threads / patch / 30147

v11, 9 partsdifftool: teach difftool to handle directory diffs

Subject: [PATCH 8/9 v11] difftool: teach difftool to handle directory diffs

## tl;dr

9 messages between Apr 4, 2012 and Apr 12, 2012. Diffs are folded; open one to read it.

replies: 8people: 3as markdown or json

Tim Henigan· Apr 4, 2012, 19:21 UTC · lore

When 'difftool' is called to compare a range of commits that modify more than one file, it opens a separate instance of the diff tool for each file that changed.

The new '--dir-diff' option copies all the modified files to a temporary location and runs a directory diff on them in a single instance of the diff tool.

Signed-off-by: Tim Henigan <tim.henigan@gmail.com>
---

This replaces commit 4d6efe0 in the 'th/difftool-diffall' branch based on pu. This version corrects some problems with submodule support in difftool.

Changes in v11:
  - Fix handling of submodules in subdirectories.
    
    If the diff includes a submodule that lives in a subdirectory of
    the project, the script fails with "No such file or directory".
    For example, if a submodule is located at '<repo root>/lib/sub',
    the script fails when it tries to write the submodule SHA1 to
    '/tmp/.../lib/sub' because the 'lib' directory does not exist.
    This commit teaches difftool to create the subdirectory path in
    the tmp location prior to writing the submodule file.
  - Append "dirty" to submodule SHA1 when needed
    
    If a submodule is dirty (i.e. has uncommitted changes) then
      'git diff' reports "<SHA1 of parent>-dirty"
    while
      'git diff --raw' reports "<SHA1 of parent>"
    difftool uses the '--raw' diff format, so the "dirty" flag is lost.
    To make the output more closely match that of 'git diff', teach
    difftool to append "-dirty" to the submodule SHA1 when needed.
 Documentation/git-difftool.txt |    6 ++
 git-difftool--helper.sh        |   19 +++--
 git-difftool.perl              |  170 +++++++++++++++++++++++++++++++++++-----
 t/t7800-difftool.sh            |   39 +++++++++
 4 files changed, 209 insertions(+), 25 deletions(-)
Show changes to 4 files +209 −25

Documentation/git-difftool.txt, git-difftool--helper.sh, git-difftool.perl, t/t7800-difftool.sh

diff --git a/Documentation/git-difftool.txt b/Documentation/git-difftool.txt
index fe38f66..aba5e76 100644
--- a/Documentation/git-difftool.txt
+++ b/Documentation/git-difftool.txt
@@ -19,6 +19,12 @@ linkgit:git-diff[1].
 
 OPTIONS
 -------
+-d::
+--dir-diff::
+	Copy the modified files to a temporary location and perform
+	a directory diff on them. This mode never prompts before
+	launching the diff tool.
+
 -y::
 --no-prompt::
 	Do not prompt before launching a diff tool.
diff --git a/git-difftool--helper.sh b/git-difftool--helper.sh
index e6558d1..3d0fe0c 100755
--- a/git-difftool--helper.sh
+++ b/git-difftool--helper.sh
@@ -73,9 +73,16 @@ then
 	fi
 fi
 
-# Launch the merge tool on each path provided by 'git diff'
-while test $# -gt 6
-do
-	launch_merge_tool "$1" "$2" "$5"
-	shift 7
-done
+if test -n "$GIT_DIFFTOOL_DIRDIFF"
+then
+	LOCAL="$1"
+	REMOTE="$2"
+	run_merge_tool "$merge_tool" false
+else
+	# Launch the merge tool on each path provided by 'git diff'
+	while test $# -gt 6
+	do
+		launch_merge_tool "$1" "$2" "$5"
+		shift 7
+	done
+fi
diff --git a/git-difftool.perl b/git-difftool.perl
index d4fe998..5bb01e1 100755
--- a/git-difftool.perl
+++ b/git-difftool.perl
@@ -1,21 +1,29 @@
 #!/usr/bin/env perl
 # Copyright (c) 2009, 2010 David Aguilar
+# Copyright (c) 2012 Tim Henigan
 #
 # This is a wrapper around the GIT_EXTERNAL_DIFF-compatible
 # git-difftool--helper script.
 #
 # This script exports GIT_EXTERNAL_DIFF and GIT_PAGER for use by git.
-# GIT_DIFFTOOL_NO_PROMPT, GIT_DIFFTOOL_PROMPT, and GIT_DIFF_TOOL
-# are exported for use by git-difftool--helper.
+# GIT_DIFFTOOL_NO_PROMPT, GIT_DIFFTOOL_PROMPT, GIT_DIFFTOOL_DIRDIFF,
+# and GIT_DIFF_TOOL are exported for use by git-difftool--helper.
 #
 # Any arguments that are unknown to this script are forwarded to 'git diff'.
 
