On Wed, Jan 02, 2008 at 09:47:32PM +0100, Simon Bertrang wrote: > On Tue, Jan 01, 2008 at 03:11:13PM +0100, Eric Faurot wrote: > > Hi and happy new year. > > > > While working on another port, simon and I noticed that the current doxygen > > is broken. Generated HTML filenames for directory listing are constructed by > > md5 summing the original path, but the md5 algorithm used is broken on 64bit > > archs, so the PLIST differed. This is fixed in recent versions. Now another > > problem is that the original path itself (the one where the sources to be > > documented are located) may vary when building a port needing doxygen: one > > could use WRKOBJDIR=/usr/obj/ports, or the port could be built from > > /usr/ports/mystuff. This is very inconvenient for PLISTs. > > > > So, I thought the simplest/nicest way to fix this would be to use an env var > > set to the prefix to strip from the paths used to generate the md5 sum. > > Ports > > using doxygen would just have something like: > > > > MAKE_ENV= ENCODE_DIRNAME_PREFIX_STRIP=${WRKSRC} > > > > (the variable name is not too bright maybe) > > > > As the chcksums should IMO be built relative to WRKSRC anyway i think > the prefix strip does exactly what it should. > > > Diff with update and hack attached. > > > > Comments? > > > > Checked with your gdal port on amd64 - no plist changes anymore. > > Ok from my side.
Builds and installs fine. I didn't test ports using it, but on a small app here @i386 with an old doxyfile, works fine. Diff ok for me too, and hack seems nice. Maybe mentionning it in DESCR would be useful for future ports using it ? Landry