Implement `for<>` lifetime binder for closures
This PR implements RFC 3216 ([TI](https://github.com/rust-lang/rust/issues/97362)) and allows code like the following:
```rust
let _f = for<'a, 'b> |a: &'a A, b: &'b B| -> &'b C { b.c(a) };
// ^^^^^^^^^^^--- new!
```
cc ``@Aaron1011`` ``@cjgillot``
Pull Derefer before ElaborateDrops
_Follow up work to #97025#96549#96116#95887 #95649_
This moves `Derefer` before `ElaborateDrops` and creates a new `Rvalue` called `VirtualRef` that allows us to bypass many constraints for `DerefTemp`.
r? `@oli-obk`
Lower let-else in MIR
This MR will switch to lower let-else statements in MIR building instead.
To lower let-else in MIR, we build a mini-switch two branches. One branch leads to the matching case, and the other leads to the `else` block. This arrangement will allow temporary lifetime analysis running as-is so that the temporaries are properly extended according to the same rule applied to regular `let` statements.
cc https://github.com/rust-lang/rust/issues/87335Fix#98672
don't allow ZST in ScalarInt
There are several indications that we should not ZST as a ScalarInt:
- We had two ways to have ZST valtrees, either an empty `Branch` or a `Leaf` with a ZST in it.
`ValTree::zst()` used the former, but the latter could possibly arise as well.
- Likewise, the interpreter had `Immediate::Uninit` and `Immediate::Scalar(Scalar::ZST)`.
- LLVM codegen already had to special-case ZST ScalarInt.
So I propose we stop using ScalarInt to represent ZST (which are clearly not integers). Instead, we can add new ZST variants to those types that did not have other variants which could be used for this purpose.
Based on https://github.com/rust-lang/rust/pull/98831. Only the commits starting from "don't allow ZST in ScalarInt" are new.
r? `@oli-obk`
There are several indications that we should not ZST as a ScalarInt:
- We had two ways to have ZST valtrees, either an empty `Branch` or a `Leaf` with a ZST in it.
`ValTree::zst()` used the former, but the latter could possibly arise as well.
- Likewise, the interpreter had `Immediate::Uninit` and `Immediate::Scalar(Scalar::ZST)`.
- LLVM codegen already had to special-case ZST ScalarInt.
So instead add new ZST variants to those types that did not have other variants
which could be used for this purpose.
Clarify MIR semantics of storage statements
Seems worthwhile to start closing out some of the less controversial open questions about MIR semantics. Hopefully this is fairly non-controversial - it's what we implement already, and I see no reason to do anything more restrictive. cc ``@tmiasko`` who commented on this when it was discussed in the original PR that added these docs.
Miscellaneous inlining improvements
Add `#[inline]` to a few trivial non-generic methods from a perf report
that otherwise wouldn't be candidates for inlining.
don't succeed `evaluate_obligation` query if new opaque types were registered
fixes#98608fixes#98604
The root cause of all this is that in type flag computation we entirely ignore nongeneric things like struct fields and the signature of function items. So if a flag had to be set for a struct if it is set for a field, that will only happen if the field is generic, as only the generic parameters are checked.
I now believe we cannot use type flags to handle opaque types. They seem like the wrong tool for this.
Instead, this PR replaces the previous logic by adding a new variant of `EvaluatedToOk`: `EvaluatedToOkModuloOpaqueTypes`, which says that there were some opaque types that got hidden types bound, but that binding may not have been legal (because we don't know if the opaque type was in its defining scope or not).
Highlight conflicting param-env candidates
This could probably be further improved by noting _why_ equivalent param-env candidates (modulo regions) leads to ambiguity.
Fixes#98786
Rollup of 9 pull requests
Successful merges:
- #97917 (Implement ExitCodeExt for Windows)
- #98844 (Reword comments and rename HIR visiting methods.)
- #98979 (interpret: use AllocRange in UninitByteAccess)
- #98986 (Fix missing word in comment)
- #98994 (replace process exit with more detailed exit in src/bootstrap/*.rs)
- #98995 (Add a test for #80471)
- #99002 (suggest adding a derive for #[default] applied to variants)
- #99004 (Add a test for #70408)
- #99017 (Replace boolean argument for print_where_clause with an enum to make code more clear)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
interpret: use AllocRange in UninitByteAccess
also use nice new format string syntax in `interpret/error.rs`, and use the `#` flag to add `0x` prefixes where applicable.
r? ``@oli-obk``
Make lowering a query
Split from https://github.com/rust-lang/rust/pull/88186.