 use 5.008;
 use strict;
 use warnings;
+use File::Basename qw(dirname);
+use File::Copy;
+use File::stat;
+use File::Path qw(mkpath);
+use File::Temp qw(tempdir);
 use Getopt::Long qw(:config pass_through);
 use Git;
 
+my @working_tree;
+
 sub usage
 {
 	my $exitcode = shift;
@@ -24,15 +32,121 @@ usage: git difftool [-t|--tool=<tool>]
                     [-x|--extcmd=<cmd>]
                     [-g|--gui] [--no-gui]
                     [--prompt] [-y|--no-prompt]
+                    [-d|--dir-diff]
                     ['git diff' options]
 USAGE
 	exit($exitcode);
 }
 
+sub setup_dir_diff
+{
+	# Run the diff; exit immediately if no diff found
+	my $repo = Git->repository();
+	my $diffrtn = $repo->command_oneline(['diff', '--raw', '--no-abbrev', '-z', @ARGV]);
+	exit(0) if (length($diffrtn) == 0);
+
+	# Setup temp directories
+	my $tmpdir = tempdir('git-diffall.XXXXX', CLEANUP => 1, TMPDIR => 1);
+	my $ldir = "$tmpdir/left";
+	my $rdir = "$tmpdir/right";
+	mkpath($ldir) or die $!;
+	mkpath($rdir) or die $!;
+
+	# Build index info for left and right sides of the diff
+	my $submodule_mode = "160000";
+	my $null_mode = 0 x 6;
+	my $null_sha1 = 0 x 40;
+	my $lindex = "";
+	my $rindex = "";
+	my %submodule;
+	my @rawdiff = split('\0', $diffrtn);
+
+	for (my $i=0; $i<@rawdiff; $i+=2) {
+		my ($lmode, $rmode, $lsha1, $rsha1, $status) = split(' ', substr($rawdiff[$i], 1));
+		my $path = $rawdiff[$i + 1];
+
+		if (($lmode eq $submodule_mode) or ($rmode eq $submodule_mode)) {
+			$submodule{$path}{left} = $lsha1;
+			if ($lsha1 ne $rsha1) {
+				$submodule{$path}{right} = $rsha1;
+			} else {
+				$submodule{$path}{right} = "$rsha1-dirty";
+			}
+			next;
+		}
+
+		if ($lmode ne $null_mode) {
+			$lindex .= "$lmode $lsha1\t$path\0";
+		}
+
+		if ($rmode ne $null_mode) {
+			if ($rsha1 ne $null_sha1) {
+				$rindex .= "$rmode $rsha1\t$path\0";
+			} else {
+				push(@working_tree, $path);
+			}
+		}
+	}
+
+	# Populate the left and right directories based on each index file
+	my ($inpipe, $ctx);
+	$ENV{GIT_DIR} = $repo->repo_path();
+	$ENV{GIT_INDEX_FILE} = "$tmpdir/lindex";
+	($inpipe, $ctx) = $repo->command_input_pipe(qw/update-index -z --index-info/);
+	print($inpipe $lindex);
+	$repo->command_close_pipe($inpipe, $ctx);
+	system(('git', 'checkout-index', '--all', "--prefix=$ldir/"));
+
+	$ENV{GIT_INDEX_FILE} = "$tmpdir/rindex";
+	($inpipe, $ctx) = $repo->command_input_pipe(qw/update-index -z --index-info/);
+	print($inpipe $rindex);
+	$repo->command_close_pipe($inpipe, $ctx);
+	system(('git', 'checkout-index', '--all', "--prefix=$rdir/"));
+
+	# Changes in the working tree need special treatment since they are
+	# not part of the index
+	my $workdir = $repo->repo_path() . "/..";
+	for (@working_tree) {
+		my $dir = dirname($_);
+		unless (-d "$rdir/$dir") {
+			mkpath("$rdir/$dir") or die $!;
+		}
+		copy("$workdir/$_", "$rdir/$_") or die $!;
+		chmod(stat("$workdir/$_")->mode, "$rdir/$_") or die $!;
+	}
+
+	# Changes to submodules require special treatment. This loop writes a
+	# temporary file to both the left and right directories to show the
+	# change in the recorded SHA1 for the submodule.
+	foreach my $path (keys %submodule) {
+		if (defined $submodule{$path}{left}) {
+			my $dir = dirname($path);
+			unless (-d "$ldir/$dir") {
+				mkpath("$ldir/$dir") or die $!;
+			}
+			open(my $fh, ">", "$ldir/$path") or die $!;
+			print($fh "Subproject commit $submodule{$path}{left}");
+			close($fh);
+		}
+		if (defined $submodule{$path}{right}) {
+			my $dir = dirname($path);
+			unless (-d "$rdir/$dir") {
+				mkpath("$rdir/$dir") or die $!;
+			}
+			open(my $fh, ">", "$rdir/$path") or die $!;
+			print($fh "Subproject commit $submodule{$path}{right}");
+			close($fh);
+		}
+	}
+
+	return ($ldir, $rdir);
+}
+
 # parse command-line options. all unrecognized options and arguments
 # are passed through to the 'git diff' command.
-my ($difftool_cmd, $extcmd, $gui, $help, $prompt);
+my ($difftool_cmd, $dirdiff, $extcmd, $gui, $help, $prompt);
 GetOptions('g|gui!' => \$gui,
+	'd|dir-diff' => \$dirdiff,
 	'h' => \$help,
 	'prompt!' => \$prompt,
 	'y' => sub { $prompt = 0; },
@@ -65,22 +179,40 @@ if ($gui) {
 		$ENV{GIT_DIFF_TOOL} = $guitool;
 	}
 }
-if (defined($prompt)) {
-	if ($prompt) {
-		$ENV{GIT_DIFFTOOL_PROMPT} = 'true';
+
+# In directory diff mode, 'git-difftool--helper' is called once
+# to compare the a/b directories.  In file diff mode, 'git diff'
+# will invoke a separate instance of 'git-difftool--helper' for
+# each file that changed.
+if (defined($dirdiff)) {
+	my ($a, $b) = setup_dir_diff();
+	if (defined($extcmd)) {
+		system(($extcmd, $a, $b));
 	} else {
-		$ENV{GIT_DIFFTOOL_NO_PROMPT} = 'true';
+		$ENV{GIT_DIFFTOOL_DIRDIFF} = 'true';
+		system(('git', 'difftool--helper', $a, $b));
+	}
+
+	# If the diff including working copy files and those
+	# files were modified during the diff, then the changes
+	# should be copied back to the working tree
+	my $repo = Git->repository();
+	my $workdir = $repo->repo_path() . "/..";
+	for (@working_tree) {
+		copy("$b/$_", "$workdir/$_") or die $!;
+		chmod(stat("$b/$_")->mode, "$workdir/$_") or die $!;
+	}
+} else {
+	if (defined($prompt)) {
+		if ($prompt) {
+			$ENV{GIT_DIFFTOOL_PROMPT} = 'true';
+		} else {
+			$ENV{GIT_DIFFTOOL_NO_PROMPT} = 'true';
+		}
 	}
-}
 
