Re: [PATCH 2/4] repair: add canonical names for the XR_INO_ constants

["Darrick J. Wong" <djwong@kernel.org>]

On Tue, Dec 02, 2025 at 08:33:07AM +0100, Christoph Hellwig wrote:
> On Mon, Dec 01, 2025 at 02:47:16PM -0800, Darrick J. Wong wrote:
> > BUILD_BUG_ON(ARRAY_SIZED(xr_ino_type_name) != XR_INO_MAX);
> 
> Or static_assert for the standard userspace version?
> 
> > > +_("size of %s inode %" PRIu64 " != 0 (%" PRId64 " bytes)\n"),
> > > +				xr_ino_type_name[type], lino,
> > 
> > i18n nit: In the old code, the inode description ("realtime bitmap")
> > would be translated, now they aren't.  I don't know of a good way to
> > maintain that though...
> > 
> > "el tamaño del realtime bitmap nodo 55 != 0"
> > 
> >                ^^^^^^^^^^^^^^^ whee, English in the middle of Spanish!
> > 
> > I guess you could write "_(xr_ino_type_name[type])" for that second
> > argument.
> 
> Seems like the standard gettext thing to do is to mark the array
> initializers as N_(), which we do in quite a few places.  Not sure
> where the actual translation actually gets applied with that, though.

That N_() thing is #defined out of existence for compilation in
platform_defs.h:

# define N_(x)			 x

and include/buildrules passes "--keyword=N_" to xgettext so that it'll
find all the N_("string") strings and put them in the message catalog.
I think you still have to use regular _() (aka gettext()) when you pass
them to printf.

IOWs, you'd still have to define the strings array like this:

static const char *xr_ino_type_name[] = {
	[XR_INO_UNKNOWN]	= N_("unknown"),
...
};

to capture "unknown" in the .po file, but I think the printf argument
still has to call gettext() to look up the localized version.

--D