This PR refactors the relationship between lowering and the resolver outputs in order to make lowering itself a query.
In a first part, lowering is changed to avoid modifying resolver outputs, by maintaining its own data structures for creating new `NodeId`s and so.
Then, the `TyCtxt` is modified to allow creating new `LocalDefId`s from inside it. This is done by:
- enclosing `Definitions` in a lock, so as to allow modification;
- creating a query `register_def` whose purpose is to declare a `LocalDefId` to the query system.
See `TyCtxt::create_def` and `TyCtxt::iter_local_def_id` for more detailed explanations of the design.
This makes it possible to mutably borrow different fields of the MIR
body without resorting to methods like `basic_blocks_local_decls_mut_and_var_debug_info`.
To preserve validity of control flow graph caches in the presence of
modifications, a new struct `BasicBlocks` wraps together basic blocks
and control flow graph caches.
The `BasicBlocks` dereferences to `IndexVec<BasicBlock, BasicBlockData>`.
On the other hand a mutable access requires explicit `as_mut()` call.
Do not fetch HIR to compute variances.
Everything can be done using higher-level queries. This simplifies the code, and should allow better incremental caching.
macros: `LintDiagnostic` derive
- Move `LintDiagnosticBuilder` into `rustc_errors` so that a diagnostic derive can refer to it.
- Introduce a `DecorateLint` trait, which is equivalent to `SessionDiagnostic` or `AddToDiagnostic` but for lints. Necessary without making more changes to the lint infrastructure as `DecorateLint` takes a `LintDiagnosticBuilder` and re-uses all of the existing logic for determining what type of diagnostic a lint should be emitted as (e.g. error/warning).
- Various refactorings of the diagnostic derive machinery (extracting `build_field_mapping` helper and moving `sess` field out of the `DiagnosticDeriveBuilder`).
- Introduce a `LintDiagnostic` derive macro that works almost exactly like the `SessionDiagnostic` derive macro except that it derives a `DecorateLint` implementation instead. A new derive is necessary for this because `SessionDiagnostic` is intended for when the generated code creates the diagnostic. `AddToDiagnostic` could have been used but it would have required more changes to the lint machinery.
~~At time of opening this pull request, ignore all of the commits from #98624, it's just the last few commits that are new.~~
r? `@oli-obk`
Fix repr(align) enum handling
`enum`, for better or worse, supports `repr(align)`. That has already caused a bug in https://github.com/rust-lang/rust/issues/92464, which was "fixed" in https://github.com/rust-lang/rust/pull/92932, but it turns out that that fix is wrong and caused https://github.com/rust-lang/rust/issues/96185.
So this reverts #92932 (which fixes#96185), and attempts another strategy for fixing #92464: special-case enums when doing a cast, re-using the code to load the discriminant rather than assuming that the enum has scalar layout. This works fine for the interpreter.
However, #92464 contained another testcase that was previously not in the test suite -- and after adding it, it ICEs again. This is not surprising; codegen needs the same patch that I did in the interpreter. Probably this has to happen [around here](d32ce37a17/compiler/rustc_codegen_ssa/src/mir/rvalue.rs (L276)). Unfortunately I don't know how to do that -- the interpreter can load a discriminant from an operand, but codegen can only do that from a place. `@oli-obk` `@eddyb` `@bjorn3` any idea?
lints: mostly translatable diagnostics
As lints are created slightly differently than other diagnostics, intended to try make them translatable first and then look into the applicability of diagnostic structs but ended up just making most of the diagnostics in the crate translatable (which will still be useful if I do make a lot of them structs later anyway).
r? ``@compiler-errors``
Change enum->int casts to not go through MIR casts.
follow-up to https://github.com/rust-lang/rust/pull/96814
this simplifies all backends and even gives LLVM more information about the return value of `Rvalue::Discriminant`, enabling optimizations in more cases.
Interpret: AllocRange Debug impl, and use it more consistently
The two commits are pretty independent but it did not seem worth having two PRs for them.
r? ``@oli-obk``
Add method to mutate MIR body without invalidating CFG caches.
