Linting shouldn't be run on optimized MIR

[ecstatic-morse]

For example, const-checking needs to run before optimizations, otherwise it might miss some invalid operations that get optimized away. Specifically, it should run on the result of the mir_const query, but currently it uses optimized_mir. This is because the result of the mir_const query is "stolen" instead of cloned, so it no longer exists when clippy runs. clippy tries to work around this by setting mir-opt-level=0, but this doesn't disable every optimization pass.

Until this is fixed, missing_const_for_fn will have false positives and/or ICE (due to some assertions about the state of the MIR at the point const-checking is run).

[camsteffen]

According to @oli-obk on zulip, we can now use mir_for_ctfe.

[camsteffen]

I tried to switch over to mir_for_ctfe but was harshly denied, "Turn aside, heathen! None shall enter except he be const!". Chatted again with @oli-obk and he suggested we should add tcx.unoptimized_mir to simply bypasses the assertion.

[b-NC]

I tried to switch over to mir_for_ctfe but was harshly denied, "Turn aside, heathen! None shall enter except he be const!". Chatted again with @oli-obk and he suggested we should add tcx.unoptimized_mir to simply bypasses the assertion.

@camsteffen is this still something to be fixed ?

[camsteffen]

@InfRandomness yes I believe so

[1c3t3a]

Checking in once more: Would it be possible to add tcx.unoptimized_mir? I hit this issue again in #14003 and I think this would be a clean solution? Don't know about the performance cost, though.