add FETCH_HEAD under the
620                 # specified directory.  Allowing a refspec might be
621                 # misleading because we won't do anything with any other
622                 # branches fetched via the refspec.
623                 ensure_valid_ref_format "$2"
624
625                 cmd_add_repository "$@"
626         else
627                 say "error: parameters were '$@'"
628                 die "Provide either a commit or a repository and commit."
629         fi
630 }
631
632 cmd_add_repository () {
633         echo "git fetch" "$@"
634         repository=$1
635         refspec=$2
636         git fetch "$@" || exit $?
637         revs=FETCH_HEAD
638         set -- $revs
639         cmd_add_commit "$@"
640 }
641
642 cmd_add_commit () {
643         revs=$(git rev-parse $default --revs-only "$@") || exit $?
644         set -- $revs
645         rev="$1"
646
647         debug "Adding $dir as '$rev'..."
648         git read-tree --prefix="$dir" $rev || exit $?
649         git checkout -- "$dir" || exit $?
650         tree=$(git write-tree) || exit $?
651
652         headrev=$(git rev-parse HEAD) || exit $?
653         if test -n "$headrev" && test "$headrev" != "$rev"
654         then
655                 headp="-p $headrev"
656         else
657                 headp=
658         fi
659
660         if test -n "$squash"
661         then
662                 rev=$(new_squash_commit "" "" "$rev") || exit $?
663                 commit=$(add_squashed_msg "$rev" "$dir" |
664                         git commit-tree "$tree" $headp -p "$rev") || exit $?
665         else
666                 revp=$(peel_committish "$rev") &&
667                 commit=$(add_msg "$dir" $headrev "$rev" |
668                         git commit-tree "$tree" $headp -p "$revp") || exit $?
669         fi
670         git reset "$commit" || exit $?
671
672         say "Added dir '$dir'"
673 }
674
675 cmd_split () {
676         debug "Splitting $dir..."
677         cache_setup || exit $?
678
679         if test -n "$onto"
680         then
681                 debug "Reading history for --onto=$onto..."
682                 git rev-list $onto |
683                 while read rev
684                 do
685                         # the 'onto' history is already just the subdir, so
686                         # any parent we find there can be used verbatim
687                         debug "  cache: $rev"
688                         cache_set "$rev" "$rev"
689                 done
690         fi
691
692         if test -n "$ignore_joins"
693         then
694                 unrevs=
695         else
696                 unrevs="$(find_existing_splits "$dir" "$revs")"
697         fi
698
699         # We can't restrict rev-list to only $dir here, because some of our
700         # parents have the $dir contents the root, and those won't match.
701         # (and rev-list --follow doesn't seem to solve this)
702         grl='git rev-list --topo-order --reverse --parents $revs $unrevs'
703         revmax=$(eval "$grl" | wc -l)
704         revcount=0
705         createcount=0
706         eval "$grl" |
707         while read rev parents
708         do
709                 revcount=$(($revcount + 1))
710                 progress "$revcount/$revmax ($createcount)"
711                 debug "Processing commit: $rev"
712                 exists=$(cache_get "$rev")
713                 if test -n "$exists"
714                 then
715                         debug "  prior: $exists"
716                         continue
717                 fi
718                 createcount=$(($createcount + 1))
719                 debug "  parents: $parents"
720                 newparents=$(cache_get $parents)
721                 debug "  newparents: $newparents"
722
723                 tree=$(subtree_for_commit "$rev" "$dir")
724                 debug "  tree is: $tree"
725
726                 check_parents $parents
727
728                 # ugly.  is there no better way to tell if this is a subtree
729                 # vs. a mainline commit?  Does it matter?
730                 if test -z "$tree"
731                 then
732                         set_notree "$rev"
733                         if test -n "$newparents"
734                         then
735                                 cache_set "$rev" "$rev"
736                         fi
737                         continue
738                 fi
739
740                 newrev=$(copy_or_skip "$rev" "$tree" "$newparents") || exit $?
741                 debug "  newrev is: $newrev"
742                 cache_set "$rev" "$newrev"
743                 cache_set latest_new "$newrev"
744                 cache_set latest_old "$rev"
745         done || exit $?
746
747         latest_new=$(cache_get latest_new)
748         if test -z "$latest_new"
749         then
750                 die "No new revisions were found"
751         fi
752
753         if test -n "$rejoin"
754         then
755                 debug "Merging split branch into HEAD..."
756                 latest_old=$(cache_get latest_old)
757                 git merge -s ours \
758                         --allow-unrelated-histories \
759                         -m "$(rejoin_msg "$dir" "$latest_old" "$latest_new")" \
760                         "$latest_new" >&2 || exit $?
761         fi
762         if test -n "$branch"
763         then
764                 if rev_exists "refs/heads/$branch"
765                 then
766                         if ! rev_is_descendant_of_branch "$latest_new" "$branch"
767                         then
768                                 die "Branch '$branch' is not an ancestor of commit '$latest_new'."
769                         fi
770                         action='Updated'
771                 else
772                         action='Created'
773                 fi
774                 git update-ref -m 'subtree split' \
775                         "refs/heads/$branch" "$latest_new" || exit $?
776                 say "$action branch '$branch'"
777         fi
778         echo "$latest_new"
779         exit 0
780 }
781
782 cmd_merge () {
783         revs=$(git rev-parse $default --revs-only "$@") || exit $?
784         ensure_clean
785
786         set -- $revs
787         if test $# -ne 1
788         then
789                 die "You must provide exactly one revision.  Got: '$revs'"
790         fi
791         rev="$1"
792
793         if test -n "$squash"
794         then
795                 first_split="$(find_latest_squash "$dir")"
796                 if test -z "$first_split"
797                 then
798                         die "Can't squash-merge: '$dir' was never added."
799                 fi
800                 set $first_split
801                 old=$1
802                 sub=$2
803                 if test "$sub" = "$rev"
804                 then
805                         say "Subtree is already at commit $rev."
806                         exit 0
807                 fi
808                 new=$(new_squash_commit "$old" "$sub" "$rev") || exit $?
809                 debug "New squash commit: $new"
810                 rev="$new"
811         fi
812
813         version=$(git version)
814         if test "$version" \< "git version 1.7"
815         then
816                 if test -n "$message"
817                 then
818                         git merge -s subtree --message="$message" "$rev"
819                 else
820                         git merge -s subtree "$rev"
821                 fi
822         else
823                 if test -n "$message"
824                 then
825                         git merge -Xsubtree="$prefix" \
826                                 --message="$message" "$rev"
827                 else
828                         git merge -Xsubtree="$prefix" $rev
829                 fi
830         fi
831 }
832
833 cmd_pull () {
834         if test $# -ne 2
835         then
836                 die "You must provide <repository> <ref>"
837         fi
838         ensure_clean
839         ensure_valid_ref_format "$2"
840         git fetch "$@" || exit $?
841         revs=FETCH_HEAD
842         set -- $revs
843         cmd_merge "$@"
844 }
845
846 cmd_push () {
847         if test $# -ne 2
848         then
849                 die "You must provide <repository> <ref>"
850         fi
851         ensure_valid_ref_format "$2"
852         if test -e "$dir"
853         then
854                 repository=$1
855                 refspec=$2
856                 echo "git push using: " "$repository" "$refspec"
857                 localrev=$(git subtree split --prefix="$prefix") || die
858                 git push "$repository" "$localrev":"refs/heads/$refspec"
859         else
860                 die "'$dir' must already exist. Try 'git subtree add'."
861         fi
862 }
863
864 "cmd_$command" "$@"