Am Mittwoch, 13. Oktober 2004 10:01 schrieb Jean-Marc Lasgouttes:
> >>>>> "Georg" == Georg Baum <[EMAIL PROTECTED]> writes:
>
> OK, here are some comments. Better late than never ;)
Thanks. Unfortunately I am late with the reply, too ;-)
> Georg> Questions: 1. Do we ever use any other gettext than GNU
> Georg> gettext?
>
> I do not know. A case of interest is probably the gettext used on *bsd.
Ok, then we are on the safe side with messages-2.diff.
> Georg> 2. Is HAVE_LC_MESSAGES ever undefined?
>
> Don't know.
Ok, it does not matter for the second version.
> What happens is that we may need two languages at the same time: one
> for the interface and one for the document strings (this is new to
> 1.4.0cvs). Therefore, the locale has to be reset. I suspect however
> that we could do these changes less often.
I disabled the complete localization in messages.C by doing
return m;
at the very beginning of Messages::Pimpl::get(). Several runs of
time ./lyx-notrans -x 'command-sequence file-open UserGuide.lyx
; lyx-quit ;'
show on average 50 ms less than a normal build, so I guess we don't need
to worry about this point.
> I propose a different solution, which is a mix of what you did and
> what I proposed earlier.
>
> 1/ use a clear notation for translators hints that does not risk to be
> used in normal code, like [[double brackets]] or /*C-style comments*/
I don't like this very much, because we limit the domain of translatable
strings, but I think I would prefer the double bracket version if we are
going to do it like this.
> 2/ Once a string has been translated, remove translators hints (using
> a regexp) if they are still present.
>
> It seems to me that this avoids nicely the use of a en.po.
Why do you want to avoid it? The amount of code is almost the same as for
your proposal. IMHO using en.po is a nice solution (I may say that, it is
Angus' idea, not mine :-) )
> Would that work?
Yes. See attached diff (undefine USE_EN_PO, parts of my previous patch are
needed, too). The regex is preliminary, bacause like this I did not need
to change the dialogs again. The same coarse timing method as above shows
no difference above noise between a normal build, the method with en.po
and your proposal.
Georg
--- lyx-1.4-clean/src/messages.C 2004-06-09 18:53:41.000000000 +0200
+++ lyx-1.4-cvs/src/messages.C 2004-10-15 21:43:16.000000000 +0200
@@ -13,6 +13,11 @@
#include "support/filetools.h"
#include "support/path_defines.h"
+#undef USE_EN_PO
+#ifndef USE_EN_PO
+#include <boost/regex.hpp>
+#endif
+
using lyx::support::GetEnvPath;
using lyx::support::lyx_localedir;
@@ -98,11 +103,37 @@
bindtextdomain(PACKAGE, lyx_localedir().c_str());
textdomain(PACKAGE);
const char* msg = gettext(m.c_str());
+ // Some english words have different translations, depending
+ // on context. These cases are handled with an english
+ // translation: The original string is augmented by context
+ // information (e.g.
+ // "To:[translation hint: as in 'From page x to page y']" and
+ // "To:[translation hint: as in 'From format x to format y']",
+ // and the english translation of both strings is "To:".
+ // This means that we need the english translation as
+ // fallback, if no translation for the current language is
+ // available, otherwise the user sees bogus messages.
+ string translated(msg);
+ if (n && translated == m) {
+#ifdef USE_EN_PO
+ n = setlocale(LC_ALL, "en_US");
+ bindtextdomain(PACKAGE, lyx_localedir().c_str());
+ textdomain(PACKAGE);
+ msg = gettext(m.c_str());
+ translated = msg;
+#else
+ static boost::regex const reg("^([^\\[]*)\\[translation hint:[^\\]]*\\](.*)$");
+ boost::smatch sub;
+ if (regex_match(translated, sub, reg)) {
+ translated = sub.str(1) + sub.str(2);
+ }
+#endif
+ }
setlocale(LC_ALL, old);
free(old);
// If we are unable to honour the request we just
// return what we got in.
- return (!n ? m : string(msg));
+ return (!n ? m : translated);
}
private:
///
