[PATCH aiaiai 7/7] git-find-base: rewritten to use newer design
Jacob Keller
jacob.e.keller at intel.com
Fri Apr 4 15:06:52 PDT 2014
Instead of just checking a few extra headers, add support for all header
types that git can output, and be a bit more robust about renames and
deletes. For now, assume a rename won't attempt to overwrite an existing
file.. Ideally we should do a full check on the base as well..
This model should allow support of all git commit types, not just simple
ones. In addition, in order to allow a certain patch format, we add an
option to remove duplicates. Do it by default, but enable option to keep
duplicate diff chunks. Primarily useful for patches which contain both
an inline and attachment.
Signed-off-by: Jacob Keller <jacob.e.keller at intel.com>
---
helpers/git-find-base | 300 ++++++++++++++++++++++++++++++++------------------
1 file changed, 190 insertions(+), 110 deletions(-)
diff --git a/helpers/git-find-base b/helpers/git-find-base
index cf4b9393cdf9..db7900ff1ec1 100755
--- a/helpers/git-find-base
+++ b/helpers/git-find-base
@@ -55,157 +55,237 @@ standard out will be a single commit id. If nothing was found, no standard
output will be generated, and this utility will exit with a non-zero exit code.
Options:
+ -k, --keep Keep duplicate diff chunks.
-?, -h Show this text and exit.
END
}
-# subroutine to check whether two blob indexes match, (ie: one
-# contains the other regardless of which one is larger)
sub match_index {
- my ( $x, $y ) = @_;
+ my ($x, $y) = @_;
- my $lx = length $x;
- my $ly = length $y;
-
- # Find which length is shortest
- my $l = $lx >= $ly ? $ly : $lx;
-
- # Truncate the indexes to the shortest
- my $tx = substr $lx,0,$l;
- my $ty = substr $ly,0,$l;
-
- # Return the match
- return $tx == $ty;
+ return ( index $x,$y ) == 0 or ( index $y,$x ) == 0;
}
+sub hash_comp(\%\%) {
+ my %x = %{ shift @_ };
+ my %y = %{ shift @_ };
+
+ ( grep { not ( ( exists $y{$_} ) and $x{$_} eq $y{$_} ) } keys %x ) == 0;
+}
+
+sub path_exists(\%$) {
+ my %tree = %{ shift @_ };
+ my $path = shift @_;
+
+ return exists $tree{$path} and $tree{$path}->{status} eq "";
+}
+
+my $duplicates = '';
+
Getopt::Long::Configure("pass_through");
-GetOptions('h|?' => sub { show_usage; exit 0; });
+GetOptions('h|?' => sub { show_usage; exit 0; },
+ 'keep!' => \$duplicates );
# Slurp the contents into $mbox for processing
my $mbox = do { local $/; <STDIN> };
-# Hash of file-index relations
-my %files = ();
+# Array of hrefs to chunk contexts
+my @chunks = ();
-# Split mbox apart by diff lines, preserving the filename we matched against,
-# as well as the full index line. This should handle even the rename case from
-# git diff output. Note, we assume that mbox has correct ordering of patches.
-while ($mbox =~ /^diff --git [iwcoab]\/(?<oldfile>\S+) [iwcoab]\/(?<newfile>\S+)\n(?<new>new file mode [0-7]+\n)?(?<rename>^similarity index .*\n)?(?<from>^rename from \g{oldfile}\n)?(?<to>^rename to \g{newfile}\n)?(?<index>^index .*$)?\n/gm) {
- my $file = $+{oldfile};
- my $rename = $+{similarity};
- my $new = $+{new};
- my $index = $+{index};
- $file or die "Could not parse file from diff context.";
+# The possible list of extended headers supported by git-diff output
+my $extended_headers = qr/(old mode|new mode|deleted file mode|new file mode|copy from|copy to|rename from|rename to|similarity index|dissimilarity index|index)/;
- # If we get a rename without an index, simply note that a file was renamed,
- # and ignore it, since there were no real changes.
