Reduce the number of connects when fetching
[gitweb.git] / git-filter-branch.sh
index d77902d34d63583122eb76fda5d28ee7eeccd77d..ff716cabb084526ddb5655c3b7c6dd02f4ed1b39 100755 (executable)
@@ -8,11 +8,9 @@
 # a new branch. You can specify a number of filters to modify the commits,
 # files and trees.
 
-set -e
-
-USAGE="git-filter-branch [-d TEMPDIR] [FILTERS] DESTBRANCH [REV-RANGE]"
-. git-sh-setup
+# The following functions will also be available in the commit filter:
 
+functions=$(cat << \EOF
 warn () {
         echo "$*" >&2
 }
@@ -28,6 +26,20 @@ map()
        fi
 }
 
+# if you run 'skip_commit "$@"' in a commit filter, it will print
+# the (mapped) parents, effectively skipping the commit.
+
+skip_commit()
+{
+       shift;
+       while [ -n "$1" ];
+       do
+               shift;
+               map "$1";
+               shift;
+       done;
+}
+
 # override die(): this version puts in an extra line break, so that
 # the progress is still visible
 
@@ -37,6 +49,10 @@ die()
        echo "$*" >&2
        exit 1
 }
+EOF
+)
+
+eval "$functions"
 
 # When piped a commit, output a script to set the ident of either
 # "author" or "committer
@@ -50,17 +66,17 @@ set_ident () {
                        h
                        s/^'$lid' \([^<]*\) <[^>]*> .*$/\1/
                        s/'\''/'\''\'\'\''/g
-                       s/.*/export GIT_'$uid'_NAME='\''&'\''/p
+                       s/.*/GIT_'$uid'_NAME='\''&'\''; export GIT_'$uid'_NAME/p
 
                        g
                        s/^'$lid' [^<]* <\([^>]*\)> .*$/\1/
                        s/'\''/'\''\'\'\''/g
-                       s/.*/export GIT_'$uid'_EMAIL='\''&'\''/p
+                       s/.*/GIT_'$uid'_EMAIL='\''&'\''; export GIT_'$uid'_EMAIL/p
 
                        g
                        s/^'$lid' [^<]* <[^>]*> \(.*\)$/\1/
                        s/'\''/'\''\'\'\''/g
-                       s/.*/export GIT_'$uid'_DATE='\''&'\''/p
+                       s/.*/GIT_'$uid'_DATE='\''&'\''; export GIT_'$uid'_DATE/p
 
                        q
                }
@@ -68,9 +84,23 @@ set_ident () {
 
        LANG=C LC_ALL=C sed -ne "$pick_id_script"
        # Ensure non-empty id name.
-       echo "[ -n \"\$GIT_${uid}_NAME\" ] || export GIT_${uid}_NAME=\"\${GIT_${uid}_EMAIL%%@*}\""
+       echo "case \"\$GIT_${uid}_NAME\" in \"\") GIT_${uid}_NAME=\"\${GIT_${uid}_EMAIL%%@*}\" && export GIT_${uid}_NAME;; esac"
 }
 
+USAGE="[--env-filter <command>] [--tree-filter <command>] \
+[--index-filter <command>] [--parent-filter <command>] \
+[--msg-filter <command>] [--commit-filter <command>] \
+[--tag-name-filter <command>] [--subdirectory-filter <directory>] \
+[--original <namespace>] [-d <directory>] [-f | --force] \
+[<rev-list options>...]"
+
+OPTIONS_SPEC=
+. git-sh-setup
+
+git diff-files --quiet &&
+       git diff-index --cached --quiet HEAD -- ||
+       die "Cannot rewrite branch(es) with a dirty working directory."
+
 tempdir=.git-rewrite
 filter_env=
 filter_tree=
@@ -80,13 +110,20 @@ filter_msg=cat
 filter_commit='git commit-tree "$@"'
 filter_tag_name=
 filter_subdir=
-while case "$#" in 0) usage;; esac
+orig_namespace=refs/original/
+force=
+while :
 do
        case "$1" in
        --)
                shift
                break
                ;;
+       --force|-f)
+               shift
+               force=t
+               continue
+               ;;
        -*)
                ;;
        *)
@@ -120,7 +157,7 @@ do
                filter_msg="$OPTARG"
                ;;
        --commit-filter)
-               filter_commit="$OPTARG"
+               filter_commit="$functions; $OPTARG"
                ;;
        --tag-name-filter)
                filter_tag_name="$OPTARG"
@@ -128,39 +165,67 @@ do
        --subdirectory-filter)
                filter_subdir="$OPTARG"
                ;;
+       --original)
+               orig_namespace=$(expr "$OPTARG/" : '\(.*[^/]\)/*$')/
+               ;;
        *)
                usage
                ;;
        esac
 done
 
