On 31.08.2017 00:42, Greg Troxel wrote: > > Kamil Rytarowski <n54%gmx.com@localhost> writes: > >>> But we're only ~20 days from freeze start, and this structurally matches >>> some recent destabilizing changes. In general I'm opposed to signficant >>> large-scale-impact changes right before freeze, such as (aside from >>> micro/security updates) upgrading perl, clang, or gtk3, or changing php >>> versions. (I realize gtk3 and php haven't caused trouble, and the >>> fallout from the new clang was minor, but nobody expected perl to cause >>> trouble, and it really did. I don't mean to object to progress, just to >>> separate recovery from progress from the freeze.) >> >> I would like to get upgraded clang to 5.0.0, it's going to be released >> just now or in few days. It has been already released with the latest RC >> and is busy waiting for feedback. > > I really wonder if clang should be versioned in pkgsc, like gcc. > Presumably you are talking about changing the clang package, not adding > a clang5 package. > > I fully expect there to be fallout from clang moving to 5 from 4. I > remember, not very clearly, some from 3.3 to 3.4, but I could be off. > I'm not saying 5 is buggy, but it seems likely that it will catch errors > in code that weren't caught before and result in build failures. Or > perhaps you (or Joerg?) have done a full bulk build with the new clang > vs the old clang, and know that this worry is unfounded. Most of these > issues can probably be fixed fairly quickly, but right before freeze is > a bad time since it imposes churn on the freeze process. > > I've performed quasi bulk-builds to catch problems with Clang/LLVM in pkgsrc packages with something close to 5.0.0. The only ones were fixed in pkgsrc, NetBSD-current and backported to -8. This clang upgrade will involve llvm, clang, lldb, polly etc. We don't version clang and perhaps we shouldn't unless there is a good reason like a dependency from an important package. Clang is rather ascending with the number of supported features, backends, frontends etc; contrary to GNU toolchain that keeps reducing undermaintained code.
Attachment:
signature.asc
Description: OpenPGP digital signature