-$ENV{GIT_PAGER} = '';
-$ENV{GIT_EXTERNAL_DIFF} = 'git-difftool--helper';
-my @command = ('git', 'diff', @ARGV);
-
-# ActiveState Perl for Win32 does not implement POSIX semantics of
-# exec* system call. It just spawns the given executable and finishes
-# the starting program, exiting with code 0.
-# system will at least catch the errors returned by git diff,
-# allowing the caller of git difftool better handling of failures.
-my $rc = system(@command);
-exit($rc | ($rc >> 8));
+	$ENV{GIT_PAGER} = '';
+	$ENV{GIT_EXTERNAL_DIFF} = 'git-difftool--helper';
+	my $rc = system(('git', 'diff', @ARGV));
+	exit($rc | ($rc >> 8));
+}
diff --git a/t/t7800-difftool.sh b/t/t7800-difftool.sh
index e716d06..478c1be 100755
--- a/t/t7800-difftool.sh
+++ b/t/t7800-difftool.sh
@@ -319,4 +319,43 @@ test_expect_success PERL 'say no to the second file' '
 	echo "$diff" | stdin_doesnot_contain br2
 '
 
+test_expect_success PERL 'setup change in subdirectory' '
+	git checkout master &&
+	mkdir sub &&
+	echo master >sub/sub &&
+	git add sub/sub &&
+	git commit -m "added sub/sub" &&
+	echo test >>file &&
+	echo test >>sub/sub &&
+	git add . &&
+	git commit -m "modified both"
+'
+
+test_expect_success PERL 'difftool -d' '
+	diff=$(git difftool -d --extcmd ls branch) &&
+	echo "$diff" | stdin_contains sub &&
+	echo "$diff" | stdin_contains file
+'
+
+test_expect_success PERL 'difftool --dir-diff' '
+	diff=$(git difftool --dir-diff --extcmd ls branch) &&
+	echo "$diff" | stdin_contains sub &&
+	echo "$diff" | stdin_contains file
+'
+
+test_expect_success PERL 'difftool --dir-diff ignores --prompt' '
+	diff=$(git difftool --dir-diff --prompt --extcmd ls branch) &&
+	echo "$diff" | stdin_contains sub &&
+	echo "$diff" | stdin_contains file
+'
+
+test_expect_success PERL 'difftool --dir-diff from subdirectory' '
+	(
+		cd sub &&
+		diff=$(git difftool --dir-diff --extcmd ls branch) &&
+		echo "$diff" | stdin_contains sub &&
+		echo "$diff" | stdin_contains file
+	)
+'
+
 test_done
-- 
1.7.10.rc2.22.ga52d4
Tim Henigan· Apr 4, 2012, 19:31 UTC · re: Tim Henigan · lore

