--disable-nls vs. libiconv

[Jeff Sturm]

On Fri, 9 May 2003, Ranjit Mathew wrote:
> > Jeff> a) looks like the current behavior, but I don't remember if this
> > Jeff> was a deliberate choice or not.
> >
> > I don't think it was.  I don't remember thinking about --disable-nls
> > at all.
>
> I think PR java/2812 could explain this a bit. I'm guessing that
> AM_ICONV is SNAFU-ing slightly on the submitter's build machine.

My finding is that libiconv is correctly detected for the
*bootstrap* compiler.  However, for the stage1 compiler to link with the
same libiconv, it may need -L and/or -I arguments that aren't yet provided
by the configure machinery.

The fix for java/2812 solved a different problem, that gcj couldn't
correctly detect and use iconv() due to the use of renaming macros.

> Whether iconv( ) is used or not when --disable-nls is specified
> depends on what *exactly* we take this configure option to
> mean - does it only mean that the compiler messages would be in
> the standard US-English or even that the compiler cannot parse
> non-ASCII files.
>
> The latter might not augur well with what Java compilers are
> supposed to be able to do, AFAICT.

I agree.

Jeff