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.
Rollup of 10 pull requests
Successful merges:
- #64273 (Stabilize attribute macros on inline modules)
- #67287 (typeck: note other end-point when checking range pats)
- #67564 (docs: Iterator adapters have unspecified results after a panic)
- #67622 (Some keyword documentation.)
- #67657 (Clean up const-hack PRs now that const if / match exist.)
- #67677 (resolve: Minor cleanup of duplicate macro reexports)
- #67687 (Do not ICE on lifetime error involving closures)
- #67698 (Move reachable_set and diagnostic_items to librustc_passes.)
- #67701 (tidy: Enforce formatting rather than just check it if `--bless` is specified)
- #67715 (Typo fix)
Failed merges:
r? @ghost
Work around a resolve bug in const prop
r? @wesleywiser @anp
This isn't exposed right now, but further changes to rustc may start causing bugs without this.
We only want to return specializations when `Reveal::All` is passed, not
when `Reveal::UserFacing` is. Resolving this fixes several issues with
the `ConstProp`, `SimplifyBranches`, and `Inline` MIR optimization
passes.
Fixes#66901
Various const eval and pattern matching ICE fixes
r? @RalfJung
cc @spastorino
This PR does not change existing behaviour anymore and just fixes a bunch of ICEs reachable from user code (sometimes even on stable via obscure union transmutes).
The callbacks have precisely two states: the default, and the one
present throughout almost all of the rustc run (the filled in value
which has access to TyCtxt).
We used to store this as a thread local, and reset it on each thread to
the non-default value. But this is somewhat wasteful, since there is no
reason to set it globally -- while the callbacks themselves access TLS,
they do not do so in a manner that fails in when we do not have TLS to
work with.
Initial implementation of `#![feature(bindings_after_at)]`
Following up on #16053, under the gate `#![feature(bindings_after_at)]`, `x @ Some(y)` is allowed subject to restrictions necessary for soundness.
The implementation and test suite should be fairly complete now.
One aspect that is not covered is the interaction with nested `#![feature(or_patterns)]`.
This is not possible to test at the moment in a good way because that feature has not progressed sufficiently and has fatal errors in MIR building. We should make sure to add such tests before we stabilize both features (but shipping one of them is fine).
r? @pnkfelix
cc @nikomatsakis @matthewjasper @pcwalton
cc https://github.com/rust-lang/rust/issues/65490
Add simpler entry points to const eval for common usages.
I found the `tcx.const_eval` API to be complex/awkward to work with, because of the inherent complexity from all of the different situations it is called from. Though it mainly used in one of the following ways:
- Evaluates the value of a constant without any substitutions, e.g. evaluating a static, discriminant, etc.
- Evaluates the value of a resolved instance of a constant. this happens when evaluating unevaluated constants or normalising trait constants.
- Evaluates a promoted constant.
This PR adds three new functions `const_eval_mono`, `const_eval_resolve`, and `const_eval_promoted` to `TyCtxt`, which each cater to one of the three ways `tcx.const_eval`
is normally used.
Allocate HIR on an arena 1/4
This PR is the first in a series of 4, aiming at allocating the HIR on an arena, as a memory optimisation.
1. This first PR lays the groundwork and migrates some low-hanging fruits.
2. The second PR will migrate `hir::Expr`, `hir::Pat` and related.
3. The third PR will migrate `hir::Ty` and related.
4. The final PR will be dedicated to eventual cleanups.
In order to make the transition as gradual as possible, some lowering routines receive `Box`-allocated data and move it into the arena. This is a bit wasteful, but hopefully temporary.
Nonetheless, special care should be taken to avoid double arena allocations.
Work mentored by @Zoxc.
Make GATs less ICE-prone.
After this PR simple lifetime-generic associated types can now be used in a compiling program. There are two big limitations:
* #30472 has not been addressed in any way (see src/test/ui/generic-associated-types/iterable.rs)
* Using type- and const-generic associated types errors because bound types and constants aren't handled by trait solving.
* The errors are technically non-fatal, but they happen in a [part of the compiler](4abb0ad273/src/librustc_typeck/lib.rs (L298)) that fairly aggressively stops compiling on errors.
closes#47206closes#49362closes#62521closes#63300closes#64755closes#67089
Add a raw "address of" operator
* Parse and feature gate `&raw [const | mut] expr` (feature gate name is `raw_address_of`)
* Add `mir::Rvalue::AddressOf`
* Use the new `Rvalue` for:
* the new syntax
* reference to pointer casts
* drop shims for slices and arrays
* Stop using `mir::Rvalue::Cast` with a reference as the operand
* Correctly evaluate `mir::Rvalue::{Ref, AddressOf}` in constant propagation
cc @Centril @RalfJung @oli-obk @eddyb
cc #64490