Re: [PATCH 8/9 v11] difftool: teach difftool to handle directory diffs

> This replaces commit 4d6efe0 in the 'th/difftool-diffall' branch based
> on pu. This version corrects some problems with submodule support in
> difftool.
Please s/4d6efe0/4b6efe0/.  Sorry for the confusion.
David Aguilar· Apr 9, 2012, 12:14 UTC · re: Tim Henigan · lore

Re: [PATCH 8/9 v11] difftool: teach difftool to handle directory diffs

On Wed, Apr 4, 2012 at 12:21 PM, Tim Henigan <tim.henigan@gmail.com> wrote:
Show 17 quoted lines
> diff --git a/git-difftool.perl b/git-difftool.perl
> index d4fe998..5bb01e1 100755
> --- a/git-difftool.perl
> +++ b/git-difftool.perl
> @@ -1,21 +1,29 @@
>  #!/usr/bin/env perl
>  # Copyright (c) 2009, 2010 David Aguilar
> +# Copyright (c) 2012 Tim Henigan
>  #
>  # This is a wrapper around the GIT_EXTERNAL_DIFF-compatible
>  # git-difftool--helper script.
>  #
>  # This script exports GIT_EXTERNAL_DIFF and GIT_PAGER for use by git.
> -# GIT_DIFFTOOL_NO_PROMPT, GIT_DIFFTOOL_PROMPT, and GIT_DIFF_TOOL
> -# are exported for use by git-difftool--helper.
> +# GIT_DIFFTOOL_NO_PROMPT, GIT_DIFFTOOL_PROMPT, GIT_DIFFTOOL_DIRDIFF,
> +# and GIT_DIFF_TOOL are exported for use by git-difftool--helper.
What if we punt on enumerating each variable and reword this to:

# This script exports GIT_EXTERNAL_DIFF and GIT_PAGER for use by git. # The GIT_DIFF* variables are exported for use by git-difftool--helper.

I also think we should change the shebang line to #!/usr/bin/perl.
Show 28 quoted lines
> @@ -24,15 +32,121 @@ usage: git difftool [-t|--tool=<tool>]
>                     [-x|--extcmd=<cmd>]
>                     [-g|--gui] [--no-gui]
>                     [--prompt] [-y|--no-prompt]
> +                    [-d|--dir-diff]
>                     ['git diff' options]
>  USAGE
>        exit($exitcode);
>  }
>
> +sub setup_dir_diff
> +{
> +       # Run the diff; exit immediately if no diff found
> +       my $repo = Git->repository();
> +       my $diffrtn = $repo->command_oneline(['diff', '--raw', '--no-abbrev', '-z', @ARGV]);
> +       exit(0) if (length($diffrtn) == 0);
> +
> +       # Setup temp directories
> +       my $tmpdir = tempdir('git-diffall.XXXXX', CLEANUP => 1, TMPDIR => 1);
> +       my $ldir = "$tmpdir/left";
> +       my $rdir = "$tmpdir/right";
> +       mkpath($ldir) or die $!;
> +       mkpath($rdir) or die $!;
> +
> +       # Build index info for left and right sides of the diff
> +       my $submodule_mode = "160000";
> +       my $null_mode = 0 x 6;
> +       my $null_sha1 = 0 x 40;
I know Perl allows it, but my eyes get confused.

How about writing "0" x 6 and "0" x 40? That way we can visually see that the result is a string.

Show 6 quoted lines
> +       my $lindex = "";
> +       my $rindex = "";
> +       my %submodule;
> +       my @rawdiff = split('\0', $diffrtn);
> +
> +       for (my $i=0; $i<@rawdiff; $i+=2) {

We use $i + 1 to grab list elements, so how about $#rawdiff instead of @rawdiff? It doesn't matter in practice, though...

