[PATCH] gitweb: Use git-show-ref instead of git-peek-remote

[PATCH] gitweb: Use git-show-ref instead of git-peek-remote

[Jakub Narebski]

Use "git show-ref --dereference" instead of "git peek-remote ." in
git_get_references. git-show-ref is faster than git-peek-remote; even
faster is reading info/refs file (if it exists), but the information
in info/refs can be stale.

git-show-ref is available since v1.4.4; the output format is slightly
different than git-peek-remote output format.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
---
 gitweb/gitweb.perl |    9 +++++----
 1 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/gitweb/gitweb.perl b/gitweb/gitweb.perl
index f06cd3e..290751f 100755
--- a/gitweb/gitweb.perl
+++ b/gitweb/gitweb.perl
@@ -1154,14 +1154,15 @@ sub git_get_last_activity {
 sub git_get_references {
 	my $type = shift || "";
 	my %refs;
-	# 5dc01c595e6c6ec9ccda4f6f69c131c0dd945f8c	refs/tags/v2.6.11
-	# c39ae07f393806ccf406ef966e9a15afc43cc36a	refs/tags/v2.6.11^{}
-	open my $fd, "-|", $GIT, "peek-remote", "$projectroot/$project/"
+	# 5dc01c595e6c6ec9ccda4f6f69c131c0dd945f8c refs/tags/v2.6.11
+	# c39ae07f393806ccf406ef966e9a15afc43cc36a refs/tags/v2.6.11^{}
+	open my $fd, "-|", git_cmd(), "show-ref", "--dereference",
+		($type ? ("--", "refs/$type") : ()) # use -- <pattern> if $type
 		or return;
 
 	while (my $line = <$fd>) {
 		chomp $line;
-		if ($line =~ m/^([0-9a-fA-F]{40})\trefs\/($type\/?[^\^]+)/) {
+		if ($line =~ m/^([0-9a-fA-F]{40}) refs\/($type\/?[^\^]+)/) {
 			if (defined $refs{$1}) {
 				push @{$refs{$1}}, $2;
 			} else {
-- 
1.4.4.1

Re: [PATCH] gitweb: Use git-show-ref instead of git-peek-remote

[Junio C Hamano]

Jakub Narebski <jnareb@gmail.com> writes:

> Use "git show-ref --dereference" instead of "git peek-remote ." in
> git_get_references. git-show-ref is faster than git-peek-remote; even
> faster is reading info/refs file (if it exists), but the information
> in info/refs can be stale.

More importantly, it is for dumb protocol transports, not for gitweb.

You forgot to mention that you fixed the last place that
directly used "$GIT" to invoke the command, bypassing sub
git_cmd.  That is a consistency clean-up worth mentioning.

> git-show-ref is available since v1.4.4; the output format is slightly
> different than git-peek-remote output format.

> -		if ($line =~ m/^([0-9a-fA-F]{40})\trefs\/($type\/?[^\^]+)/) {
> +		if ($line =~ m/^([0-9a-fA-F]{40}) refs\/($type\/?[^\^]+)/) {

I would rather do:

	m|^([0-9a-f]{40})\srefs/($type/?[^^]+)|

which would catch both space and tab.

Re: [PATCH] gitweb: Use git-show-ref instead of git-peek-remote

[Jakub Narebski]

Junio C Hamano wrote:

> Jakub Narebski <jnareb@gmail.com> writes:
> 
>> Use "git show-ref --dereference" instead of "git peek-remote ." in

It should be 
   Use "git show-ref --dereference" instead of "git peek-remote <project>"

>> git_get_references. git-show-ref is faster than git-peek-remote; even
>> faster is reading info/refs file (if it exists), but the information
>> in info/refs can be stale.
> 
> More importantly, it is for dumb protocol transports, not for gitweb.

Yes, but if it is there, why not make use of it? Especially that it
is used only (I think) for refs markers, and not for anything important.
On the other side a bit of performance here doesn't matter much.

