The function used to translate a function how requires a
ModuleTranslationState, which contains state collected after reading early
sections of a wasm module. Currently, this is only used in Cranelift for the
multi-value feature (to read block type signatures), so we can use a dummy
state that's shared across all functions.

Gary, Christian: how big an issue it is for you when updating to a specific changeset results in not being able to build? I imagine your bisection script would use the skip option of git bisect, but just wanted to double-check this.

Context of this question: in a bug such as this one, it seems preferable to be able to look at what actual changes happened, separately of the output of the mach vendor rust command, which adds a lot of noise. So we'd prefer to be able to look in retrospect at commits one by one, instead of squashing them, but this will create small commit ranges in which it's impossible to build.

I think we are entirely moving away from compiled bisections and want to use TC builds instead. For that purpose, having small ranges in history that don't build seem fine to me.

In any case, the current bisection scripts work on Mercurial but not Git for now. Historically they have been able to skip "impossible-to-build" changesets, up till a certain arbitrary number.

Good to know, thanks! It's ideal because it means we can land as is, and be able to look at the history of commits and see actual changes. I assume the arbitrary number of skip revisions is low, but greater than or equal to 2, which is our case here.

https://hg.mozilla.org/mozilla-central/rev/307a54c830da
https://hg.mozilla.org/mozilla-central/rev/b321de6e4733
https://hg.mozilla.org/mozilla-central/rev/f1000962724e