fix miri argument unpacking issue

[dwrensha]

This should fix rust-lang/miri#360. I haven't actually tested it.

I believe this fix is correct because a similar fix works for Seer: dwrensha/seer@a071e1c

cc @oli-obk

[rust-highfive]

r? @michaelwoerister

(rust_highfive has picked a reviewer for you, use r? to override)

[oli-obk]

I don't think this'll work for ((),(),((), &i32)) or similarly nested things.

[michaelwoerister]

r? @eddyb

[eddyb]

IMO this should "just" do untupling by doing what arg0 = tuple.0; arg1 = tuple.1; ... would do.
Which means using project_field or w/e and avoid all of this broken makeshift logic.

[kennytm]

Hi @dwrensha, just to check if you're still working on this PR 😊

Could you apply @eddyb's suggestion in #47239 (comment)?

[dwrensha]

I've updated Seer with a somewhat simplified version of this, including the testcase @oli-obk was concerned about: dwrensha/seer@f2962ea.

Unfortunately, the https://github.com/solson/miri testsuite is not currently runnable at rustc master (see rust-lang/miri#361), so it is difficult to verify whether changes like this are correct. I'll defer to @oli-obk and @eddyb on what to do with this pull request. My intent was mainly to bring attention to the problem.

[eddyb]

My point was that the entire match args[1].value { can become a single loop using place_field.

[kennytm]

Hi @dwrensha, checking again. Would you like to apply the suggested change in #47239 (comment), i.e. to refactor “the entire match into a single loop using place_field”? Or should we just close it since the intention was “mainly to bring attention to the problem”?

Also cc @oli-obk, as OP mentioned in #47239 (comment), it seems difficult to move this PR forward without miri testing re-enabled, which in turn is blocked by #46882.

[pietroalbini]

@dwrensha and @oli-obk, ping from triage!

[oli-obk]

Let's close this until I get the miri submodule working again. Until then it's not testable and additionally this PR is not the preferred way to solve this

[pietroalbini]

Ok, closing this. @dwrensha, thank you for this PR anyway!