> You forgot to mention that you fixed the last place that
> directly used "$GIT" to invoke the command, bypassing sub
> git_cmd.  That is a consistency clean-up worth mentioning.

This is not fix, this is change of style. The style was to use
  git peek-remote $projectroot/$project
instead of equivalent
  git --git-dir=$projectroot/$project peek-remote .

We don't have this choice with git-show-ref.

On the other hand that makes all command invocation (except the one
used for getting git core version) pass through git_cmd() subroutine.

>> git-show-ref is available since v1.4.4; the output format is slightly
>> different than git-peek-remote output format.
> 
>> -		if ($line =~ m/^([0-9a-fA-F]{40})\trefs\/($type\/?[^\^]+)/) {
>> +		if ($line =~ m/^([0-9a-fA-F]{40}) refs\/($type\/?[^\^]+)/) {
> 
> I would rather do:
> 
> 	m|^([0-9a-f]{40})\srefs/($type/?[^^]+)|
> 
> which would catch both space and tab.

Feel free to do that (one nitpick: is [^^] more readable than [^\^]?).

Should I resend the patch?
-- 
Jakub Narebski

[PATCH (amend)] gitweb: Use git-show-ref instead of git-peek-remote

[Jakub Narebski]

Use "git show-ref --dereference" instead of "git peek-remote
$projectroot/project" in git_get_references. git-show-ref is faster
than git-peek-remote (40ms vs 56ms user+sys for git.git repository);
even faster is reading info/refs file (if it exists), but the
information in info/refs can be stale; that and the fact that
info/refs is meant for dumb protocol transports, not for gitweb.

git-show-ref is available since v1.4.4; the output format is slightly
different than git-peek-remote output format.

While at it make git_get_references return hash in list context,
and reference to hash (as it used to do) in scalar and void contexts.

Additionally this change makes all git commands invocations (except
the one used to get version of git core tools) go through git_cmd()
subroutine.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
---
Junio C Hamano wrote:

> Jakub Narebski <jnareb@gmail.com> writes:
> 
>> Use "git show-ref --dereference" instead of "git peek-remote ." in
>> git_get_references. git-show-ref is faster than git-peek-remote; even
>> faster is reading info/refs file (if it exists), but the information
>> in info/refs can be stale.
> 
> More importantly, it is for dumb protocol transports, not for gitweb.

Mentioned in commit message.

> You forgot to mention that you fixed the last place that
> directly used "$GIT" to invoke the command, bypassing sub
> git_cmd.  That is a consistency clean-up worth mentioning.

This is not fix, this is change of style. The style was to use
  git peek-remote $projectroot/$project
instead of equivalent
  git --git-dir=$projectroot/$project peek-remote .

We don't have this choice with git-show-ref.

On the other hand that makes all command invocation (except the one
used for getting git core version) pass through git_cmd() subroutine.
So yes, it is consistency clean-up.

>> git-show-ref is available since v1.4.4; the output format is slightly
>> different than git-peek-remote output format.
> 
>> -		if ($line =~ m/^([0-9a-fA-F]{40})\trefs\/($type\/?[^\^]+)/) {
>> +		if ($line =~ m/^([0-9a-fA-F]{40}) refs\/($type\/?[^\^]+)/) {
> 
> I would rather do:
> 
> 	m|^([0-9a-f]{40})\srefs/($type/?[^^]+)|
> 
> which would catch both space and tab.

This addresses those concerns.

 gitweb/gitweb.perl |   23 +++++++++++++++--------
 1 files changed, 15 insertions(+), 8 deletions(-)

