Tobias Nygren <tnn%NetBSD.org@localhost> writes:
> There's no fundamental reason why i386 should be treated differently
> from amd64 (or NetBSD from SunOS, or Linux). This is how we end up
> with random library breakage like this. We need to move away from the
> ad-hoc nature of how this currently works. I would like the bootstrap
> process to be:
Basically agreed, but as I see it there is
- A: a build-from bootstrap method (as you describe) that compiles lang/rust
- this package, or something, can also build a new bootstrap
- B: on platforms where upstream publishes binaries, and those work, a
rust-bin package, as a way to cope when plan A does not work
- some consensus decision when to set the default RUST_TYPE to bin vs
src, wtih the notion that if ought to be bin, then we have a problem
with plan A and should have a get-well plan
> - documented
> - built from pkgsrc sources, with pkgsrc patches
> - reproducable
> - automated
> - testable
> - as statically linked needed to get rid of compat80
> this could be togglable with a "static" PKG_OPTION
It could, but I'd be fine with static only unless somebody wants to do
the work to make dynamic optional
> - the same for all pkgsrc platforms
> - the same for all architectures
>
> Obviously this won't happen in time for the freeze so someone will have
> to manually mend the i386 breakage for now.
True. Are you able to build a NetBSD/i386 bootstrap that is statically
linked? Somebody else?
> I think it's not that much work to set something up on Gitlab CI that
> can build and publish for all our currently supported platforms. I will
> try it out during the freeze.
We need to be careful about chain of custody, but CI that reports if it
works or not would be great.
Attachment:
signature.asc
Description: PGP signature