On Wed, Nov 30, 2005 at 07:10:37PM -0600, William Rowe wrote: > Colm MacCarthaigh wrote: > >If apr 1.0 or 1.1 happen to be installed, I don't see why it's not > >reasonable to fail to configure. The administrator may intend to link > >against the system version, they may not want httpd having its own > >libapr. And they're the only people capable of making that decision and > >hence resolving the conflict. They can decide to install over their > >existing apr, or to install a new one just for httpd. > > > >I brought this exact issue up weeks ago, and it didn't go very far. I > >was originally -1, for the very same reasons you are, but having thought > >about it decided that yes, while the present system introduces some > >inconvienence for a small fraction of users, it doesn't try to second > >guess them either, and unbundling apr/apr-util would represent a huge > >inconvienence to a large fraction of users. > > I read this a bit backwards of your interpretation; > > * admins who install 1.1 for some specific reason are responsible to > ensure they deal with the new package correctly (e.g., we give them > a message upon configure "Found old APR 1.1.0, installing APR 1.2.2 > for you" and let them decide what to do. 99% of the time, they must > follow our advise and install 1.2.2 in the same prefix/ as httpd.) > > * the vast majority of users, who only have apr 1.0/1.1 due to svn or > other intrapackage dependencies, get a free apr 1.2 without having > to think about it. Make this whole headache a noop for them.
If some random user has APR 1.1 installed in /usr/local/apr, and builds httpd 2.2 with --prefix=/usr/local/httpd-2.2, it would be a Bad Thing (and certainly, very surprising behaviour) if that httpd install went ahead and silently upgraded that APR install. (what if the APR configure options were wrong/different? what if the APR 1.1 build had been custom-patched? etc) Therefore I maintain that the current behaviour having configure fail if the system APR/apr-util is not of sufficient version is the right thing to do. The user can always force the use of the bundled copies (to be installed in the same $prefix as httpd) as had been said many times. > And I for one don't want the headaches of the users@ trouble reports. I'd > really prefer to see those who help out on users@ answering this objection, > as opposed to the hackers who are detached from the user community pushing > this out +1 over those user-supporters objections. Any users who run httpd are unlikely to have installed APR 1.[01] given that APR 1.x has never been supported by an httpd release to date. It's really only httpd/APR developers who are likely to get into this situation. (APR 1.x has never been shipped in a Subversion tarball) joe
