The attached patch applied over current ebuild "fixes" a little problem with berkdb includes. I used the quotation marks as it's not a problem for general users, but it's a real problem for fbsd users. Little explanation: /usr/include/db.h is a system include in freebsd. At the moment sys-libs/db creates a link db.h which points to db4.2/db.h, and this is not good. I have modified the eclass to skip that link, but i'm not sure for the fix as for now. The attached patch make sure that the db.h file included by subversion is the one in db4.2 and not the system include. This changes nothing for normal users, but makes subversion work for fbsd users. I also removed the autoconf dying when found 2.58, as the depends states clearly 2.59. P.S.: i've CCed pauldv as it's subversion's maintainer. HTH, Diego
Created attachment 53700 [details, diff] subversion-1.1.3-r1.patch
I'm also the db maintainer. I'd like your suggestions how to once and for all fix the db problems.
Created attachment 53923 [details, diff] db.eclass patch Good to know :) This is the patch I used on g/fbsd, which simply removes inclues which are put in /usr/include. This would create troubles to packages, similarly as it does with subversion, so they should add the right directory to the include path.
Created attachment 55356 [details, diff] subversion-1.1.4.patch A little more of a cleanup, removing a few $myconf lines and moved them directly on econf call.
I'll try to look at this tonight. The include files must however not be 4.2 but the ones that correspond to the db version that apr-util was build against.
This bug does NOT apply just to BSD, why don't change the Product? Thanks.
Well actually on non-bsd systems there shouldn't be too much problems with that. Main problem is that /usr/include/db.h is a system include instead of a local one which is on Linux. The other cleanups in the patch are mainly cosmetics. Oh, Paul if the problem is with apr, why not just modify apr-config script in apr to point to the right include path directly? This would require a pair of changes more, but should work, right?
The fact is that until I've tried thr subversion-1.1.4.patch I had not been able to compile this.
Diego that's right. Fransisco, this is most a problem for berkdb. But to get rid of many problems I might decide to remove the whole symlink business for the header files allthough that would require packages to be more explicit in their berkeley db header files.
The patch for 1.1.4 applies cleanly over 1.1.4-r1 (with offset, of course).
Fixed with latest version which uses apr-util (when bug #95468 will be fixed).
