Re: [PATCH 2/4] gitweb: Make feed title valid utf8

From: Jakub Narębski <jnareb@gmail.com>
Date: Apr 9, 2013, 19:58 UTC
Message-ID: <516472F1.4060903@gmail.com>
In-Reply-To: <m2ip3va4ro.fsf@zahir.fritz.box>

W dniu 09.04.2013 21:22, Jürgen Kreileder napisał:

Show 6 quoted lines

> Jakub Narębski <jnareb@gmail.com> writes: 
>> W dniu 09.04.2013 19:40, Jürgen Kreileder napisał:
>>> Jakub Narębski <jnareb@gmail.com> writes:
>>>> Jürgen Kreileder wrote:
>>>>
>>>>> Properly encode site and project names for RSS and Atom feeds.

Show 22 quoted lines

>>> Good point.  But it doesn't fix the string in question:
>>> It looks like to_utf8("$a $b") != (to_utf8($a) . " " . to_utf8($b)).
>>
>> Strange.  I wonder if the bug is in our to_utf8() implementation,
>> or in Encode, or in Perl... and whether this bug can be triggered
>> anywhere else in gitweb.
> 
> I don't think it's a bug, more like a consequence of concatenating utf8
> and non-utf8 strings:
> 
>     my $a = "ü";
>     my $b = "ü";
>     my $c = "$a - $b";
>     print "$c -> ". to_utf8($c) . ": " . (utf8::is_utf8($c) ? "utf8" : "not utf8") . "\n"; # GOOD
>     $b = to_utf8($b);
>     $c = "$a - $b";
>     print "$c -> ". to_utf8($c) . ": " . (utf8::is_utf8($c) ? "utf8" : "not utf8") . "\n"; # GOOD
> 
> yields (hopefully the broken encoding shows up correctly here):
> 
>     Ã¼ - Ã¼ -> ü - ü: not utf8
>     Ã¼ - ü -> Ã¼ - ü: utf8

Ah, so it looks like it is misfeature of the way Perl handles Unicode; concatenating adds 'UTF8' flag if either of concatenates strings has it to the result.

[Which I have checked using Devel::Peek with
 perl -MDevel::Peek -E '
  my $a = "ż"; my $b = "\x{17c}";
  Dump $a; Dump $b; Dump "$b - $a"'
]

Show 5 quoted lines

> In gitweb we have the bad case: 
> 
>    my $title = "$site_name - $project/$action";
> 
> $project and $action are apparently utf8 already but $site_name isn't.

$project and $action are taken from URL, and we have to run decode_utf8 (at least for query params) for gitweb to work correctly.

$site_name is usually taken from config file, and gitweb doesn't have "use utf8" pragma.

> The resulting string is marked as utf8 - although the encoding of
> $site_name was never fixed.  The to_utf8() in esc_html() returns the string
> without fixing anything because of that.

O.K.

_Maybe_ it would be worth adding explanation of this to commit message (and I see I should audit gitweb for similar problems elsewhere), but anyway

Acked-by: Jakub Narebski <jnareb@gmail.com>

-- 
Jakub Narębski