In addition to adding this method, a handful of passes are updated to use it. There's still quite a few passes that could in principle make use of this as well, but do not at the moment because they use `VisitorMut` or `MirPatch`, which needs additional support for this.
The method name is slightly unwieldy, but I don't expect anyone to be writing it a lot, and at least it says what it does. If anyone has a suggestion for a better name though, would be happy to rename.
r? rust-lang/mir-opt
CTFE interning: don't walk allocations that don't need it
The interning of const allocations visits the mplace looking for references to intern. Walking big aggregates like big static arrays can be costly, so we only do it if the allocation we're interning contains references or interior mutability.
Walking ZSTs was avoided before, and this optimization is now applied to cases where there are no references/relocations either.
---
While initially looking at this in the context of #93215, I've been testing with smaller allocations than the 16GB one in that issue, and with different init/uninit patterns (esp. via padding).
In that example, by default, `eval_to_allocation_raw` is the heaviest query followed by `incr_comp_serialize_result_cache`. So I'll show numbers when incremental compilation is disabled, to focus on the const allocations themselves at 95% of the compilation time, at bigger array sizes on these minimal examples like `static ARRAY: [u64; LEN] = [0; LEN];`.
That is a close construction to parts of the `ctfe-stress-test-5` benchmark, which has const allocations in the megabytes, while most crates usually have way smaller ones. This PR will have the most impact in these situations, as the walk during the interning starts to dominate the runtime.
Unicode crates (some of which are present in our benchmarks) like `ucd`, `encoding_rs`, etc come to mind as having bigger than usual allocations as well, because of big tables of code points (in the hundreds of KB, so still an order of magnitude or 2 less than the stress test).
In a check build, for a single static array shown above, from 100 to 10^9 u64s (for lengths in powers of ten), the constant factors are lowered:
(log scales for easier comparisons)
(linear scale for absolute diff at higher Ns)
For one of the alternatives of that issue
```rust
const ROWS: usize = 100_000;
const COLS: usize = 10_000;
static TWODARRAY: [[u128; COLS]; ROWS] = [[0; COLS]; ROWS];
```
we can see a similar reduction of around 3x (from 38s to 12s or so).
For the same size, the slowest case IIRC is when there are uninitialized bytes e.g. via padding
```rust
const ROWS: usize = 100_000;
const COLS: usize = 10_000;
static TWODARRAY: [[(u64, u8); COLS]; ROWS] = [[(0, 0); COLS]; ROWS];
```
then interning/walking does not dominate anymore (but means there is likely still some interesting work left to do here).
Compile times in this case rise up quite a bit, and avoiding interning walks has less impact: around 23%, from 730s on master to 568s with this PR.
Fix FFI-unwind unsoundness with mixed panic mode
UB maybe introduced when an FFI exception happens in a `C-unwind` foreign function and it propagates through a crate compiled with `-C panic=unwind` into a crate compiled with `-C panic=abort` (#96926).
To prevent this unsoundness from happening, we will disallow a crate compiled with `-C panic=unwind` to be linked into `panic-abort` *if* it contains a call to `C-unwind` foreign function or function pointer. If no such call exists, then we continue to allow such mixed panic mode linking because it's sound (and stable). In fact we still need the ability to do mixed panic mode linking for std, because we only compile std once with `-C panic=unwind` and link it regardless panic strategy.
For libraries that wish to remain compile-once-and-linkable-to-both-panic-runtimes, a `ffi_unwind_calls` lint is added (gated under `c_unwind` feature gate) to flag any FFI unwind calls that will cause the linkable panic runtime be restricted.
In summary:
```rust
#![warn(ffi_unwind_calls)]
mod foo {
#[no_mangle]
pub extern "C-unwind" fn foo() {}
}
extern "C-unwind" {
fn foo();
}
fn main() {
// Call to Rust function is fine regardless ABI.
foo::foo();
// Call to foreign function, will cause the crate to be unlinkable to panic-abort if compiled with `-Cpanic=unwind`.
unsafe { foo(); }
//~^ WARNING call to foreign function with FFI-unwind ABI
let ptr: extern "C-unwind" fn() = foo::foo;
// Call to function pointer, will cause the crate to be unlinkable to panic-abort if compiled with `-Cpanic=unwind`.
ptr();
//~^ WARNING call to function pointer with FFI-unwind ABI
}
```
Fix#96926
`@rustbot` label: T-compiler F-c_unwind
Enable MIR inlining
Continuation of https://github.com/rust-lang/rust/pull/82280 by `@wesleywiser.`
#82280 has shown nice compile time wins could be obtained by enabling MIR inlining.
