On Wed, Dec 7, 2011 at 12:56 PM, Graham Leggett <minf...@sharp.fm> wrote: > On 07 Dec 2011, at 7:32 PM, Jeff Trawick wrote: > >> Using the same autoconf and libtool as before should cost you less >> than 10 minutes, it makes it trivial to review more of the proposed >> tarball, and it essentially guarantees no regressions in certain >> aspects of the release. >> >> What could be better than that? > > Using the same autoconf and libtool as a well supported vendor takes zero > minutes, takes advantage of bug and security fixes provided by that vendor, > and completely eliminates the risk that some manual thing I did messes up the > release.
You are the one introducing variations into the release process. That is what is introducing risk into the release process. You are throwing away fixes in later autotools that we have been shipping with for a number of apr and apr-util releases. As far as how to install, it is nothing more than the obvious mechanism, which looks something like this: wget autoconf cd autoconf ./configure --prefix=$HOME/autotools_for_apr_releases && make && make install cd .. wget libtool cd libtool ./configure --prefix=$HOME/autotools_for_apr_releases && make && make install cd .. Then add the new autotools_for_apr_releases bin director to PATH first so that it gets used instead of the system libtool/autoconf. > People build APR using their vendor supplied compilers and tools which will > change as issues are found and fixed. Freezing the versions of autoconf and > libtool to an arbitrary version makes no sense to me. The "People build APR" part is irrelevant. The versions of autoconf and libtool used so far are anything but arbitrary. They are known to work for APR users on a variety of platforms. You're throwing away fixes we have been shipping with for some time, and AFAICT it is not practical to evaluate what the real impact is. (Some AIX level no longer recognized? Some new fooBSD not handled properly? Who knows?) Perhaps the potential concern with apr-util is much less than with apr (is it just expat that has nitty gritty considerations?), but I still think it is generally wrong to deviate without cause, definitely wrong to backlevel, and aggravating to release testers that the uninteresting changes to configure-generated files make them hard to compare with previous releases.
