On 7/9/24 09:51, Paolo Bonzini wrote:
On Tue, Jul 9, 2024 at 2:18 PM Daniel P. Berrangé <berra...@redhat.com> wrote:
My thought is that the initial merge focuses only on the build system
integration. So that's basically patches 1 + 2 in this series.
Patch 3, the high level APIs is where I see most of the work and
collaboration being needed, but that doesn't need to be rushed into
the first merge. We would have a "rust" subsystem + maintainer who
would presumably have a staging tree, etc in the normal way we work
and collaborate
It's complicated. A "perfect" build system integration would include
integration with Kconfig, but it's not simple work and it may not be
Manos's preference for what to work on (or maybe it is), and it's also
not a blocker for further work on patches 3-4.
On the other hand, patches 3 and 4 are _almost_ ready except for
requiring a very new Rust - we know how to tackle that, but again it
may take some time and it's completely separate work from better build
system integration.
In other words, improving build system integration is harder until
merge, but merge is blocked by independent work on lowering the
minimum supported Rust version. This is why I liked the idea of having
either a development tree to allow a merge into early 9.2.
On the other hand, given the exceptional scope (completely new code
that can be disabled at will) and circumstances, even a very early
merge into 9.1 (disabled by default) might be better to provide
developers with the easiest base for experimenting. The requirements
for merging, here, would basically amount to a good roadmap and some
established good habits.
An merge into early 9.2 would be a bit harder for experimenting, while
merging it now would sacrifice CI integration in the initial stages of
the work but make cooperation easier.
I would be in favor of a 9.1 merge (disabled, not auto).
Noting that soft-freeze is two weeks from today, so no dilly dallying. :-)
The only reasonable alternative that I see is the kind of development branch that qemu is
not used to having: a long-term shared branch on qemu-projects. With which I would be ok,
up to and including a branch merge at the end, as I'm used to working that way with other
projects.
I think the only reason to delay the current development work until 9.2 is if we were
still questioning whether using Rust *at all* is a good idea. I think we're beyond that
point.
r~
