@matheusbsbs / matheus-bagliano / commits / aaf962df4a

Reintegrate: cope with delete-modify conflicts

A more proper solution is probably to teach "rerere" to remember what modification (probably indexed via a patch ID for flexibility, but either the pre- or post-image blob ID without such flexibility is probably good enough) was deemed "uninteresting" to resolve in favor of deletion and reapply it when we see another delete-modify conflict with the same modification, but for now, treat all delete-modify conflict to be with uninteresting modification. The way the Reintegrate machinery is used makes it somewhat safer than it looks---the initial attempt to merge (and come up with a merge-fix if needed) will always be done using "git merge" and the resulting commit is manually examined (if only to see if there is a need for a further merge-fix) before redo-foo script is written out of the resulting history. We'll need to deal with a case where a modifying side need to be kept over the deleting side if/when it arises, but so far I haven't seen such a case in real life.

Junio C Hamano committed Dec 11, 2023 at 12:40 UTC aaf962df4a807886abb21c1a95ba6652a64943eb

1 file changed +21

Reintegrate

+21

@@ -123,6 +123,27 @@ cocci_mark="treewide: apply cocci patch"

123 case "$generate" in

124 no)

125 accept_rerere () {

126 + git ls-files -u -z |

127 + perl -0 -e '

128 + my %path_stage = ();

129 + my @to_remove = ();

130 + while (<>) {

131 + my ($mode, $sha1, $stage, $path) =

132 + /^([0-7]+) ([0-9a-f]+) ([0-3]) (.*)$/;

133 + $path_stage{$path} ||= 0;

134 + $path_stage{$path} |= (1 << ($stage - 1));

135 + }

136 +

137 + while (my ($path, $bits) = each %path_stage) {

138 + if ($bits == 3 || $bits == 5) {

139 + push @to_remove, $path;

140 + }

141 + }

142 + if (@to_remove) {

143 + system(qw(git rm -f), @to_remove);

144 + }

145 + '

146 +

147 if ! git write-tree 2>/dev/null >/dev/null

148 then

149 git rerere remaining