Fixing RTL search in Evince

[Uri Sivan <tartif gmail com>]

Hi,

There's a long-standing bug [1] regarding searching RTL (Arabic, Hebrew) 
strings in evince. This is because text is laid out "visually" instead 
of "logically" in PDF, and that is what poppler searches. This is indeed 
a poppler limitation but as the discussion [2] on freedesktop.org 
explains, it's fiendishly hard to do correctly (essentially, reconstruct 
logical text from visual).

A simple workaround that would help Evince users is to change the search 
string itself from logical to visual before searching. This has no 
effect on LTR text but flips RTL text so it is searched correctly. It's 
not the ideal solution but it solves 98% of the problem, and is 
infinitely better than the user performing the logical-to-visual 
translation manually when typing in the search string. I can point out 
some of the remaining issues if necessary.

I have a working prototype and can finish it and send a patch if this 
solution is acceptable. It generally amounts to replacing the line

job->text = g_strdup(text);

in the function ev_job_find_new() with something like:

job->text = ev_bidi_reorder(text);

where ev_bidi_reorder() is a small helper function that does 
logical->visual using libfribidi.

Uri


[1] https://bugzilla.gnome.org/show_bug.cgi?id=313230, 
https://bugzilla.gnome.org/show_bug.cgi?id=546828

[2] https://bugs.freedesktop.org/show_bug.cgi?id=2981