Most of the issues in https://github.com/rust-lang/rust/issues/81567 are now fixed,
except the interaction with polymorphization which is worked around specifically.
I believe we can proceed with enabling MIR inlining in the near future
(preferably just after beta branching, in case we discover new issues).
Steps before merging:
- [x] figure out the interaction with polymorphization;
- [x] figure out how miri should deal with extern types;
- [x] silence the extra arithmetic overflow warnings;
- [x] remove the codegen fulfilment ICE;
- [x] remove the type normalization ICEs while compiling nalgebra;
- [ ] tweak the inlining threshold.
cleanup mir visitor for `rustc::pass_by_value`
by changing `& $($mutability)?` to `$(& $mutability)?`
I also did some formatting changes because I started doing them for the visit methods I changed and then couldn't get myself to stop xx, I hope that's still fairly easy to review.
Accept `DiagnosticMessage` in `LintDiagnosticBuilder::build` so that
lints can be built with translatable diagnostic messages.
Signed-off-by: David Wood <david.wood@huawei.com>
move MIR syntax into a dedicated file and ping some people whenever it changes
Adding or changing MIR operations/statements/whatever should be under significant scrutiny wrt their wider impact, specified semantics, and so on. So let's start by putting all that into a dedicated file and pinging some people whenever that file changes.
This PR only moves definitions around, and then fiddles with imports until it all works again.
Only keep a single query for well-formed checking
There are currently 3 queries to perform wf checks on different item-likes. This complexity is not required.
This PR replaces the query by:
- one query per item;
- one query to invoke it for a whole module.
This allows to remove HIR `ParItemLikeVisitor`.
Check ADT field is well-formed before checking it is sized
Fixes#96810.
There is one diagnostics regression, in [`src/test/ui/generic-associated-types/bugs/issue-80626.stderr`](https://github.com/rust-lang/rust/pull/97780/files#diff-53795946378e78a0af23a10277c628ff79091c18090fdc385801ee70c1ba6963). I am not super concerned about it, since it's GAT related.
We _could_ fix it, possibly by using the `FieldSized` obligation cause code instead of `BuiltinDerivedObligation`. But that would require changing `Sized` trait confirmation and the `adt_sized_constraint` query.
Reverse folder hierarchy
#91318 introduced a trait for infallible folders distinct from the fallible version. For some reason (completely unfathomable to me now that I look at it with fresh eyes), the infallible trait was a supertrait of the fallible one: that is, all fallible folders were required to also be infallible. Moreover the `Error` associated type was defined on the infallible trait! It's so absurd that it has me questioning whether I was entirely sane.
This trait reverses the hierarchy, so that the fallible trait is a supertrait of the infallible one: all infallible folders are required to also be fallible (which is a trivial blanket implementation). This of course makes much more sense! It also enables the `Error` associated type to sit on the fallible trait, where it sensibly belongs.
There is one downside however: folders expose a `tcx` accessor method. Since the blanket fallible implementation for infallible folders only has access to a generic `F: TypeFolder`, we need that trait to expose such an accessor to which we can delegate. Alternatively it's possible to extract that accessor into a separate `HasTcx` trait (or similar) that would then be a supertrait of both the fallible and infallible folder traits: this would ensure that there's only one unambiguous `tcx` method, at the cost of a little additional boilerplate. If desired, I can submit that as a separate PR.
r? ````@jackh726````
Improve suggestion for calling fn-like expr on type mismatch
1.) Suggest calling values of with RPIT types (and probably TAIT) when we expect `Ty` and have `impl Fn() -> Ty`
2.) Suggest calling closures even when they're not assigned to a local variable first
3.) Drive-by fix of a pretty-printing bug (`impl Fn()-> Ty` => `impl Fn() -> Ty`)
r? ```@estebank```
Remove (transitive) reliance on sorting by DefId in pretty-printer
This moves us a step closer to removing the `PartialOrd/`Ord` impls
for `DefId`. See #90317
implement `iter_projections` function on `PlaceRef`
this makes the api more flexible. the original function now calls the PlaceRef
version to avoid duplicating the code.
