Re: CVS commit: pkgsrc/lang

[Greg Troxel <gdt%lexort.com@localhost>]

nia <nia%NetBSD.org@localhost> writes:

> On Wed, Jun 03, 2020 at 08:01:06PM -0400, Greg Troxel wrote:

>> The real issue is that it seems that you want is in direct conflict with
>> what other people want, which is that rust (not bin) can be built on a
>> system that's currently pointed at bin.  At least 3 of us have spoken in
>> favor of following our norm of -bin suffix for binary programs, and
>> having bulk builds attempt to build both, so that we can get data for
>> our get-well plan.
>
> I don't think the norm holds up here, but gathering data does sound
> useful (providing people actually read bulk logs...). You're welcome

I actually do look at the bulk build results and have been mailing
people about them offlist, for things that are a big deal.

> to rename the package for that reason, but I do think it's regretful
> if rust only ends up being available under the rust-bin name.

OK - I will do that then.  I do see your point about users and naming.
I think people trying to use rust have already had a lot of difficulty
and "pkgin add rust-bin" is very easy, and is a lesser total harm than
not even trying to do bulk builds from source on rust-bin platform.
Certainly it can be included as a big section in the branch
announcement.

>> It seems also that rust is building in all the pkgbuild builds on the
>> last branch.  The issue seems to have been one of MAKE_JOBS_SAFE.  So
>> perhaps you should be deleting the bin package now.  But I don't mind,
>> and I think no one else does, if it remains -- as rust-bin -- so that we
>> could flip it back on if needed.
>
> It's been building unreliably. There are remaining unresolved issues.

OK - I withdraw the suggestion of flipping back to source.  We can see
over time how the source package builds in bulk build environments.

> There's also the matter of the build process for rust actively driving
> people towards things like firefox52 on supported platforms.

Yes, but that is fundamentally about upstream rust, how beastly it is,
and how there is no reasonable bootstrap path from C.   I agree that us
trying to mitigate that problem is good.

I really am glad you added rust-bin as a package, and I didn't mean to
sound at all opposed to that.   My only issue is the PKGNAME conflict
and the resulting omission of lang/rust from bulk builds.

> It's a beast to bootstrap, and I don't think there's any advantage building
> it ourselves over letting upstream cross-compile it for us. We're relying
> on a difficult to reconstruct binary bootstrapping process regardless.

I see your point, but I don't think we are going to totally give up on
rust-from-bootstrap-and-source in pkgsrc.   I can see a typical path of
using -bin when that is available from upstream as an outcome, and of
course we'll have to see how that goes.