fix(frontend): make path_to breadth-first

[W95Psp]

Fixes #692

Status: seems like this is introducing a bug, I have to debug.
Some implementations cannot be found after the change

[maximebuyse]

A first fix to this implementation (or to a bug revealed by it) is in #797

There is another problem remaining (reaching https://github.com/hacspec/hax/blob/main/frontend/exporter/src/traits.rs#L349)

[maximebuyse]

A first fix to this implementation (or to a bug revealed by it) is in #797

There is another problem remaining (reaching https://github.com/hacspec/hax/blob/main/frontend/exporter/src/traits.rs#L349)

This bug introduced by the initial version of this PR is now fixed. It was just that switching from a recursive depth-first search to a breadth-first search with a queue meant the parents and associated traits had to be taken from the candidate path and not self. In other words we were only checking paths of length 1 with path_to.

[W95Psp]

Nice, thank you @maximebuyse! That looks good to me, and @franziskuskiefer is currently trying to break it by throwing it on bigger crates 😅 🤞

[franziskuskiefer]

The frontend panics now in most cases where it looped before.

thread 'rustc' panicked at /rustc/6b0f4b5ec3aa707ecaa78230722117324a4ce23c/compiler/rustc_type_ir/src/binder.rs:713:9:
type parameter `T/#1` (T/#1/1) out of range when instantiating, args=[Alias(Projection, AliasTy { args: [Self/#0], def_id: DefId(5:71 ~ crypto_common[e66d]::BlockSizeUser::BlockSize) })]

[W95Psp]

This is #683, which is not related. IMO we should merge that PR, that's fixing the looping issue

[franziskuskiefer]

that's fixing the looping issue

How do we know that it is fixing the issue?

[maximebuyse]

that's fixing the looping issue

How do we know that it is fixing the issue?

You're right, it doesn't work on #692 . The looping behavior was gone because the breadth-first search had a bug.

[W95Psp]

Oh, you mean that it's still looping? :/

[W95Psp]

I pushed make-instantiating-bug-nonfatal which is making the instantiating bug non-fatal. The looping behavior is gone on https://github.com/RustCrypto/traits/tree/master/cipher at least.

[W95Psp]

LGTM, with a few nits!

[maximebuyse]

LGTM, with a few nits!

Thanks for the tips, I made the corrections.

[W95Psp]

Nice, thanks! Have you tried the patch on both #686 and #692? (I updated make-instantiating-bug-nonfatal)
If #692 goes through with make-instantiating-bug-nonfatal, let's document that and close it in favor of #747.

Also, you can resolve conversations on the comments I added, otherwise the PR will be blocked from merging :)

[maximebuyse]

Nice, thanks! Have you tried the patch on both #686 and #692? (I updated make-instantiating-bug-nonfatal) If #692 goes through with make-instantiating-bug-nonfatal, let's document that and close it in favor of #747.

Also, you can resolve conversations on the comments I added, otherwise the PR will be blocked from merging :)

This seems to solve #692, I added a test with this example.
But for #686, no change (the memory consumption keeps increasing until I have to kill it). So it might be another problem.
I didn't try with make-instantiating-bug-non-fatal but I didn't get instantiating issues apparently.

[franziskuskiefer]

Alright, then let's get this in and keep looking for the issue in #686.

[W95Psp]

Mmh, ok. Added a comment there: #686 (comment).