-dstbranch="$1"
-shift
-test -n "$dstbranch" || die "missing branch name"
-git show-ref "refs/heads/$dstbranch" 2> /dev/null &&
-       die "branch $dstbranch already exists"
-
-test ! -e "$tempdir" || die "$tempdir already exists, please remove it"
-mkdir -p "$tempdir/t"
-cd "$tempdir/t"
-workdir="$(pwd)"
-
-case "$GIT_DIR" in
-/*)
+case "$force" in
+t)
+       rm -rf "$tempdir"
+;;
+'')
+       test -d "$tempdir" &&
+               die "$tempdir already exists, please remove it"
+esac
+mkdir -p "$tempdir/t" &&
+tempdir="$(cd "$tempdir"; pwd)" &&
+cd "$tempdir/t" &&
+workdir="$(pwd)" ||
+die ""
+
+# Remove tempdir on exit
+trap 'cd ../..; rm -rf "$tempdir"' 0
+
+# Make sure refs/original is empty
+git for-each-ref > "$tempdir"/backup-refs
+while read sha1 type name
+do
+       case "$force,$name" in
+       ,$orig_namespace*)
+               die "Namespace $orig_namespace not empty"
        ;;
-*)
-       GIT_DIR="$(pwd)/../../$GIT_DIR"
+       t,$orig_namespace*)
+               git update-ref -d "$name" $sha1
        ;;
-esac
-export GIT_DIR GIT_WORK_TREE=.
+       esac
+done < "$tempdir"/backup-refs
 
-export GIT_INDEX_FILE="$(pwd)/../index"
-git read-tree # seed the index file
+ORIG_GIT_DIR="$GIT_DIR"
+ORIG_GIT_WORK_TREE="$GIT_WORK_TREE"
+ORIG_GIT_INDEX_FILE="$GIT_INDEX_FILE"
+GIT_WORK_TREE=.
+export GIT_DIR GIT_WORK_TREE
 
-ret=0
+# The refs should be updated if their heads were rewritten
+git rev-parse --no-flags --revs-only --symbolic-full-name --default HEAD "$@" |
+sed -e '/^^/d' >"$tempdir"/heads
 
+test -s "$tempdir"/heads ||
+       die "Which ref do you want to rewrite?"
 
-mkdir ../map # map old->new commit ids for rewriting parents
+GIT_INDEX_FILE="$(pwd)/../index"
+export GIT_INDEX_FILE
+git read-tree || die "Could not seed the index"
+
+ret=0
+
+# map old->new commit ids for rewriting parents
+mkdir ../map || die "Could not create map/ directory"
 
 case "$filter_subdir" in
 "")
@@ -170,11 +235,13 @@ case "$filter_subdir" in
 *)
        git rev-list --reverse --topo-order --default HEAD \
                --parents --full-history "$@" -- "$filter_subdir"
-esac > ../revs
-commits=$(cat ../revs | wc -l | tr -d " ")
+esac > ../revs || die "Could not get the commits"
+commits=$(wc -l <../revs | tr -d " ")
 
 test $commits -eq 0 && die "Found nothing to rewrite"
 
+# Rewrite the commits
+
 i=0
 while read commit parents; do
        i=$(($i+1))
@@ -186,10 +253,12 @@ while read commit parents; do
                ;;
        *)
                git read-tree -i -m $commit:"$filter_subdir"
-       esac
+       esac || die "Could not initialize the index"
 
-       export GIT_COMMIT=$commit
-       git cat-file commit "$commit" >../commit
+       GIT_COMMIT=$commit
+       export GIT_COMMIT
+       git cat-file commit "$commit" >../commit ||
+               die "Cannot read commit $commit"
 
        eval "$(set_ident AUTHOR <../commit)" ||
                die "setting author failed for commit $commit"
@@ -199,14 +268,15 @@ while read commit parents; do
                die "env filter failed: $filter_env"
 
        if [ "$filter_tree" ]; then
-               git checkout-index -f -u -a
+               git checkout-index -f -u -a ||
+                       die "Could not checkout the index"
                # files that $commit removed are now still in the working tree;
                # remove them, else they would be added again
                git ls-files -z --others | xargs -0 rm -f
                eval "$filter_tree" < /dev/null ||
                        die "tree filter failed: $filter_tree"
 
-               git diff-index -r $commit | cut -f 2- | tr '\n' '\0' | \
+               git diff-index -r $commit | cut -f 2- | tr '\012' '\000' | \
                        xargs -0 git update-index --add --replace --remove
                git ls-files -z --others | \
                        xargs -0 git update-index --add --replace --remove
@@ -233,21 +303,73 @@ while read commit parents; do
                $(git write-tree) $parentstr < ../message > ../map/$commit
 done <../revs
 