Show 34 quoted lines
> +               my ($lmode, $rmode, $lsha1, $rsha1, $status) = split(' ', substr($rawdiff[$i], 1));
> +               my $path = $rawdiff[$i + 1];
> +
> +               if (($lmode eq $submodule_mode) or ($rmode eq $submodule_mode)) {
> +                       $submodule{$path}{left} = $lsha1;
> +                       if ($lsha1 ne $rsha1) {
> +                               $submodule{$path}{right} = $rsha1;
> +                       } else {
> +                               $submodule{$path}{right} = "$rsha1-dirty";
> +                       }
> +                       next;
> +               }
> +
> +               if ($lmode ne $null_mode) {
> +                       $lindex .= "$lmode $lsha1\t$path\0";
> +               }
> +
> +               if ($rmode ne $null_mode) {
> +                       if ($rsha1 ne $null_sha1) {
> +                               $rindex .= "$rmode $rsha1\t$path\0";
> +                       } else {
> +                               push(@working_tree, $path);
> +                       }
> +               }
> +       }
> +
> +       # Populate the left and right directories based on each index file
> +       my ($inpipe, $ctx);
> +       $ENV{GIT_DIR} = $repo->repo_path();
> +       $ENV{GIT_INDEX_FILE} = "$tmpdir/lindex";
> +       ($inpipe, $ctx) = $repo->command_input_pipe(qw/update-index -z --index-info/);
> +       print($inpipe $lindex);
> +       $repo->command_close_pipe($inpipe, $ctx);
> +       system(('git', 'checkout-index', '--all', "--prefix=$ldir/"));

Please drop the extra parens. Perl ignores them. We should also check the return value from system() here.

Show 11 quoted lines
> +
> +       $ENV{GIT_INDEX_FILE} = "$tmpdir/rindex";
> +       ($inpipe, $ctx) = $repo->command_input_pipe(qw/update-index -z --index-info/);
> +       print($inpipe $rindex);
> +       $repo->command_close_pipe($inpipe, $ctx);
> +       system(('git', 'checkout-index', '--all', "--prefix=$rdir/"));
> +
> +       # Changes in the working tree need special treatment since they are
> +       # not part of the index
> +       my $workdir = $repo->repo_path() . "/..";
> +       for (@working_tree) {
Please use the "for my $foo (@list)" form so we can say $foo instead of $_.
Show 12 quoted lines
> +               my $dir = dirname($_);
> +               unless (-d "$rdir/$dir") {
> +                       mkpath("$rdir/$dir") or die $!;
> +               }
> +               copy("$workdir/$_", "$rdir/$_") or die $!;
> +               chmod(stat("$workdir/$_")->mode, "$rdir/$_") or die $!;
> +       }
> +
> +       # Changes to submodules require special treatment. This loop writes a
> +       # temporary file to both the left and right directories to show the
> +       # change in the recorded SHA1 for the submodule.
> +       foreach my $path (keys %submodule) {

I think it's better to use "for" instead of "foreach" since we do not modify $path.

> +               if (defined $submodule{$path}{left}) {

In some places we write "defined(...)" but here it's "defined ...". We should be consistent...

Show 27 quoted lines
> @@ -65,22 +179,40 @@ if ($gui) {
>                $ENV{GIT_DIFF_TOOL} = $guitool;
>        }
>  }
> -if (defined($prompt)) {
> -       if ($prompt) {
> -               $ENV{GIT_DIFFTOOL_PROMPT} = 'true';
> +
> +# In directory diff mode, 'git-difftool--helper' is called once
> +# to compare the a/b directories.  In file diff mode, 'git diff'
> +# will invoke a separate instance of 'git-difftool--helper' for
> +# each file that changed.
> +if (defined($dirdiff)) {
> +       my ($a, $b) = setup_dir_diff();
> +       if (defined($extcmd)) {
> +               system(($extcmd, $a, $b));
>        } else {
> -               $ENV{GIT_DIFFTOOL_NO_PROMPT} = 'true';
> +               $ENV{GIT_DIFFTOOL_DIRDIFF} = 'true';
> +               system(('git', 'difftool--helper', $a, $b));
> +       }
> +
> +       # If the diff including working copy files and those
> +       # files were modified during the diff, then the changes
> +       # should be copied back to the working tree
> +       my $repo = Git->repository();
> +       my $workdir = $repo->repo_path() . "/..";
Does this work when $GIT_WORK_TREE / core.worktree are defined?
> +       for (@working_tree) {
> +               copy("$b/$_", "$workdir/$_") or die $!;
> +               chmod(stat("$b/$_")->mode, "$workdir/$_") or die $!;
> +       }
"for my ..."
Show 26 quoted lines
> +} else {
> +       if (defined($prompt)) {
> +               if ($prompt) {
> +                       $ENV{GIT_DIFFTOOL_PROMPT} = 'true';
> +               } else {
> +                       $ENV{GIT_DIFFTOOL_NO_PROMPT} = 'true';
> +               }
>        }
> -}
>
> -$ENV{GIT_PAGER} = '';
> -$ENV{GIT_EXTERNAL_DIFF} = 'git-difftool--helper';
> -my @command = ('git', 'diff', @ARGV);
> -
> -# ActiveState Perl for Win32 does not implement POSIX semantics of
> -# exec* system call. It just spawns the given executable and finishes
> -# the starting program, exiting with code 0.
> -# system will at least catch the errors returned by git diff,
> -# allowing the caller of git difftool better handling of failures.
> -my $rc = system(@command);
> -exit($rc | ($rc >> 8));
> +       $ENV{GIT_PAGER} = '';
> +       $ENV{GIT_EXTERNAL_DIFF} = 'git-difftool--helper';
> +       my $rc = system(('git', 'diff', @ARGV));
> +       exit($rc | ($rc >> 8));
> +}

We went back and forth a few times on this section, eventually landing back on using system().