- if ( $rename and not $index ) {
- print STDERR "Found rename of $file\n";
- next;
- }
+# Split mbox apart by diff header chunks, finding a diff line followed by any number of extended header lines
+while ($mbox =~ /^(?<chunk>diff (?s:.*?))(?=^(?!$extended_headers))/gm) {
- # Check the index line for proper formatting.
- $index =~ /^index ([0-9a-f]+)[.]{2}([0-9a-f]+) [0-7]{6}$/;
- my $initialshortblob = $1;
- my $modifiedshortblob = $2;
- $initialshortblob or die "Could not parse short blob index from diff context. Is the mbox corrupted?";
+ # Capture the block
+ my $rawchunk = $+{chunk};
- # If we have a new file, store the initial setting as "new", and keep the
- # modified blob for checking future changes in this series.
- if ($new) {
- print STDERR "Found new file at $file\n";
- $files{$file}{"initial"} = "new";
- $files{$file}{"modified"} = $modifiedshortblob;
- next;
- };
+ print STDERR "Found a diff chunk\n";
+ print STDERR $rawchunk;
- # If we already have this file, simply update the modified blob index
- if (exists $files{$file}) {
- # Check if the blob matches the last known result of the file
- if (match_index($initialshortblob, $files{$file}{"modified"})) {
- print STDERR "Found further modification of $file, ($initialshortblob -> $modifiedshortblob).\n";
- $files{$file}{"modified"} = $modifiedshortblob;
- next;
- } elsif (match_index($modifiedshortblob, $files{$file}{"modified"}) and match_index($initialshortblob, $files{$file}{"initial"})) {
- print STDERR "Found duplicate modification of $file. Possible duplicate patch blob, or an incorrect patch format? Ignoring for now.\n";
- } else {
- die "Found futher modification of $file that does not match expected index, ($initialshortblob -> $modifiedshortblob). Is the patch sequence out of order?";
+ # Check whether it has expected format
+ if ( $rawchunk =~ /^diff --git [iwcoab]\/(?<oldpath>\S+) [iwcoab]\/(?<newpath>\S+)$/m ) {
+ # We have a standard git diff chunk. Now, we need to parse the extended
+ # headers from the section.
+
+ my %chunk = ();
+ $chunk{oldpath} = $+{oldpath};
+ $chunk{newpath} = $+{newpath};
+ $chunk{oldindex} = "";
+ $chunk{newindex} = "";
+ $chunk{action} = "none";
+
+ if ( $rawchunk =~ /^index (?<oldindex>[0-9a-fA-F]+)[.]{2}(?<newindex>[0-9a-fA-F]+)( (?<mode>[0-7]{6}))?$/m ) {
+ $chunk{oldindex} = $+{oldindex};
+ $chunk{newindex} = $+{newindex};
+ $chunk{oldmode} = $+{mode};
+ $chunk{newmode} = $+{mode};
}
+
+
+ if ( $rawchunk =~ /^old mode (?<mode>[0-7]{6})$/m ) {
+ $chunk{oldmode} = $+{mode};
+ }
+
+ if ( $rawchunk =~ /^new mode (?<mode>[0-7]{6})$/m ) {
+ $chunk{newmode} = $+{mode};
+ }
+
+ if ( $rawchunk =~ /^deleted file mode (?<mode>[0-7]{6})$/m ) {
+ $chunk{oldmode} = $+{mode};
+ $chunk{action} = "delete";
+ }
+
+ if ( $rawchunk =~ /^new file mode (?<mode>[0-7]{6})$/m ) {
+ $chunk{newmode} = $+{mode};
+ $chunk{action} = "create";
+ }
+
+ if ( $rawchunk =~ /^rename from \Q$chunk{oldpath}\E$/m ) {
+ $chunk{action} = "rename";
+ }
+
+ if ( $rawchunk =~ /^rename to \Q$chunk{newpath}\E$/m ) {
+ $chunk{action} = "rename";
+ }
+
+ if ( $rawchunk =~ /^similarity index (?<similarity>[0-9]{1,3}%)$/m ) {
+ $chunk{similarity} = $+{similarity};
+ }
+
+ if ( $rawchunk =~ /^dissimilarity index (?<dissimilarity>[0-9]{1,3}%)$/m ) {
+ $chunk{similarity} = 100 - $+{dissimilarity};
+ }
+
+ if ( not $duplicates and ( grep { hash_comp ( %$_, %chunk ) } @chunks ) > 0 ) {
+ print STDERR "Skipping duplicate diff chunk. Disable this behavior with --keep.\n";
+ } else {
+ push (@chunks, \%chunk);
+ }
+