Point at return expression for RPIT-related error
Certainly this needs some diagnostic refining, but I wanted to show that it was possible first and foremost. Not sure if this is the right approach. Open to feedback.
Fixes#80583
#91318 introduced a trait for infallible folders distinct from the fallible version. For some reason (completely unfathomable to me now that I look at it with fresh eyes), the infallible trait was a supertrait of the fallible one: that is, all fallible folders were required to also be infallible. Moreover the `Error` associated type was defined on the infallible trait! It's so absurd that it has me questioning whether I was entirely sane.
This trait reverses the hierarchy, so that the fallible trait is a supertrait of the infallible one: all infallible folders are required to also be fallible (which is a trivial blanket implementation). This of course makes much more sense! It also enables the `Error` associated type to sit on the fallible trait, where it sensibly belongs.
There is one downside however: folders expose a `tcx` accessor method. Since the blanket fallible implementation for infallible folders only has access to a generic `F: TypeFolder`, we need that trait to expose such an accessor to which we can delegate. Alternatively it's possible to extract that accessor into a separate `HasTcx` trait (or similar) that would then be a supertrait of both the fallible and infallible folder traits: this would ensure that there's only one unambiguous `tcx` method, at the cost of a little additional boilerplate. If desired, I can submit that as a separate PR.
r? @jackh726
`try_fold_unevaluated` for infallible folders
#97447 added folding of unevaluated constants, but did not include an override of the default (fallible) operation in the blanket impl of `FallibleTypeFolder` for infallible folders. Here we provide that missing override.
r? ```@nnethercote```
Remove dereferencing of Box from codegen
Through #94043, #94414, #94873, and #95328, I've been fixing issues caused by Box being treated like a pointer when it is not a pointer. However, these PRs just introduced special cases for Box. This PR removes those special cases and instead transforms a deref of Box into a deref of the pointer it contains.
Hopefully, this is the end of the Box<T, A> ICEs.
Don't omit comma when suggesting wildcard arm after macro expr
* Also adds `Span::eq_ctxt` to consolidate the various usages of `span.ctxt() == other.ctxt()`
* Also fixes an unhygenic usage of spans which caused the suggestion to render weirdly when we had one arm match in a macro
* Also always suggests a comma (i.e. even after a block) if we're rendering a wildcard arm in a single-line match (looks prettier 🌹)
Fixes#94866
#97447 added folding of unevaluated constants, but did not include an override of the default (fallible) operation in the blanket impl of `FallibleTypeFolder` for infallible folders. Here we provide that missing override.
r? @nnethercote
Rollup of 5 pull requests
Successful merges:
- #98105 (rustdoc: remove tuple link on round braces)
- #98136 (Rename `impl_constness` to `constness`)
- #98146 (Remove --memory-init-file flag when linking with Emscripten)
- #98219 (Skip late bound regions in GATSubstCollector)
- #98233 (Remove accidental uses of `&A: Allocator`)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Rename `impl_constness` to `constness`
The current code is a basis for `is_const_fn_raw`, and `impl_constness`
is no longer a valid name, which is previously used for determining the
constness of impls, and not items in general.
r? `@oli-obk`
ctfe: limit hashing of big const allocations when interning
Const allocations are only hashed for interning. However, they can be large, making the hashing expensive especially since it uses `FxHash`: it's better suited to short keys, not potentially big buffers like the actual bytes of allocation and the associated 1/8th sized `InitMask`.
We can partially hash these fields when they're large, hashing the length, and head and tail of these buffers, to
limit possible collisions while avoiding most of the hashing work.
r? `@ghost`
Big const allocations hash a large amount of data for interning:
the whole bytes buffer, and the 1/8th sized initmask, with FxHash.
This hash function is made for shorter keys.
This only hashes the length, and head and tail of these buffers, to
limit possible collisions while avoiding most of the hashing work.
Support lint expectations for `--force-warn` lints (RFC 2383)
Rustc has a `--force-warn` flag, which overrides lint level attributes and forces the diagnostics to always be warn. This means, that for lint expectations, the diagnostic can't be suppressed as usual. This also means that the expectation would not be fulfilled, even if a lint had been triggered in the expected scope.
