From: Jiang Xin <worldhello.net@gmail.com>
Date: Tue, 16 Dec 2025 06:18:06 GMT
Subject: Re: [PATCH v2 2/7] strbuf: split out logic to humanise byte values
Message-ID: <CANYiYbExjGoCw4n92a75xtREE_EhjEySVSmk=NwJd3GoMAoVLg@mail.gmail.com>
In-Reply-To: <xmqqqzsv3uus.fsf@gitster.g>

On Tue, Dec 16, 2025 at 12:37 PM Junio C Hamano <gitster@pobox.com> wrote:
>
> Jiang Xin <worldhello.net@gmail.com> writes:
>
> > On Sat, Dec 13, 2025 at 6:37 AM Justin Tobler <jltobler@gmail.com> wrote:
> >> +               return humanise_rate ?
> >> +                              /* TRANSLATORS: IEC 80000-13:2008 byte/second */
> >> +                              xstrfmt(Q_("byte/s", "bytes/s", bytes)) :
> >> +                              /* TRANSLATORS: IEC 80000-13:2008 byte */
> >> +                              xstrfmt(Q_("byte", "bytes", bytes));
> >
> > We have already defined "byte" as a 10n string without plural forms in the
> > file "t/helper/test-simple-ipc.c" via commit 36a7eb6876 (t0052: add simple-ipc
> > tests and t/helper/test-simple-ipc tool, 2021-03-22 10:29:48 +0000).
> >
> >     OPT_STRING(0, "byte", &bytevalue, N_("byte"), N_("ballast character")),
> >
> > The newly introduced usage of "byte" is now marked as having a plural form
> > (via Q_("byte", "bytes", bytes)), which causes a conflict. This results in make
> > pot failing with the following error:
> >
> >     msgcat: msgid 'byte' is used without plural and with plural.
> >
> > This happens because gettext requires that a given msgid be treated
> > consistently—either exclusively as a singular string or as part of a plural
> > construct—but not both.
> >
> > To resolve this conflict, we can unmark the singular "byte" in
> > t/helper/test-simple-ipc.c, allowing it to reuse the translation from the
> > plural-form definition of "byte".
>
> I learned a new thing today and am happy :).
>
> But how does one "unmark" the singular "byte" there, exactly?
>
> Would something like this ...
>
>      OPT_STRING(0, "byte", &bytevalue, Q_("byte", "bytes", 1), N_("ballast character")),
>
> ... a good idea, to "mark" it as a countable noun that has a plural
> form?
>
> Or did you mean that we can simply drop N_() around it, i.e.,
> N_("byte") -> "byte", to discard the i18n, because it merely is a
> test helper?

I prefer dropping N_() for "byte" in "t/helper/test-simple-ipc.c", and
the i18n for the test helper will continue to work as before if we also
mark the plural-form of "byte" in this patch series. (i.e., drop the N_()
for "byte" in the test helper in this patch.)

This is because N_() is a macro that does not invoke any gettext
function, only returns msgid as in gettext.h:

    #define N_(msgid) msgid

And the actual translation for the msgid (the argh field of an option)
occurs later by calling:

    opts->argh ? _(opts->argh) : _("...")

in "parse-options.c".

However, replacing N_() with Q_() would cause the string to be
processed by gettext twice: once at runtime via Q_(), and again
when _(opts->argh) is evaluated.