Re: Question and possibly bugs about string encoding in gtk-perl

[Torsten Schönfeld <kaffeetisch gmx de>]

On 16.12.2016 20:02, Dominique Dumont wrote:
> On Wednesday, 14 December 2016 10:54:16 CET Boyuan Yang wrote:
> > The original messy output, as indicated in screenshot in the Ubuntu bug,
> > looks  like treating a latin-1-encoded binary data as UTF-8-encoded data
> > and showing them anyway.
>
> In more details, the problematic code boils down to:
>
>   my $wnck_screen = Gnome2::Wnck::Screen->get_default;
>   my $win= $wnck_screen->get_windows_stacked; # Gnome2::Wnck object
>   my $name = $win->get_name;
>   my $window_item = Gtk2::ImageMenuItem->new_with_label( $name );
>
> $name contains window name apparently in octet format instead of an utf8
> string. As a consequence, the list of windows shown by shutter contains
> mojibake.

Yes, Gnome2::Wnck does not do any decoding at all in get_name.  This is 
probably by accident, not by intention, but if you look at what 
wnck_window_get_name boils down to, 
<https://git.gnome.org/browse/libwnck/tree/libwnck/xutils.c#n1229>, you 
see that it returns UTF8-encoded strings in most cases, but sometimes 
also strings in X11's compound text encoding, according to 
<http://stackoverflow.com/questions/7706589/is-there-any-difference-with-the-x11-atoms-xa-wm-name-and-net-wm-name>.

So it seems like the safest bet would be to try to decode the window 
name from UTF8, and if that fails, try Encode::X11 and its 
'x11-compound-text' (Hi Kevin!).

> > The hacky patch proposed (by me) is using
> > Encode::_utf8_on() to turn on the internal flag for string and mark it as
> > UTF-8.
>
> And I'm worried that shutter may crash if used in a non-utf8 environment.
>
> After some experimentation, I've come up with a safer solution:
>
>   use 5.12.0;
>   use Encode::Locale;
>   use Encode qw/decode/;
>   # ...
>   my $name = decode( 'locale' , $win->get_name);
>
> This works in utf8 locale and is safer than turning utf8 flag on.

It seems like this would fail in a non-UTF8 locale, however, as 
gnome_wnck_window_get_name seems to always return UTF8 strings, 
regardless of locale.