Should we retain this comment to help future readers from having to re-learn it the hard way again?

We could also link to the ML threads, if you think that's helpful.
-- 
David
Junio C Hamano· Apr 9, 2012, 18:02 UTC · re: David Aguilar · lore

Re: [PATCH 8/9 v11] difftool: teach difftool to handle directory diffs

David Aguilar <davvid@gmail.com> writes:
Show 9 quoted lines
> On Wed, Apr 4, 2012 at 12:21 PM, Tim Henigan <tim.henigan@gmail.com> wrote:
>> diff --git a/git-difftool.perl b/git-difftool.perl
>> index d4fe998..5bb01e1 100755
>> --- a/git-difftool.perl
>> +++ b/git-difftool.perl
>> @@ -1,21 +1,29 @@
>>  #!/usr/bin/env perl
> ...
> I also think we should change the shebang line to #!/usr/bin/perl.

Good point; it would not make a difference in the end result, but consistency is good.

Show 19 quoted lines
>> -# ActiveState Perl for Win32 does not implement POSIX semantics of
>> -# exec* system call. It just spawns the given executable and finishes
>> -# the starting program, exiting with code 0.
>> -# system will at least catch the errors returned by git diff,
>> -# allowing the caller of git difftool better handling of failures.
>> -my $rc = system(@command);
>> -exit($rc | ($rc >> 8));
>> +       $ENV{GIT_PAGER} = '';
>> +       $ENV{GIT_EXTERNAL_DIFF} = 'git-difftool--helper';
>> +       my $rc = system(('git', 'diff', @ARGV));
>> +       exit($rc | ($rc >> 8));
>> +}
>
> 
> We went back and forth a few times on this section,
> eventually landing back on using system().
>
> Should we retain this comment to help future readers from
> having to re-learn it the hard way again?

Well, I kept typing "good point" but ended up agreeing with everything you said in your message, and removed all "Me too"s ;-).

Thanks for a thorough review.
Tim Henigan· Apr 10, 2012, 15:01 UTC · re: David Aguilar · lore

Re: [PATCH 8/9 v11] difftool: teach difftool to handle directory diffs

Hi David,

Thank you for the review. I am working on the changes you suggested, but want to discuss one of them:

On Mon, Apr 9, 2012 at 8:14 AM, David Aguilar <davvid@gmail.com> wrote:
Show 9 quoted lines
> On Wed, Apr 4, 2012 at 12:21 PM, Tim Henigan <tim.henigan@gmail.com> wrote:
>> +
>> +       # If the diff including working copy files and those
>> +       # files were modified during the diff, then the changes
>> +       # should be copied back to the working tree
>> +       my $repo = Git->repository();
>> +       my $workdir = $repo->repo_path() . "/..";
>
> Does this work when $GIT_WORK_TREE / core.worktree are defined?

I had not considered this case and the existing code will certainly ignore alternate work tree locations. To fix the problem, I plan to add something like this:

+sub find_worktree
+{
+       my $worktree;
+       my $env_worktree = $ENV{GIT_WORK_TREE};
+       my $core_worktree = Git::config('core.worktree');
+
+       if (length($env_worktree) > 0) {
+               $worktree = $env_worktree;
+       } elsif (length($core_worktree) > 0) {
+               $worktree = $core_worktree;
+       } else {
+               my $repo = Git->repository();
+               $worktree = $repo->repo_path() . "/..";
+       }
+
+       return $worktree;
+}

Does this look like a reasonable solution? I don't typically use '$GIT_WORK_TREE' or 'core.worktree'. The above function works for me, but I would appreciate some feedback before I send the next revision of the patch (v12!).

