[Texi2html] gettext-like framework for string translations

[Patrice Dumas]

Hello,

I just committed an important change, now a gettext-like approach is used
for in document string translations. It is also still possible to use the
former approach.

libintl-perl is used as a gettext implementation, and more precisely the
pure perl implementation is used, to be sure to have a consistent gettext-like
implementation which is not the case if the system one is used. libintl-perl
is shipped in texi2html and installed to be sure that it is available. It is
also possible to use the system libintl (currently decided at build-time).

Things are setup like that (this is also how it was done previously):

* translated strings are texinfo strings, which may have @-commands
* the variables parts of the string are not denoted by %s and the like, but 
  by {arg_name}. This is for 2 reasons, first changing the order of
  printf arguments is only available since perl 5.8.0, second the order
  of the argument may not be predictable when @-commands expansion may lead 
  to different orders depending on the output format.
* When a translated string is needed, it is possible to give a state as 
  argument which determines the context of expansion (use the document state,
  expansion in string, no expansion...).

Here is how things happen:

1. First the string is translated. The locale is @address@hidden

  If the @documentlanguage is like ll_CC, ll_CC is tried first, and then ll. 
  If the encoding is not us-ascii, us-ascii is also tried. The idea is that 
  if there is a us-ascii encoding, it means that all the characters in the 
  charset may be expressed as @-commands. For example there is a fr.us-ascii 
  locale that can accomodate any encoding, since all the latin1 characters
  have associated @-commands. For the ja translations, there is only ja.utf-8
  since there are no @-commands for ja letters.

2. Next the args in string are protected, for example {arg_name} becomes
   @address@hidden

3. Next the string is expanded as a texinfo string. 
   @internal_translation_open_brace{} expands as { and 
   @internal_translation_close_brace{} expands as }, such that in the end
   one still gets {arg_name} within an expanded string.

4. Then the in string arguments are substituted, for example {arg_name} is 
   substituted by the corresponding argument.

(2. and 3. are skipped when there is no expansion).


For example, in the following {date}, {program_homepage} and {program}
are the argument of the string. Since they are used in @uref, their
order in not predictable. The {'duplicate'=>1} means the the document state
should be used when expanding the string. {date}, {program_homepage} 
and {program} are substituted after the expansion, which means that they 
should already be acceptable output.

gdt('This document was generated on @i{{date}} using @uref{{program_homepage}, 
@i{{program}}}.', {
           'date' => $date, 'program_homepage' => 
$Texi2HTML::THISDOC{'program_homepage'}, 'program' => 
$Texi2HTML::THISDOC{'program'} },{'duplicate'=>1});


This approach is a bit complicated, however what is interesting is that
it allows to have translation available in different encodings for charset
that are covered by @-commands, and also to specify how the formatting for
some commands is done independently of the output format but still allow it to
be language dependent. For example, the @pxref string may be:

  see {node_file_href} section `{section}\' in @cite{{book}}

which allows to specify a string independently of the output format but with a
rich formatting that may be differently translated in other languages.


It is also possible to use more regular %s escapes, and also avoid any 
expansion (with 'keep_texi' in the state).


(for the record, I reused the existing translations already in texi2html).

--
Pat