diff --git a/gitweb/gitweb.perl b/gitweb/gitweb.perl
index f06cd3e..2ebd9d7 100755
--- a/gitweb/gitweb.perl
+++ b/gitweb/gitweb.perl
@@ -1154,14 +1154,15 @@ sub git_get_last_activity {
 sub git_get_references {
 	my $type = shift || "";
 	my %refs;
-	# 5dc01c595e6c6ec9ccda4f6f69c131c0dd945f8c	refs/tags/v2.6.11
-	# c39ae07f393806ccf406ef966e9a15afc43cc36a	refs/tags/v2.6.11^{}
-	open my $fd, "-|", $GIT, "peek-remote", "$projectroot/$project/"
+	# 5dc01c595e6c6ec9ccda4f6f69c131c0dd945f8c refs/tags/v2.6.11
+	# c39ae07f393806ccf406ef966e9a15afc43cc36a refs/tags/v2.6.11^{}
+	open my $fd, "-|", git_cmd(), "show-ref", "--dereference",
+		($type ? ("--", "refs/$type") : ()) # use -- <pattern> if $type
 		or return;
 
 	while (my $line = <$fd>) {
 		chomp $line;
-		if ($line =~ m/^([0-9a-fA-F]{40})\trefs\/($type\/?[^\^]+)/) {
+		if ($line =~ m!^([0-9a-fA-F]{40})\srefs/($type/?[^^]+)!) {
 			if (defined $refs{$1}) {
 				push @{$refs{$1}}, $2;
 			} else {
@@ -1170,7 +1171,7 @@ sub git_get_references {
 		}
 	}
 	close $fd or return;
-	return \%refs;
+	return wantarray ? %refs : \%refs;
 }
 
 sub git_get_rev_name_tags {
@@ -1293,8 +1294,9 @@ sub parse_commit {
 			$co{'author'} = $1;
 			$co{'author_epoch'} = $2;
 			$co{'author_tz'} = $3;
-			if ($co{'author'} =~ m/^([^<]+) </) {
-				$co{'author_name'} = $1;
+			if ($co{'author'} =~ m/^([^<]+) <([^>]*)>/) {
+				$co{'author_name'}  = $1;
+				$co{'author_email'} = $2;
 			} else {
 				$co{'author_name'} = $co{'author'};
 			}
@@ -1303,7 +1305,12 @@ sub parse_commit {
 			$co{'committer_epoch'} = $2;
 			$co{'committer_tz'} = $3;
 			$co{'committer_name'} = $co{'committer'};
-			$co{'committer_name'} =~ s/ <.*//;
+			if ($co{'committer'} =~ m/^([^<]+) <([^>]*)>/) {
+				$co{'committer_name'}  = $1;
+				$co{'committer_email'} = $2;
+			} else {
+				$co{'committer_name'} = $co{'committer'};
+			}
 		}
 	}
 	if (!defined $co{'tree'}) {
-- 
1.4.4.1

Re: [PATCH (amend)] gitweb: Use git-show-ref instead of git-peek-remote

[Jakub Narebski]

Jakub Narebski wrote:

Ooops. This is a part of next patch to send. Please delete this two chunks.
Sorry for the mistake.


> @@ -1293,8 +1294,9 @@ sub parse_commit {
>                       $co{'author'} = $1;
>                       $co{'author_epoch'} = $2;
>                       $co{'author_tz'} = $3;
> -                     if ($co{'author'} =~ m/^([^<]+) </) {
> -                             $co{'author_name'} = $1;
> +                     if ($co{'author'} =~ m/^([^<]+) <([^>]*)>/) {
> +                             $co{'author_name'}  = $1;
> +                             $co{'author_email'} = $2;
>                       } else {
>                               $co{'author_name'} = $co{'author'};
>                       }
> @@ -1303,7 +1305,12 @@ sub parse_commit {
>                       $co{'committer_epoch'} = $2;
>                       $co{'committer_tz'} = $3;
>                       $co{'committer_name'} = $co{'committer'};
> -                     $co{'committer_name'} =~ s/ <.*//;
> +                     if ($co{'committer'} =~ m/^([^<]+) <([^>]*)>/) {
> +                             $co{'committer_name'}  = $1;
> +                             $co{'committer_email'} = $2;
> +                     } else {
> +                             $co{'committer_name'} = $co{'committer'};
> +                     }
>               }
>       }
>       if (!defined $co{'tree'}) {

-- 
Jakub Narebski
Warsaw, Poland
ShadeHawk on #git

[PATCH (take 3)] gitweb: Use git-show-ref instead of git-peek-remote

[Jakub Narebski]

Use "git show-ref --dereference" instead of "git peek-remote
$projectroot/project" in git_get_references. git-show-ref is faster
than git-peek-remote (40ms vs 56ms user+sys for git.git repository);
even faster is reading info/refs file (if it exists), but the
information in info/refs can be stale; that and the fact that
info/refs is meant for dumb protocol transports, not for gitweb.

git-show-ref is available since v1.4.4; the output format is slightly
different than git-peek-remote output format.