Thanks, Tim

Tim Henigan· Apr 10, 2012, 17:24 UTC · re: David Aguilar · lore

Re: [PATCH 8/9 v11] difftool: teach difftool to handle directory diffs

On Mon, Apr 9, 2012 at 8:14 AM, David Aguilar <davvid@gmail.com> wrote:
Show 6 quoted lines
> On Wed, Apr 4, 2012 at 12:21 PM, Tim Henigan <tim.henigan@gmail.com> wrote:
>> diff --git a/git-difftool.perl b/git-difftool.perl
>> index d4fe998..5bb01e1 100755
>> --- a/git-difftool.perl
>> +++ b/git-difftool.perl
>> @@ -1,21 +1,29 @@
...snip...
Show 6 quoted lines
>> +       $ENV{GIT_DIR} = $repo->repo_path();
>> +       $ENV{GIT_INDEX_FILE} = "$tmpdir/lindex";
>> +       ($inpipe, $ctx) = $repo->command_input_pipe(qw/update-index -z --index-info/);
>> +       print($inpipe $lindex);
>> +       $repo->command_close_pipe($inpipe, $ctx);
>> +       system(('git', 'checkout-index', '--all', "--prefix=$ldir/"));
...snip...
Show 7 quoted lines
>> +       # If the diff including working copy files and those
>> +       # files were modified during the diff, then the changes
>> +       # should be copied back to the working tree
>> +       my $repo = Git->repository();
>> +       my $workdir = $repo->repo_path() . "/..";
>
> Does this work when $GIT_WORK_TREE / core.worktree are defined?
Should I also be concerned that the existing code always overrides $GIT_DIR?
For example, should I squash in the following before calling 'git update-index'?
-       $ENV{GIT_DIR} = $repo->repo_path();
+      if (not defined($ENV{GIT_DIR})) {
+              $ENV{GIT_DIR} = $repo->repo_path();
+      }

This seems like the right thing to do, but again I have not used $GIT_DIR except for the implementation of this script.

David Aguilar· Apr 11, 2012, 09:02 UTC · re: Tim Henigan · lore

Re: [PATCH 8/9 v11] difftool: teach difftool to handle directory diffs

On Tue, Apr 10, 2012 at 10:24 AM, Tim Henigan <tim.henigan@gmail.com> wrote:
Show 34 quoted lines
> On Mon, Apr 9, 2012 at 8:14 AM, David Aguilar <davvid@gmail.com> wrote:
>> On Wed, Apr 4, 2012 at 12:21 PM, Tim Henigan <tim.henigan@gmail.com> wrote:
>>> diff --git a/git-difftool.perl b/git-difftool.perl
>>> index d4fe998..5bb01e1 100755
>>> --- a/git-difftool.perl
>>> +++ b/git-difftool.perl
>>> @@ -1,21 +1,29 @@
>
> ...snip...
>
>>> +       $ENV{GIT_DIR} = $repo->repo_path();
>>> +       $ENV{GIT_INDEX_FILE} = "$tmpdir/lindex";
>>> +       ($inpipe, $ctx) = $repo->command_input_pipe(qw/update-index -z --index-info/);
>>> +       print($inpipe $lindex);
>>> +       $repo->command_close_pipe($inpipe, $ctx);
>>> +       system(('git', 'checkout-index', '--all', "--prefix=$ldir/"));
>
> ...snip...
>
>>> +       # If the diff including working copy files and those
>>> +       # files were modified during the diff, then the changes
>>> +       # should be copied back to the working tree
>>> +       my $repo = Git->repository();
>>> +       my $workdir = $repo->repo_path() . "/..";
>>
>> Does this work when $GIT_WORK_TREE / core.worktree are defined?
>
> Should I also be concerned that the existing code always overrides $GIT_DIR?
> For example, should I squash in the following before calling 'git update-index'?
>
> -       $ENV{GIT_DIR} = $repo->repo_path();
> +      if (not defined($ENV{GIT_DIR})) {
> +              $ENV{GIT_DIR} = $repo->repo_path();
> +      }
I think the right thing to do would be to not override GIT_DIR at all.
 I haven't read it deeply enough to know whether it was being
overridden for a specific reason, but I think it should be safe to
leave it as-is.
Git.pm ends up overriding these variables itself anyways when calling commands.

The GIT_WORK_TREE check should use $repo->wc_path(). Git.pm's already done all the hard work ;-)

-- 
David
Tim Henigan· Apr 11, 2012, 16:46 UTC · re: David Aguilar · lore

Re: [PATCH 8/9 v11] difftool: teach difftool to handle directory diffs

On Wed, Apr 11, 2012 at 5:02 AM, David Aguilar <davvid@gmail.com> wrote:
Show 10 quoted lines
> On Tue, Apr 10, 2012 at 10:24 AM, Tim Henigan <tim.henigan@gmail.com> wrote:
>> On Mon, Apr 9, 2012 at 8:14 AM, David Aguilar <davvid@gmail.com> wrote:
>>> On Wed, Apr 4, 2012 at 12:21 PM, Tim Henigan <tim.henigan@gmail.com> wrote:
>
> I think the right thing to do would be to not override GIT_DIR at all.
>  I haven't read it deeply enough to know whether it was being
> overridden for a specific reason, but I think it should be safe to
> leave it as-is.
>
> Git.pm ends up overriding these variables itself anyways when calling commands.

