|
From: Martin C. <cos...@wa...> - 2010-06-14 13:25:27
|
Alexander Hansen wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > On 6/14/10 7:22 AM, Duncan Galloway wrote: >> Hi Martin, >> this doesn't seem to do anything (assuming I'm doing what you >> suggest)... >> >> > fink update automake1.5 >> Information about 10377 packages read in 1 seconds. >> No packages to install. >> >> ... and fink update-all fails just as before. I even tried fink remove >> automake1.5, but that didn't work either >> Any other ideas? >> D. >> > > > The message was saying that automake1.5 had been installed, and it > wanted autoconf, rather than autconf2.6, so this probably won't work. > > The most robust workaround is not to use 'update-all', and to update > packages individually, or in smaller subsets, to simplify the build tree > not to try to want both automake1.5 and autoconf2.6 at the same time. The only package that is still interested in automake1.5 is gettext. Perhaps if you update gettext first, then removing automake1.5 will work. What this whole thing shows is that the automake packages are abusing the fact that the autoconf packages are not marked builddependsonly, which they certainly should. They are the prototype of packages that are only needed at build time and should be freely exchangeable. Packages builddepending on automake should know themselves which version of autoconf they need to builddepend on at the same time. I remember that I recently was in a similar automake/autoconf hell, but unfortunately I forgot how I got out of it. I probably modified the automake package so that it depended not on a single autoconf package, but on a choice of several. This is what the recent automakex.y packages with x>=9 are doing. Or maybe I removed all [Build]Dependencies in automakex.y.info. -- Martin |