Re: an i18n sample

[Daniel Jacobowitz <drow at false dot org>]

On Sun, Oct 24, 2004 at 09:23:43PM +0200, Eli Zaretskii wrote:
> > -          fprintf_unfiltered (gdb_stderr, "Type ");
> > +          fprintf_unfiltered (gdb_stderr, _("Type "));
> >            if (type == NULL)
> > -            fprintf_unfiltered (gdb_stderr, "(null)");
> > +            fprintf_unfiltered (gdb_stderr, _("(null)"));
> >            else
> >              type_print (type, "", gdb_stderr, -1);
> > -          error (" is not a structure or union type");
> > +          error (_(" is not a structure or union type"));
> 
> This fragment should be rewritten to use full sentences, like in
> "Type (null) is not a structure or union type", instead of marking
> each part separately.  That way, a translator will be able to
> translate the entire sentence as a single entity, rather than each
> part.  (Translating sentences in parts usually leads to bad
> translations, unless the translator makes an effort to read the
> source.)

I agree about partial translations, of course.  In this particular
case, it looks like the file needs type_sprint or type_to_string.
The type_print interface is unfriendly for i18n.

> > @@ -6005,7 +6005,7 @@ empty_record (struct objfile *objfile)
> >    TYPE_CODE (type) = TYPE_CODE_STRUCT;
> >    TYPE_NFIELDS (type) = 0;
> >    TYPE_FIELDS (type) = NULL;
> > -  TYPE_NAME (type) = "<empty>";
> > +  TYPE_NAME (type) = _("<empty>");
> >    TYPE_TAG_NAME (type) = NULL;
> >    TYPE_FLAGS (type) = 0;
> >    TYPE_LENGTH (type) = 0;

I'm not sure this sort of string should be translated.  I guess it does
no harm (if nothing checks for the string, of course).

> > @@ -6167,8 +6167,8 @@ ada_template_to_fixed_record_type_1 (str
> >       the current RTYPE length might be good enough for our purposes.  */
> >    if (TYPE_LENGTH (type) <= 0)
> >      {
> > -      warning ("Invalid type size for `%s' detected: %d.",
> > -               TYPE_NAME (rtype) ? TYPE_NAME (rtype) : "<unnamed>",
> > +      warning (_("Invalid type size for `%s' detected: %d."),
> > +               TYPE_NAME (rtype) ? TYPE_NAME (rtype) : _("<unnamed>"),
> >                 TYPE_LENGTH (type));
> >      }
> 
> Same here.

Actually, I disagree about this one.  Whether or not "<unnamed>"
should be translated, is there any reason not to translate the format
string?

> > @@ -537,12 +537,12 @@ print_variant_clauses (struct type *type
> >  
> >    for (i = 0; i < TYPE_NFIELDS (var_type); i += 1)
> >      {
> > -      fprintf_filtered (stream, "\n%*swhen ", level + 4, "");
> > +      fprintf_filtered (stream, _("\n%*swhen "), level + 4, "");
> >        print_choices (var_type, i, stream, discr_type);
> >        fprintf_filtered (stream, " =>");
> >        if (print_record_field_types (TYPE_FIELD_TYPE (var_type, i),
> >  				    outer_type, stream, show, level + 4) <= 0)
> > -	fprintf_filtered (stream, " null;");
> > +	fprintf_filtered (stream, _(" null;"));
> >      }
> 
> This also, and in addition I think it plays character-counting games
> that are no-no in translatable messages.

It's only character-counting blank space, to indent appropriately -
since I don't believe it's aligning with an earlier line, it should be
OK.  That said, the "when" is a partial sentence, and the \n%*s doesn't
need to be translated.

-- 
Daniel Jacobowitz