Re: Environment variable $U

[Eric Blake]

According to Ralf Wildenhues on 2/8/2010 11:32 PM:
> Sounds like a bug to me, $U should be initialized somewhere, or removed
> from the LIBOBJS/LTLIBOBJS variables if ansi2knr is not used.

Aha.  In many projects, U _is_ initialized, but by automake during
AM_C_PROTOTYPES, and not by autoconf.  IIUC, the problem can only occur if
you are using autoconf but not automake.  So one real fix would be to quit
using U in autoconf, and to teach automake how to override
_AC_LIBOBJS_NORMALIZE to re-support U; but that has the drawback of
requiring upgrades in both automake and autoconf simultaneously.  Here's a
less-invasive fix, that only affects autoconf.

tests/semantics.at:AC_REPLACE_FUNCS also uses $U, so we could see a
spurious test failure; I'm not sure what to do there.  Thoughts before I
apply this to autoconf?

>  Or we
> could also go ahead and remove ansi2knr support (after deprecating it
> suitably).

It's always been a desire of mine to switch to C89 as default and require
extra effort on the part of people still wanting K&R, rather than the
current situation of trying to support both by default.  But it takes lots
of time and energy to implement, and hasn't been a high priority for me.

-- 
Don't work too hard, make some time for fun as well!

Eric Blake             address@hidden

autoconf.patch251
Description: Text document

signature.asc
Description: OpenPGP digital signature