+ } elsif ( $rawchunk =~ /^diff --(combined|cc) (?<newfile>\S+)$/m ) {
+ # We can't use combined diff formats, since these are used for multiple
+ # parents, and are not suitable for this process
+ print STDERR "Found a combined diff format, indicating a merge. We can't find a base commit for a merge!\n";
+ exit 1;
+ } else {
+ # Non git-formats are not supported, as we need the index information
+ print STDERR "Found a diff chunk, but it does not have a recognized format.\n";
+ exit 1;
}
-
- print STDERR "Found modification to $file, ($initialshortblob -> $modifiedshortblob).\n";
-
- # We have to process the short blob index into a full index value using
- # git-rev-parse, otherwise the lookup will fail.
- open my $rev_parse, '-|', 'git' => 'rev-parse' => '--verify', $initialshortblob
- or die "Couldn't open pipe to git-rev-parse: ", $!;
-
- my $initialblob = <$rev_parse>;
- close $rev_parse or die "Couldn't expand the blob index: ", $? >> 8;
- chomp $initialblob;
-
- # Store the initial blob, as well as the index after modification
- $files{$file}{"initial"} = $initialblob;
- $files{$file}{"modified"} = $modifiedshortblob;
}
-# Subroutine to check a commit treeish, ensuring that every blob is present at
-# the correct path. This allows us to determine whether the commit is "good",
-# ie: has all the blobs required to cleanly apply the patch, or not.
+# We have collated all the chunks. Now we need to loop over a series of commits
+# based on user input. For each commit, we will try to build up the list of
+# changes and see if it is applicable.
sub check_commit {
my ( $commit ) = @_;
- # Loop through every blob/path combination from the mbox, and check if the
- # ls-tree on that path matches the blob we need.
- for my $path ( keys %files) {
- my $blob = $files{$path}{"initial"};
+ # Our current view of the tree
+ my %tree = ();
- # We shouldn't try to find a new file, as it won't exist yet
- continue if $blob eq "new";
+ # For each chunk, we need to build up the tree. looking up from git-ls-tree
+ # for the first time we find a path. We want to see if our patch could cleanly apply to the given commit.
+ for my $chunk ( @chunks ) {
- # Fail with die on the pipe since this should always work.
- open my $ls_tree, '-|', 'git' => 'ls-tree' => '--full-tree' => $commit => '--', $path
- or die "Couldn't open pipe to git-ls-tree: ", $!;
+ # If the path doesn't exist yet, just fill in some information about it
+ # from the real tree
+ if ( not exists $tree{$chunk->{oldpath}} ) {
+ open my $ls_tree, '-|', 'git', => 'ls-tree' => '--full-tree' => $commit => '--' => $chunk->{oldpath}
+ or die "Couldn't open pipe to git-ls-tree: ", $!;
- # Return here if we fail to find the file, because it might not yet
- # exist.
- my $tree = <$ls_tree>;
- close $ls_tree or do {
- print STDERR "Couldn't find matching tree: ", $? >> 8;
- return;
- };
- chomp $tree;
+ my $ls_tree_output = <$ls_tree>;
+ close $ls_tree or do {
+ print STDERR "git-ls-tree failed: ", $? >> 8;
+ return 0;
+ };
- # Check the output formatting to ensure we didn't get any errors
- $tree =~ /\A[0-7]{6} (\S+) (\S+)/ or do {
- print STDERR "Unexpected git-ls-tree output.\n";
- return;
- };
+ # Only add the tree object if we actually have output
+ if ( defined $ls_tree_output ) {
+ chomp $ls_tree_output;
+ $ls_tree_output =~ /\A([0-7]{6}) (blob|tree|commit) (\S+)/ or do {
+ print STDERR "Unexpected git-ls-tree output.\n";
+ return 0;
+ };
- # Return undef if they don't match. This will ensure we bail at the
- # first conflicting blob, without forcing extra checks.
