Re: AC_OUTPUT_COMMANDS

[Akim Demaille]

>>>>> "Ralf" == Ralf Corsepius <address@hidden> writes:

Ralf> Akim Demaille wrote:
>> Err, maybe $INSTALL should be preserved too?
Ralf> Let me put it this way: All well-known (may they be explictly
Ralf> documented or not :) shell-variables should be preserved.

I agree.

Ralf> This comprises includedir, libdir, target, host, build, prefix,
Ralf> exec_prefix and many more.

Yep, this is why I changed AT_CHECK_CONFIGURE so that it checks that
those variables are preserved.

Ralf> AFAIS from my foo-0 example, the only relevant variables this
Ralf> happens for are top_srcdir, srcdir and INSTALL - Therefore, I
Ralf> would say yes, INSTALL should be preserved, too. If other
Ralf> variables somehow can make it into the responsible config.status
Ralf> section (I simply don't know :), they also should be preserved.

The thing is you've put the requirement much higher than it was
before: you want us to preserve vars inside config.status too.  I'm
not saying you're wrong, just that we (I) never thought about this
before.

Lemme ask a question though: if $prefix is NONE, then it's defaulted
to /usr/local by AC_OUTPUT.  That's obviously needed for the
AC_SUBST.  Would you say it should be preserved to NONE?  If you
answer yes, please be aware that it complicates our matter quite a
lot, since AC_SUBST(foo) substitutes $foo to @address@hidden  So it is somewhat
unnatural for prefix and exec_prefix, while for INSTALL, srcdir and
top_srcdir it was only a matter of prefixing them with ac_.

As for INSTALL, I'll do that in a separate patch, that one is big
enough as is.  The test suite likes it 100%.  OK to install?