Re: PATCH: realpath() and dead code

[Oliver Kiddle <okiddle@xxxxxxxxxxx>]

Mikael Magnusson wrote:
> This breaks cross-compilation, or requires manually figuring out the
> answer and injecting it in the autoconf cache. (depending on your
> definition of "break").

As I mentioned, it uses the existance or otherwise of
canonicalize_file_name in the action-if-cross-compiling part. (I'm not
aware of any system that has canonicalize_file_name but doesn't accept
NULL).

Even with a cross-compile action of zsh_cv_func_realpath_accepts_null=no
you won't have a broken build, just a suboptimal one. And without the
patch you currently get a similarly suboptimal build on FreeBSD when
compiling natively. Same goes for all the other BSDs including Darwin
judging by online man pages.

Aside from that, there's already 23 instances of AC_TRY_RUN in
configure.ac so compromising cross-compilation is surely not new or
unique to this. Do you regularly cross-compile zsh? Is this really
making a difference that wasn't there before? Or do you perhaps have
another suggestion? The only other way I can think of is to explicitly
#ifdef known systems. Anyone else got any thoughts?

Oliver