Re: gcc BZ 112671

[Bruno Haible]

[CCing bug-gettext, since AM_ICONV is documented as part of GNU gettext]

Hi Arsen,

> Would it be correct to set INCICONV and {LT,}LIBICONV when configuring
> the tree instead of (or in addition to) modifying iconv.m4?

The interface of iconv.m4 is defined here:
https://www.gnu.org/software/gettext/manual/html_node/AM_005fICONV.html
If you want to achieve compatible effects for an in-tree build, it is
  - necessary, yes, to define LIBICONV and LIBICONV,
  - but not effective to set the variable INCICONV. What the compatible
    macro should do, is to add the -I option to CPPFLAGS.

The reason why my autoconf macros add the -I option to CPPFLAGS, not
merely to some variable such as INCICONV, is that generally it is more
reliable to build an entire package with the same set of -I options.
Adding a -I option to a compilation unit here, and another -I option
to another compilation unit there, is fragile: In these directories
there may be other .h files than iconv.h, say foo.h. In some compilation
units '#include <foo.h>' would reference /usr/include/foo.h, in some
other compilation units it would reference INCICONVDIR/foo.h.
(This is, of course, what a build system does for its internal
dependencies. But for external dependencies 'foo', such a thing can
easily lead to arbitrarily complicated bugs.)

Bruno