-src_head=$(tail -n 1 ../revs | sed -e 's/ .*//')
-target_head=$(head -n 1 ../map/$src_head)
-case "$target_head" in
-'')
-       echo Nothing rewritten
+# In case of a subdirectory filter, it is possible that a specified head
+# is not in the set of rewritten commits, because it was pruned by the
+# revision walker.  Fix it by mapping these heads to the next rewritten
+# ancestor(s), i.e. the boundaries in the set of rewritten commits.
+
+# NEEDSWORK: we should sort the unmapped refs topologically first
+while read ref
+do
+       sha1=$(git rev-parse "$ref"^0)
+       test -f "$workdir"/../map/$sha1 && continue
+       # Assign the boundarie(s) in the set of rewritten commits
+       # as the replacement commit(s).
+       # (This would look a bit nicer if --not --stdin worked.)
+       for p in $( (cd "$workdir"/../map; ls | sed "s/^/^/") |
+               git rev-list $ref --boundary --stdin |
+               sed -n "s/^-//p")
+       do
+               map $p >> "$workdir"/../map/$sha1
+       done
+done < "$tempdir"/heads
+
+# Finally update the refs
+
+_x40='[0-9a-f][0-9a-f][0-9a-f][0-9a-f][0-9a-f]'
+_x40="$_x40$_x40$_x40$_x40$_x40$_x40$_x40$_x40"
+echo
+while read ref
+do
+       # avoid rewriting a ref twice
+       test -f "$orig_namespace$ref" && continue
+
+       sha1=$(git rev-parse "$ref"^0)
+       rewritten=$(map $sha1)
+
+       test $sha1 = "$rewritten" &&
+               warn "WARNING: Ref '$ref' is unchanged" &&
+               continue
+
+       case "$rewritten" in
+       '')
+               echo "Ref '$ref' was deleted"
+               git update-ref -m "filter-branch: delete" -d "$ref" $sha1 ||
+                       die "Could not delete $ref"
        ;;
-*)
-       git update-ref refs/heads/"$dstbranch" $target_head
-       if [ $(cat ../map/$src_head | wc -l) -gt 1 ]; then
-               echo "WARNING: Your commit filter caused the head commit to expand to several rewritten commits. Only the first such commit was recorded as the current $dstbranch head but you will need to resolve the situation now (probably by manually merging the other commits). These are all the commits:" >&2
-               sed 's/^/       /' ../map/$src_head >&2
-               ret=1
-       fi
+       $_x40)
+               echo "Ref '$ref' was rewritten"
+               git update-ref -m "filter-branch: rewrite" \
+                               "$ref" $rewritten $sha1 ||
+                       die "Could not rewrite $ref"
        ;;
-esac
+       *)
+               # NEEDSWORK: possibly add -Werror, making this an error
+               warn "WARNING: '$ref' was rewritten into multiple commits:"
+               warn "$rewritten"
+               warn "WARNING: Ref '$ref' points to the first one now."
+               rewritten=$(echo "$rewritten" | head -n 1)
+               git update-ref -m "filter-branch: rewrite to first" \
+                               "$ref" $rewritten $sha1 ||
+                       die "Could not rewrite $ref"
+       ;;
+       esac
+       git update-ref -m "filter-branch: backup" "$orig_namespace$ref" $sha1
+done < "$tempdir"/heads
+
+# TODO: This should possibly go, with the semantics that all positive given
+#       refs are updated, and their original heads stored in refs/original/
+# Filter tags
 
 if [ "$filter_tag_name" ]; then
        git for-each-ref --format='%(objectname) %(objecttype) %(refname)' refs/tags |
@@ -266,7 +388,8 @@ if [ "$filter_tag_name" ]; then
 
                [ -f "../map/$sha1" ] || continue
                new_sha1="$(cat "../map/$sha1")"
-               export GIT_COMMIT="$sha1"
+               GIT_COMMIT="$sha1"
+               export GIT_COMMIT
                new_ref="$(echo "$ref" | eval "$filter_tag_name")" ||
                        die "tag name filter failed: $filter_tag_name"
 
@@ -277,12 +400,22 @@ if [ "$filter_tag_name" ]; then
                        warn "unreferencing tag object $sha1t"
                fi
 
-               git update-ref "refs/tags/$new_ref" "$new_sha1"
+               git update-ref "refs/tags/$new_ref" "$new_sha1" ||
+                       die "Could not write tag $new_ref"
        done
 fi
 
 cd ../..
 rm -rf "$tempdir"
-printf "\nRewritten history saved to the $dstbranch branch\n"
+
+trap - 0
+
+unset GIT_DIR GIT_WORK_TREE GIT_INDEX_FILE
+test -z "$ORIG_GIT_DIR" || GIT_DIR="$ORIG_GIT_DIR" && export GIT_DIR
+test -z "$ORIG_GIT_WORK_TREE" || GIT_WORK_TREE="$ORIG_GIT_WORK_TREE" &&
+       export GIT_WORK_TREE
+test -z "$ORIG_GIT_INDEX_FILE" || GIT_INDEX_FILE="$ORIG_GIT_INDEX_FILE" &&
+       export GIT_INDEX_FILE
+git read-tree -u -m HEAD
 
 exit $ret