I tried to avoid setting $GIT_DIR in earlier versions of the patch. However as discussed here [1], either 'git update-index' or 'git checkout-index' did not work as expected without explicitly setting $GIT_DIR.

If $GIT_DIR is not set, 'update-index' and 'checkout-index' will only work if 'difftool' is called from the repo root. If 'difftool' is called from a subdirectory, then one of the commands fails.

I suspect that when $GIT_INDEX_FILE is set but $GIT_DIR is not, then $GIT_DIR is assumed to be 'pwd'. However, I was not able to prove that.

> The GIT_WORK_TREE check should use $repo->wc_path().  Git.pm's already
> done all the hard work ;-)

I also tried this in an earlier version of the patch. As noted here [2], I found that when 'difftool' was run from a subdirectory of the repo root, '$repo->wc_path()' returned the subdirectory rather than the repo root.

Thinking about this again now, I realize it was a side-effect of $GIT_DIR being set in the script. The man page for git config states that:

    If --git-dir or GIT_DIR is specified but none of --work-tree, GIT_WORK_TREE
    and core.worktree is specified, the current working directory is regarded as
    the top level of your working tree.

So, if I explicitly set $GIT_DIR just for the 'update-index' and 'checkout-index' commands, I need to unset it afterwards. This should allow '$repo->wc_path()' to behave as expected.

[1]: http://thread.gmane.org/gmane.comp.version-control.git/193296/focus=193302 [2]: http://thread.gmane.org/gmane.comp.version-control.git/193601/focus=193603

David Aguilar· Apr 12, 2012, 03:21 UTC · re: Tim Henigan · lore

Re: [PATCH 8/9 v11] difftool: teach difftool to handle directory diffs

On Wed, Apr 11, 2012 at 9:46 AM, Tim Henigan <tim.henigan@gmail.com> wrote:
Show 48 quoted lines
> On Wed, Apr 11, 2012 at 5:02 AM, David Aguilar <davvid@gmail.com> wrote:
>> On Tue, Apr 10, 2012 at 10:24 AM, Tim Henigan <tim.henigan@gmail.com> wrote:
>>> On Mon, Apr 9, 2012 at 8:14 AM, David Aguilar <davvid@gmail.com> wrote:
>>>> On Wed, Apr 4, 2012 at 12:21 PM, Tim Henigan <tim.henigan@gmail.com> wrote:
>>
>> I think the right thing to do would be to not override GIT_DIR at all.
>>  I haven't read it deeply enough to know whether it was being
>> overridden for a specific reason, but I think it should be safe to
>> leave it as-is.
>>
>> Git.pm ends up overriding these variables itself anyways when calling commands.
>
> I tried to avoid setting $GIT_DIR in earlier versions of the patch.
> However as discussed here [1], either 'git update-index' or 'git
> checkout-index' did not work as expected without explicitly setting
> $GIT_DIR.
>
> If $GIT_DIR is not set, 'update-index' and 'checkout-index' will only
> work if 'difftool' is called from the repo root.  If 'difftool' is
> called from a subdirectory, then one of the commands fails.
>
> I suspect that when $GIT_INDEX_FILE is set but $GIT_DIR is not, then
> $GIT_DIR is assumed to be 'pwd'.  However, I was not able to prove
> that.
>
>
>> The GIT_WORK_TREE check should use $repo->wc_path().  Git.pm's already
>> done all the hard work ;-)
>
> I also tried this in an earlier version of the patch.  As noted here
> [2], I found that when 'difftool' was run from a subdirectory of the
> repo root, '$repo->wc_path()' returned the subdirectory rather than
> the repo root.
>
> Thinking about this again now, I realize it was a side-effect of
> $GIT_DIR being set in the script.  The man page for git config states
> that:
>
>    If --git-dir or GIT_DIR is specified but none of --work-tree, GIT_WORK_TREE
>    and core.worktree is specified, the current working directory is regarded as
>    the top level of your working tree.
>
> So, if I explicitly set $GIT_DIR just for the 'update-index' and
> 'checkout-index' commands, I need to unset it afterwards.  This should
> allow '$repo->wc_path()' to behave as expected.
>
> [1]: http://thread.gmane.org/gmane.comp.version-control.git/193296/focus=193302
> [2]: http://thread.gmane.org/gmane.comp.version-control.git/193601/focus=193603

That's right. If we set it for one command we have to remember to unset it afterwards else it won't work as expected.

If we're setting GIT_DIR then we should probably set GIT_WORK_TREE too. It seems like one way would be to call repo_path() and wc_path() early, set the variables with the returned values, and then worry about managing GIT_WORK_TREE before+after calling checkout-index. That might work.

And in case the code needs it, there's a GIT_PREFIX variable that is set when the current directory is a subdir beneath the repo root.

-- 
David

← back to recent threads