This PR now also tracks the expectation ID in the `ForceWarn` level. I've also made some minor adjustments, to possibly catch more bugs and make the whole implementation more robust.
This will probably conflict with https://github.com/rust-lang/rust/pull/97718. That PR should ideally be reviewed and merged first. The conflict itself will be trivial to fix.
---
r? `@wesleywiser`
cc: `@flip1995` since you've helped with the initial review and also discussed this topic with me. 🙃
Follow-up of: https://github.com/rust-lang/rust/pull/87835
Issue: https://github.com/rust-lang/rust/issues/85549
Yeah, and that's it.
The current code is a basis for `is_const_fn_raw`, and `impl_constness`
is no longer a valid name, which is previously used for determining the
constness of impls, and not items in general.
Make `ExprKind::Closure` a struct variant.
Simple refactor since we both need it to introduce additional fields in `ExprKind::Closure`.
r? ``@Aaron1011``
Rename rustc_serialize::opaque::Encoder as MemEncoder.
This avoids the name clash with `rustc_serialize::Encoder` (a trait),
and allows lots qualifiers to be removed and imports to be simplified
(e.g. fewer `as` imports).
(This was previously merged as commit 5 in #94732 and then was reverted
in #97905 because of a perf regression caused by commit 4 in #94732.)
r? ```@bjorn3```
This avoids the name clash with `rustc_serialize::Encoder` (a trait),
and allows lots qualifiers to be removed and imports to be simplified
(e.g. fewer `as` imports).
(This was previously merged as commit 5 in #94732 and then was reverted
in #97905 because of a perf regression caused by commit 4 in #94732.)
And likewise for the `Const::val` method.
Because its type is called `ConstKind`. Also `val` is a confusing name
because `ConstKind` is an enum with seven variants, one of which is
called `Value`. Also, this gives consistency with `TyS` and `PredicateS`
which have `kind` fields.
The commit also renames a few `Const` variables from `val` to `c`, to
avoid confusion with the `ConstKind::Value` variant.
Remove unnecessary `to_string` and `String::new`
73fa217bc1 changed the type of the `suggestion` argument to `impl ToString`. This patch removes unnecessary `to_string` and `String::new`.
cc: `````@davidtwco`````
Handle `def_ident_span` like `def_span`.
`def_ident_span` had an ad-hoc status in the compiler.
This PR refactors it to be a first-class citizen like `def_span`:
- it gets encoded in the main metadata loop, instead of the visitor;
- its implementation is updated to mirror the one of `def_span`.
We do not remove the `Option` in the return type, since some items do not have an ident, AnonConsts for instance.
Wrap `HirId`s of locals into `LocalVarId`s for THIR nodes
This is the first effort to decouple `HirId`s from THIR. `HirId` is not very relevant in building THIR and MIR.
Based on the changeset, I think there are a few other pending refactoring that we could perform after this, in case we want to eliminate use of `HirId` in THIR.
- `TypeckResults::closure_min_captures` could be remapped from the variable `HirId`s to `LocalVarId` while the THIR is getting built.
- Use of `ScopeTree::var_scope` could be eliminated as well, since we will consider deprecating `ScopeTree` in the future.
Revert part of #94372 to improve performance
#94732 was supposed to give small but widespread performance improvements, as judged from three per-merge performance runs. But the performance run that occurred after merging included a roughly equal number of improvements and regressions, for unclear reasons.
This PR is for a test run reverting those changes, to see what happens.
r? `@ghost`
use precise spans for recursive const evaluation
This fixes https://github.com/rust-lang/rust/issues/73283 by using a `TyCtxtAt` with a more precise span when the interpreter recursively calls itself. Hopefully such calls are sufficiently rare that this does not cost us too much performance.
(In theory, cycles can also arise through layout computation, as layout can depend on consts -- but layout computation happens all the time so we'd have to do something to not make this terrible for performance.)
it feels arbitrary to have `Ty` and `Const` directly
in that module and to not have `GenericArg` and
`GenericArgKind` there. Writing `ty::GenericArg`
can also feel clearer than importing it. Using
`ty::subst::GenericArg` however is ugly.
