Re: Bug#355367: debuild: Preserving locale settings can affect build

[Karl Berry]

    but what I did is the following:

Wow, thanks!!

    I tested the attached patch against texinfo/makeinfo 4.8 and it worked,

Please say exactly how you tested it, ie, which environment variables
you set to what, what the makeinfo invocation was, what the input file
was.  I have had a great deal of trouble reproducing and testing i18n
stuff in the past.

    Furthermore I don't know WHEN the @docuemntlanguage is set, 

In principle, it could be changed through the document, if the document
was written in more than one language.  In practice, I have never seen
it used anywhere except at the beginning.  Anyway, I don't think it
matters.

    but it might be enough to change it in the respective callback
    function cm_documentlanguage,

Yes, that is the only thing we can do.

    but how to expand the 2 letter code to a full setting?

Well, although I see the manual says a two-letter code, clearly that is
not completely correct, although I don't recall anyone writing to me
about it.  If someone wanted to differentiate German German from
Austrian German, they should be able to use address@hidden de'
for the first and address@hidden de_AT' for the second.

So the answer is, make a lookup table which translates the two-letter
codes to their respective "common" value, e.g., fr->fr_FR, de->de_DE, etc.

Question: is de_DE, fr_FR, etc., really needed for gettext?  If you just
say "de" to gettext (in some way), doesn't it guess you mean de_DE?

Anyway, the only way in which @documentlanguage is used now is that for
some code CC, texinfo.tex tries to \input txi-CC.tex.  So there is no
reason why we couldn't have txi-de_AT.tex, if it was needed.

Just for my own information, are there in fact any differences with
German German and Austrian German with respect to these few words?  Take
a look at txi-de.tex and see if any of it would be different in Austria.
(The list of words there should also give some clues about other places
to apply __.)

      one might have managed to fix this in the last 5 years since the
      bug is open).

One would think.  I was scared off by Bruno's telling me that the
"simple" way to switch languages (which you implemented) was wrong
somehow.  I find gettext quite "magical", overall.

Now, some comments on the patch:

    +char interface_language[21] = "C";
    ...
+         strncpy(interface_language, optarg, 20);

21?  No.  Make it a char * and use xstrdup on optarg.

    +char *getdocumenttext (const char *msgid)
    ...
    +  safedlocale = setlocale(LC_ALL,NULL);

Not that it is important, and not that I'm complaining, but GNU style
is to format code like this:

char *
getdocumenttext (...)
...
  safedlocale = setlocale (LC_ALL, NULL);  /* with spaces */

Also, "safed" should be "save_" :).


Thanks very much for all this.  Actual progress, after lo these many
years :).

Best,
k