Jonathan Perkin <jperkin%joyent.com@localhost> writes: > * On 2021-09-03 at 13:10 BST, Greg Troxel wrote: > >>It is really unfortunate that the rust world thinks it is ok to depend >>on really recent rust. 1.54.0 was only released 36 days ago. So >>separately from our stability concerns about updates close to freeze, I >>don't think it's a problem to defer those updates until after branching. > > I know I'm biased in favour of Rust, which may cloud my judgement, but > I think it's important to realise that software in general is moving > towards more frequent releases and a more aggressive expectation of > things being up-to-date, and it's just something we're going to have > to deal with or be left behind. Yes, people are doing that, and we do have to cope. We don't have to think it is reasonable. Frequent releases is fine, but the idea that packages in rust depend on a days-old rust release is not ok. That's really pointing out that rust is not a stable language, but a family of langauges with a new member every few months. It is somewhat amusing to see Linux Long Term Stable on one hand and "if you compiler is 45 days old it is obsolete". on the other. I guess this ends up with rust moving itself outside the packaging systems. We do have a scheme to deal with it, and that leads to quarterly branches perhaps being a bit behind, but stable, and so far I think that's ok. What we really need is more automated preparation of bootstraps and testing, probably, but that's separate from what we're doing today. > That of course doesn't excuse drive-by updates, and we should be as > careful as possible, but it should mean we can be more confident that > as long as Rust itself builds on all important platforms we should be > good to go, at least compared to some other languages. Right now the testing/discussion/approval is only required in the two weeks leading up to the freeze. I'm willing to approve based on firefox testing on a few places and rust building on others, if we get there. I completely agree with your idea of having adequate testing but not requiring sillly testing. I gather you have pkgsrc-joyent often updated ahead of pkgsrc, but can you confirm that wip/rust is ok for illumos (for whatever version/cpu pairs matter)? Or perhaps everybody on illumos uses your pkgsrc, and we can permanently cross it off our list of worries? (serious q, not trying to be difficult)
Attachment:
signature.asc
Description: PGP signature