Not sure if this helps but in theory it should be less work than what
the current micro optimization does for `ty::Predicate` lists.
(It would explain the overhead I am seeing from `perf`.)
Fix ICE in const pretty printing and resolve FIXME
Consts now have a `fmt::Display` impl, so we can just use that to pretty-print.
This resolves an ICE in https://github.com/rust-lang/rust/issues/61936, though it hits more ICEs afterwards. I couldn't find a test case that was resolved by this that didn't hit errors later on.
Silence `TooGeneric` error
This error may be produced during intermediate failed attempts at evaluation of a generic const, which may nevertheless succeed later.
Fixes https://github.com/rust-lang/rust/issues/66962.
r? @eddyb
Fix ICE involving calling `Instance.ty` during const evaluation
Fixes#67639
`Instance.ty` assumes that we are in a fully monomorphic context (e.g.
codegen), and can therefore use an empty `ParamEnv` when performing
normalization. Howver, the MIR constant evaluator code ends up calling
`Instance.ty` as a result of us attemptign to 'speculatively'
const-evaluate generic functions during const propagation.
As a result,
we may end up with projections involving type parameters
(e.g. <T as MyTrait>::Bar>) in the type we are trying to normalize.
Normalization expects us to have proper predicates in the `ParamEnv` for
such projections, and will ICE if we don't.
This commit adds a new method `Instance.ty_env`, which takes a
`ParamEnv` for use during normalization. The MIR const-evaluator code is
changed to use this method, passing in the proper `ParamEnv` for the
context at hand.
Fixes#67639
`Instance.ty` assumes that we are in a fully monomorphic context (e.g.
codegen), and can therefore use an empty `ParamEnv` when performing
normalization. Howver, the MIR constant evaluator code ends up calling
`Instance.ty` as a result of us attemptign to 'speculatively'
const-evaluate generic functions during const propagation.
As a result,
we may end up with projections involving type parameters
(e.g. <T as MyTrait>::Bar>) in the type we are trying to normalize.
Normalization expects us to have proper predicates in the `ParamEnv` for
such projections, and will ICE if we don't.
This commit adds a new method `Instance.ty_env`, which takes a
`ParamEnv` for use during normalization. The MIR const-evaluator code is
changed to use this method, passing in the proper `ParamEnv` for the
context at hand.
Use self profile infrastructure for -Z time and -Z time-passes
There's no longer indentation for -Z time and -Z time-passes and duplicate timers between self profiling and -Z time-passes have been removed.
r? @wesleywiser
perf: Don't recurse into types that do not need normalizing
A bit speculative at this stage but profiling shows that type folding
takes up a substantial amount of time during normalization which may
indicate that many types may be folded despite there being nothing to
normalize
