On 03/16/2010 12:04 AM, Ralf Wildenhues wrote: > * Bruno Haible wrote on Sat, Mar 13, 2010 at 09:49:54PM CET: >> Ralf Wildenhues wrote: >>> Those semantics already have users out there, though. For example GCC >>> relies on --disable-libstdc___v3, so this would be an incompatible >>> change. >> >> OK, the name of the variable is not important to me (the user can define >> his own variable anyway). Here's a revised patch: > > Thanks. The patch looks ok to me, but is lacking testsuite exposure and > NEWS update. I'll let Eric decide if he wants this in before the > release, or at all, with the below squashed in, which is at least a > start at testsuite exposure.
[3 months later...] Yes, this seems nice to have in autoconf, so I've applied the two patches. > > Please note however that using this new feature may introduce > incompatibilites in a package tree where different Autoconf releases are > used. For example, 2.61-generated configure scripts will reject and > fail upon --enable-c++, so somebody using your package in a subtree and > a 2.61-using package in another part of her subtree will be in trouble. Hopefully someone piecing packages together like that can figure out how to rerun a single version of autoconf across the entire tree to unify the behavior. And if you don't use + in a variable name, it won't affect your package. > > Actually, I think standards.texi should be updated to document which > characters exactly are acceptable in a FEATURE name, and the Autoconf > test suite updated to ensure we accept exactly those names. Yes, but that should be a separate report to bug-standards. > > Cheers, > Ralf > > * tests/base.at (AC_ARG_ENABLE and AC_ARG_WITH): New test. > * NEWS: Update. > I've now pushed the combination of Bruno's and Ralf's patches, after adding a combined ChangeLog entry and re-running the testsuite to make sure I didn't see any regressions. -- Eric Blake ebl...@redhat.com +1-801-349-2682 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
