Get with field index from pattern slice instead of directly indexing

[csmoe]

Closes #82772
r? @estebank

#82789 (comment)

@estebank So the real cause is we only generate single pattern for Box here
https://github.com/csmoe/rust/blob/615b03aeaa8ce9819de7828740ab3cd7def4fa76/compiler/rustc_mir_build/src/thir/pattern/deconstruct_pat.rs#L1130-L1132
But in the replacing function, it tries to index on the 1-length pattern slice with field 1, thus out of bounds.
https://github.com/csmoe/rust/blob/615b03aeaa8ce9819de7828740ab3cd7def4fa76/compiler/rustc_mir_build/src/thir/pattern/deconstruct_pat.rs#L1346

[csmoe]

@estebank ping

[estebank]

Could you add a test for the more than one field case?

[csmoe]

@estebank kind of strange as I fixed that with vis.is_accessible_from(), then in the test --bless shows that we already have E0451 in rustc_privacy addressed such private field case. But don't know why rustc didn't catch this earlier(before typeck of pattern).

[estebank]

I have a theory: we already emit privacy errors pointing at the ADT on some cases of private fields, so the error for the field themselves would only be redundant. The ICE you're fixing only happens on non-existing fields, and the new error is just precluding getting to that. Could you verify whether my theory is correct? I'm swamped until next week to do this kind of light digging :)

[csmoe]

@estebank So the real cause is we only generate single pattern for Box here
https://github.com/csmoe/rust/blob/615b03aeaa8ce9819de7828740ab3cd7def4fa76/compiler/rustc_mir_build/src/thir/pattern/deconstruct_pat.rs#L1130-L1132
But in the replacing function, it tries to index on the 1-length pattern slice with field 1, thus out of bounds.
https://github.com/csmoe/rust/blob/615b03aeaa8ce9819de7828740ab3cd7def4fa76/compiler/rustc_mir_build/src/thir/pattern/deconstruct_pat.rs#L1346

[estebank]

@bors r+ rollup

[bors]

📌 Commit 77fb6a0 has been approved by estebank