While at it make git_get_references return hash in list context,
and reference to hash (as it used to do) in scalar and void contexts.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
---
This is the final version.

Once again, I'm extremly sorry for the confusion with the previous
version...

 gitweb/gitweb.perl |   11 ++++++-----
 1 files changed, 6 insertions(+), 5 deletions(-)

diff --git a/gitweb/gitweb.perl b/gitweb/gitweb.perl
index f06cd3e..1cded75 100755
--- a/gitweb/gitweb.perl
+++ b/gitweb/gitweb.perl
@@ -1154,14 +1154,15 @@ sub git_get_last_activity {
 sub git_get_references {
 	my $type = shift || "";
 	my %refs;
-	# 5dc01c595e6c6ec9ccda4f6f69c131c0dd945f8c	refs/tags/v2.6.11
-	# c39ae07f393806ccf406ef966e9a15afc43cc36a	refs/tags/v2.6.11^{}
-	open my $fd, "-|", $GIT, "peek-remote", "$projectroot/$project/"
+	# 5dc01c595e6c6ec9ccda4f6f69c131c0dd945f8c refs/tags/v2.6.11
+	# c39ae07f393806ccf406ef966e9a15afc43cc36a refs/tags/v2.6.11^{}
+	open my $fd, "-|", git_cmd(), "show-ref", "--dereference",
+		($type ? ("--", "refs/$type") : ()) # use -- <pattern> if $type
 		or return;
 
 	while (my $line = <$fd>) {
 		chomp $line;
-		if ($line =~ m/^([0-9a-fA-F]{40})\trefs\/($type\/?[^\^]+)/) {
+		if ($line =~ m!^([0-9a-fA-F]{40})\srefs/($type/?[^^]+)!) {
 			if (defined $refs{$1}) {
 				push @{$refs{$1}}, $2;
 			} else {
@@ -1170,7 +1171,7 @@ sub git_get_references {
 		}
 	}
 	close $fd or return;
-	return \%refs;
+	return wantarray ? %refs : \%refs;
 }
 
 sub git_get_rev_name_tags {
-- 
1.4.4.1

Re: [PATCH (take 3)] gitweb: Use git-show-ref instead of git-peek-remote

[Junio C Hamano]

Jakub Narebski <jnareb@gmail.com> writes:

> While at it make git_get_references return hash in list context,
> and reference to hash (as it used to do) in scalar and void contexts.

Why did you have to add this?  Generally, context sensitive
returns make the program much harder to maintain, because it
forces the call sites to be extremely careful to choose between
"my ($foo) = func()" vs "my $foo = func()", and people who later
call the function inevitably make mistakes.

So unless there is a compelling reason, I'd rather not see more
"wantarray" in the program.

Re: [PATCH (take 3)] gitweb: Use git-show-ref instead of git-peek-remote

[Jakub Narebski]

Junio C Hamano wrote:
> Jakub Narebski <jnareb@gmail.com> writes:
> 
>> While at it make git_get_references return hash in list context,
>> and reference to hash (as it used to do) in scalar and void contexts.
> 
> Why did you have to add this?  Generally, context sensitive
> returns make the program much harder to maintain, because it
> forces the call sites to be extremely careful to choose between
> "my ($foo) = func()" vs "my $foo = func()", and people who later
> call the function inevitably make mistakes.
> 
> So unless there is a compelling reason, I'd rather not see more
> "wantarray" in the program.

O.K. I have browsed through gitweb, and I see that we almost always
want the hashref, not hash (for passing further).

-- 
Jakub Narebski