Re: Rust and Q-branching

[Havard Eidnes <he%uninett.no@localhost>]

One thing, though...

As you may know, rust carries in its belly (as part of the source
distribution) a copy of LLVM.  You can choose to build rust with
this integrated copy of LLVM by setting

PKG_OPTIONS.rust+=      rust-llvm

The default is to use whatever other LLVM is available, be that
either an in-tree version or from the llvm pkgsrc package.

Now, in most of my build setups, /etc/mk.conf contains a line
such as the one above.  That means that I'm not testing the build
without that option set.

I suspect there's reasons why I'm using that option, again my
memory is a bit hazy, but I seem to recall not having it set
proved problematical in the past.  I also now have verification
that it's needed for the NetBSD/macppc 8.0 bulk build of rust to
succeed (1.52.0 from pkgsrc-2021Q2), and I also have a suspicion
that the i386 build needs a similar workaround, although I've not
seen a final confirmation of that.

Obviously, building the internal LLVM as part of the rust build
will prolong the build time for the package which is significant
enough by itself already.

The question I have is whether folks consider changing the
default setting for this option, and how wide-ranging we should
make that option default.  NetBSD/i386 and NetBSD/powerpc only,
NetBSD only, or make it the global default?  I'm leaning towards
the latter, so that any rust-local changes to the internally
carried LLVM are used and no version skew against either in-tree
LLVM or pkgsrc LLVM can be allowed to disturb the rust build
result.

Comments?

Regards,

- Håvard