From: Justin Tobler <jltobler@gmail.com>
Date: Mon, 15 Dec 2025 16:47:27 GMT
Subject: Re: [PATCH v2 2/7] strbuf: split out logic to humanise byte values
Message-ID: <kx3qdkm7rbd23hc66qamhq45agzofoppfhqnbbtw5cmjojevsq@2kkxiaem3fp4>
In-Reply-To: <xmqqh5ts88b1.fsf@gitster.g>

On 25/12/15 05:21PM, Junio C Hamano wrote:
> Justin Tobler <jltobler@gmail.com> writes:
> 
> > +char *strbuf_humanise_bytes_value(struct strbuf *buf, off_t bytes, unsigned flags)
> >  {
> > +	int humanise_rate = flags & STRBUF_HUMANISE_RATE;
> > +
> >  	if (bytes > 1 << 30) {
> > +		strbuf_addf(buf, "%u.%2.2u", (unsigned)(bytes >> 30),
> >  			    (unsigned)(bytes & ((1 << 30) - 1)) / 10737419);
> > +		/* TRANSLATORS: IEC 80000-13:2008 gibibyte/second and gibibyte */
> > +		return humanise_rate ? xstrfmt(_("GiB/s")) : xstrfmt(_("GiB"));
> > ...
> > }
> >  void strbuf_humanise_bytes(struct strbuf *buf, off_t bytes)
> >  {
> > -	strbuf_humanise(buf, bytes, 0);
> > +	char *unit = strbuf_humanise_bytes_value(buf, bytes, 0);
> > +	strbuf_addf(buf, " %s", unit);
> > +	free(unit);
> >  }
> 
> The old "strbuf-humanise" used to treat the whole "<number> <unit>",
> e.g., _("%u.%2.2u GiB"), as a single thing to be translated.
> However, the new code requires that in all languages:
> 
>  - Decimal point in number MUST be "." (don't some Europeans prefer
>    comma instead?);
> 
>  - Number MUST come before the unit;
> 
>  - Between the number and the unit, there has to be one and only one
>    SP.
> 
> All of which could be a severe regression from localization's point
> of view.
> 
> The first point among the above three can relatively easily
> remedied.  It is a bit more involved, but it is possible to fix the
> other two, too.

The first point could be addressed by just making "%u.%2.2u"
translatable. To address the others, we could have
strbuf_humanise_bytes_value() output two separate strings (value and
unit) instead of appending the the value and returning the unit. Maybe
something like:

  void humanise_bytes(off_t bytes, char **value, const char **unit)

We could then have another translatable string to configure the format:

  void strbuf_humanise_bytes(struct strbuf *buf, off_t bytes)
  {
    char *value;
    const char *unit;

    humanise_bytes(bytes, &value, &unit);
    strbuf_addf(buf, _("%s %s"), value, unit);
    free(value);
  }

This is certainly a bit more involved setup for translators though. But
maybe it's ok? I'll move forward with something like above in the next
version for now.

Thanks
-Justin