On 2013-06-19 03:37, Warner Losh wrote: > On Jun 18, 2013, at 11:36 AM, Peter Wemm wrote: >>> Installing SVN from ports is very painful because of the huge dependency >>> chain it carries, with the largest being Python and Perl IIRC. >> >> That was one of my main motivations. A 2 year old machine builds this >> in ~35 seconds with *no* footprint. The dependency abomination in >> ports tests my non-violent nature. This is a major source of pain on >> the freebsd.org cluster and after the stealth perl ABI changes without >> a version bump.. well.. > > Thanks for doing something! Although it came as a surprise (and seems > to have broken mips), I have found the developer load for svn isn't > building it once (that's easy, even if it does bring in the kitchen > sink). The problem comes when I update some OTHER port that updates > pyhton or something and boom, svn is busted until I rebuild it....
Python and Perl are pulled in by devel/apr1. I've sent a patch that splits up and simplifies this port to apache@ for discussion. I'll check the other subversion dependencies to see if more can be removed. Getting back to the discussion, the arguments I've seen so far are all about how bad the port is, but those are just reasons to fix the port not to tie the repo to releases. The transition period for the cvs-to- svn switch was way too long (several years) and despite that the switch was still disruptive to everybody's work flow. And now we're setting up exactly the same infrastructure just s/cvs/svn/? I'm also not convinced svnup in base would be good, because it still pulls src directly from svn and that would not be the case with a portsnap-like tool. With portsnap code distribution is separated from the underlying vcs.
signature.asc
Description: OpenPGP digital signature
