src translations follow-up: Add ngettext support at the C level

[MichaelChirico]

While this is not used by any CRAN packages the last I checked:

https://github.com/search?p=1&q=org%3Acran+dngettext&type=Code

^ Note that there are some results, but they all appear to just be copy-pastes of the #define of the dngettext macro suggested in WRE.

base does use this a few times:

./src/library/base/po/R.pot:msgid_plural "cannot allocate memory for registered native symbol (%d bytes)"
./src/library/base/po/R.pot:msgid_plural "%d arguments passed to '%s' which requires 1 or 2 arguments"
./src/library/base/po/R.pot:msgid_plural "%d arguments passed to '%s' which requires %d to %d"
./src/library/base/po/R.pot:msgid_plural "the ... list contains fewer than %d elements"
./src/library/base/po/R.pot:msgid_plural "Warning messages:"
./src/library/base/po/R.pot:msgid_plural "There were %d warnings (use warnings() to see them)"
./src/library/base/po/R.pot:msgid_plural "unused arguments %s"
./src/library/base/po/R.pot:msgid_plural "unprotect(): only %d protected items"
./src/library/base/po/R.pot:msgid_plural "R_Reprotect: only %d protected items, can't reprotect index %d"
./src/library/base/po/R.pot:msgid_plural "%d arguments passed to .Internal(%s) which requires %d"
./src/library/base/po/R.pot:msgid_plural " [ reached getOption(\"max.print\") -- omitted %d rows ]\n"
./src/library/base/po/R.pot:msgid_plural "numerical expression has %d elements: only the first used"
./src/library/base/po/R.pot:msgid_plural "%d arguments passed to '%s' which requires %d"

So support will be required to get a complete R translation with translate_package() alone.

[MichaelChirico]

Note: despite some ambiguity in the docs (as I recall), R is in fact hard-coding that it's using N_ as the plural macro:

https://github.com/wch/r-source/blob/b87d430b5e2398b2d48e124b8e8b23fde3c68b57/src/library/tools/R/translations.R#L180

[MichaelChirico]

An issue is that ngettext() is usually called directly, e.g. in errors.c

658:	    REprintf(ngettext("There was %d warning (use warnings() to see it)",
659-			      "There were %d warnings (use warnings() to see them)",
660-			      R_CollectWarnings),
661-		     R_CollectWarnings);

[MichaelChirico]

A quick note here that xgettext applies c-format when any of the inputs are templated, see

cd /tmp
echo 'ngettext("a", "b");' > test.c
echo 'ngettext("a %d", "c")' >> test.c
echo 'ngettext("d", "b %d")' >> test.c
xgettext -o /dev/stdout test.c | tail -n +21

with output:

#: test.c:1
msgid "a"
msgid_plural "b"
msgstr[0] ""
msgstr[1] ""

#: test.c:2
#, c-format
msgid "a %d"
msgid_plural "c"
msgstr[0] ""
msgstr[1] ""

#: test.c:3
#, c-format
msgid "d"
msgid_plural "b %d"
msgstr[0] ""
msgstr[1] ""

[MichaelChirico]

Note that while the thing about c-format mismatching in msgid and msgid_plural is true for xgettext, tools:::en_quot will prevent this from being allowed. msgfmt requires msgid_plural and msgid[k] to match on templates, but for en@quot, msgid[0] is just msgid, so if msgid and msgid_plural don't match on templates, this will fail.

So for now I'm leaving the logic for aggregation intact, but untested.