Separate the idea of an 'used' and an 'applied' line. Applied lines are the ones that have been moved to modify some commit whilst used lines are the ones for which we have tried to find modifications. Applied lines are given priority when looking for modifications. Thus, if one has these commits
944f57d fixup! squash! six
3512067 fixup! squash! squash! six
49af780 fixup! fixup! one
df6f708 fixup! one
6ac57d9 one
6e754da squash! squash! six
733844b squash! six
e9de3cc six
76990c9 fivethey will be correctly reordered to
pick e9de3cc six
squash 733844b squash! six
squash 6e754da squash! squash! six
fixup 944f57d fixup! squash! six
fixup 3512067 fixup! squash! squash! six
pick 6ac57d9 one
fixup df6f708 fixup! one
fixup 49af780 fixup! fixup! one
---To be honest, I'm not sure if we want to support something like this, but it seemed too crazy not to try.
This is an attempt to fix http://thread.gmane.org/gmane.comp.version-control.git/170883 (rebase autosquash doesn't recognise a chain of fixups, 2011-04-05)
Cheers, cmn
git-rebase--interactive.sh | 36 +++++++++++++++++++++++++++++++----- 1 files changed, 31 insertions(+), 5 deletions(-)
Show changes to git-rebase--interactive.sh +31 −5
diff --git a/git-rebase--interactive.sh b/git-rebase--interactive.sh index 5873ba4..5e7b393 100755 --- a/git-rebase--interactive.sh +++ b/git-rebase--interactive.sh @@ -692,16 +692,28 @@ rearrange_squash () { test -s "$1.sq" || return used= + applied= + RE="$1.re" + skiplines=0 + # Initialize the list with the first line from the input + head -n 1 "$1" > $RE while read -r pick sha1 message do + echo "dealing with $sha1 skiplines=$skiplines" >> "rebase.log" case " $used" in - *" $sha1 "*) continue ;; + *" $sha1 "*) next_input_line "$skiplines" "$1" "$RE"; skiplines="$(expr $skiplines + 1)"; continue ;; + esac + # If it's been applied, we don't want to output the "pick" line + case " $applied" in + *" $sha1 "*) ;; + *) printf '%s\n' "$pick $sha1 $message";; esac - printf '%s\n' "$pick $sha1 $message" used="$used$sha1 " + emitted=0; while read -r squash action msg do - case " $used" in + # Ignore the commits we've already dealt with + case " $used $applied " in *" $squash "*) continue ;; esac emit=0 @@ -715,15 +727,29 @@ rearrange_squash () { case "$message" in "$msg"*) emit=1;; esac ;; esac if test $emit = 1; then + emitted=1 printf '%s\n' "$action $squash $action! $msg" - used="$used$squash " + applied="$applied$squash " + echo "$action $squash $action! $msg" >> "$RE" fi done <"$1.sq" - done >"$1.rearranged" <"$1" + + # If no lines were emitted, it means we don't need take an + # early look to any more lines, so put the next one from the input into the queue + if test "$emmited" != 1; then + next_input_line "$skiplines" "$1" "$RE" + skiplines="$(expr $skiplines + 1)" + fi + + done >"$1.rearranged" <"$RE" cat "$1.rearranged" >"$1" rm -f "$1.sq" "$1.rearranged" } +next_input_line () { + tail --lines="+$1" "$2" | head -n 1 >> "$3" +} + LF=' ' parse_onto () {
-- 1.7.4.2.437.g4fc7e.dirty