- return if $2 ne $blob;
+ $tree{$chunk->{oldpath}} = {
+ mode => $1,
+ index => $3,
+ status => "",
+ };
+ }
+ }
+
+ # We have now added any known information about this path to the tree.
+ # We will now attempt to modify the tree based on the contents of the
+ # chunk.
+
+ if ( $chunk->{action} eq "create" ) {
+ if ( path_exists( %tree, $chunk->{oldpath} ) ) {
+ # This path already exists, so we can't add it!
+ print STDERR "$chunk->{oldpath} already exists.\n";
+ return 0;
+ } else {
+ # We found a patch that either doesn't exist, or is already
+ # been renamed or deleted. We can simply add it here now.
+ $tree{$chunk->{oldpath}}->{mode} = $chunk->{mode};
+ $tree{$chunk->{oldpath}}->{index} = $chunk->{newindex};
+ $tree{$chunk->{oldpath}}->{status} = "";
+ }
+ } else {
+ if ( not path_exists( %tree, $chunk->{oldpath} ) ) {
+ # This path no longer exists, we can't modify it.
+ print STDERR "$chunk->{oldpath} does not exist.\n";
+ return 0;
+ } else {
+ if ( not match_index( $tree{$chunk->{oldpath}}->{index}, $chunk->{oldindex} ) ) {
+ print STDERR "$chunk->{oldpath} does not have matching index.\n";
+ return 0;
+ }
+
+ if ( $chunk->{newindex} ) {
+ $tree{$chunk->{oldpath}}->{index} = $chunk->{newindex};
+ }
+
+ if ( $chunk->{newmode} ) {
+ $tree{$chunk->{oldpath}}->{mode} = $chunk->{newmode};
+ }
+
+ # Handle special case here for rename and delete actions
+ if ( $chunk->{action} eq "rename" ) {
+ if ( path_exists( %tree, $chunk->{newpath} ) ) {
+ print STDERR "$chunk->{newpath} already exists.\n";
+ return 0;
+ }
+
+ $tree{$chunk->{newpath}} = $tree{$chunk->{oldpath}};
+ $tree{$chunk->{oldpath}}->{status} = "renamed";
+ } elsif ( $chunk->{action} eq "delete" ) {
+ $tree{$chunk->{oldpath}}->{status} = "deleted";
+ }
+ }
+ }
}
- # If we get here, then everything matched above, so we can return true.
+ # If we get here, that means we had no issues verifying each chunk, and we
+ # can exit true.
return 1;
}
-# Open the log pipe. Pass all of our ARGV directly to the log command
-open my $log, '-|', git => log => @ARGV, '--pretty=format:%T %H'
+# Open the git-log pipe. Pass all of our ARGV directly to the rev-list command.
+open my $log, '-|', 'git' => 'log' => @ARGV => '--pretty=%H'
or die "Couldn't open pipe to git-log: ", $!;
-# Loop through each commit in the log, checking if it's tree and hash have all
-# the valid blobs. User can easily modify the log command via options to limit
-# the scope, or reverse ordering. By default we find the most recent commit
-# which has the required blobs.
+# Loop through each commit in the list, checking if the diff chunks can apply
+# cleanly to the commit. Easily allow modifying which commits are checked via
+# options to the git-log command, which allows limiting what can be checked.
while ( <$log> ) {
chomp;
- my ($tree, $commit) = split " ", $_;
- if (check_commit $commit) {
+ if (check_commit $_) {
# Print the commit hash we found, and exit with a good return status.
- print "$commit\n";
+ print "$_\n";
exit 0;
}
}
--
1.8.3.1
More information about the aiaiai
mailing list