Re: gettext 0.10.37 msgfmt charset naming compatibility on Solaris 8

[Paul Eggert]

> From: Ulrich Drepper <address@hidden>
> Date: 15 May 2001 23:48:31 -0700
> 
> Paul Eggert <address@hidden> writes:
> 
> > It is impossible to write code that will work for _everybody_.
> > Even GNU libiconv, as good as it is, does not do that.
> > However, it is helpful to improve code so that it works for more people.
> 
> No.  Because then everybody would have an excuse to demand her/his
> system to be supported which results in unacceptable bloat.

What kind of bloat are you worried about?

Surely you are not worried about the extra 3 lines of code introduced
into lib/localcharset.c.  They do not affect the runtime efficiency at
all; they only export a symbol, when gettext itself is compiled.

Nor can you be worried about the extra runtime overhead in msgfmt,
as it is quite small and is paid only by the maintainer, when new
messages are translated.

I guess that you must be worried about the hassle of maintaining the
code to do the codeset name translation.  But this code is quite small
-- only 40 lines or so -- and it is generic and is easily understood.
And it is partitioned into its own module, so that you need not worry
about it when you are maintaining the mainline code.

The goal of avoiding bloat is a good one, but it shouldn't override
all other concerns.


> If libiconv is missing something, fix it.

That advice is impractical.  It would take much, much more work than
this small patch.  Besides, not every proprietary vendor agrees how
the translations should be done, especially among the more obscure
character sets that libiconv supports.  No matter how libiconv is
modified, it still won't be entirely compatible with the native iconv.