Use existing framework for backward dataflow analyses
This PR adds support for backward analyses to the dataflow framework and adds a new live variable analysis (based on the existing one in `librustc_mir/util/liveness.rs`). By adding these to the framework instead of having a separate API, all newly implemented backward dataflow analyses get cursors/visitors, `rustc_peek` tests, and graphviz visualizations for free. In the near-term, this makes it much easier to implement global dead-store elimination, and I believe that this will enable even more MIR optimizations in the future.
This PR makes many changes to the dataflow API, since some concepts and terminology only make sense in forward dataflow. Below is a list of the important changes.
- ~~`entry_set` -> `fixpoint` (the fixpoint for backward dataflow problems is after the block's terminator)~~
- `seek_{before,after}` -> `seek_{before,after}_primary_effect` (the unprefixed dataflow effect is now referred to as the "primary" effect instead of the "after" effect. The "before" effect remains the same, although I considered changing it to the "antecedent" effect. In both backward and forward dataflow, the "before" effect is applied prior to the "primary" effect. I feel very strongly that this is the correct choice, as it means consumers don't have to switch between `seek_before` and `seek_after` based on the direction of their analysis.
- `seek_after_assume_call_returns` is now gone. Users can use `ResultsCursor::apply_custom_effect` to emulate it.
- `visit_{statement,terminator}_exit` -> `visit_{statement,terminator}_after_primary_effect`
- `visit_{statement,terminator}` -> `visit_{statement,terminator}_before_primary_effect`
Implementing this also required refactoring the dataflow cursor implementation so it could work in both directions. This is a large percentage of the diff, since the cursor code is rather complex. The fact that the cursor is exhaustively tested in both directions should reassure whomever is unlucky enough to review this 🤣.
In order to avoid computing the reverse CFG for forward dataflow analyses, I've added some hacks to the existing `mir::BodyAndCache` interface. I've requested changes to this interface that would let me implement this more efficiently.
r? @eddyb (feel free to reassign)
cc @rust-lang/wg-mir-opt
Suggest deref when coercing `ty::Ref` to `ty::RawPtr` with arbitrary mutability
Fixes#71676
1. Implement dereference suggestion when coercing `ty::Ref` to `ty::RawPtr` with arbitrary mutability.
2. Extract the dereference steps into `deref_steps()`, which removes all the `use` and `pub` noise introduced by last PR #71540, and makes the code more readable.
3. Use the `remove_prefix()` closure which makes the prefix removal more readable.
4. Introduce `Applicability` as a return value of `check_ref` to suggest `Applicability::Unspecified` suggestion.
**Special**: I found it is not possible to genereate `Applicability::MachineApplicable` suggestion for situation like this:
```rust
use std::ops::Deref;
use std::ops::DerefMut;
struct Bar(u8);
struct Foo(Bar);
struct Emm(Foo);
impl Deref for Bar{
type Target = u8;
fn deref(&self) -> &Self::Target {
&self.0
}
}
impl Deref for Foo {
type Target = Bar;
fn deref(&self) -> &Self::Target {
&self.0
}
}
impl Deref for Emm {
type Target = Foo;
fn deref(&self) -> &Self::Target {
&self.0
}
}
impl DerefMut for Bar{
fn deref_mut(&mut self) -> &mut Self::Target {
&mut self.0
}
}
impl DerefMut for Foo {
fn deref_mut(&mut self) -> &mut Self::Target {
&mut self.0
}
}
impl DerefMut for Emm {
fn deref_mut(&mut self) -> &mut Self::Target {
&mut self.0
}
}
fn main() {
let a = Emm(Foo(Bar(0)));
let _: *mut u8 = &a; //~ ERROR mismatched types
}
```
We may suggest `&mut ***a` here, but the `a` is not declared as mutable variable. And also when processing HIR, it's not possible to check if `a` is declared as a mutable variable (currently we do borrow checking with MIR). So we cannot ensure that suggestion when coercing immutable reference to mutable pointer is always machine applicable. Therefore I added a `Applicability` return value in `check_ref()`. And move the `immutable reference -> mutable pointer` situation into a sperate test file without `run-rustfix`. (It seems that `run-rustfix` will also adopt `Applicability::Unspecified` suggestion, which is strange)
Fix exceeding bitshifts not emitting for assoc. consts (properly this time, I swear!)
Fixes#69021 and fixes#71353.
As described in https://github.com/rust-lang/rust/issues/71353#issuecomment-617901923, this PR:
- adds a variant of `try_validation!` called `try_validation_pat!` that allows specific failures to be turned into validation failures (but returns the rest, unchanged), and
- allows `InvalidProgram` to be returned out of validation
r? @RalfJung
Add `RefCell::take`
Add `RefCell::take` to match `Cell` and `Option`.
I also changed a couple of calls to `.replace` to `.take`.
Tracking issue is #71395.
This is my first contribution, please tell me if there's anything I could improve, thanks!
typeck: always expose explicit enum discriminant `AnonConst`s' parent in `generics_of`.
This is similar to #70452 but for explicit `enum` discriminant constant expressions.
However, unlike #70452, this PR should have no effect on stable code, as while it alleviates #43408 errors, there is no way to actually compile an `enum` with generic parameters *and* explicit discriminants, without `#![feature(arbitrary_enum_discriminant)]`, as explicit discriminant expression don't count as uses of parameters (if they did, they would count as invariant uses).
<hr/>
There's also 2 other commits here, both related to #70453:
* "ty: use `delay_span_bug` in `ty::AdtDef::eval_explicit_discr`." - hides the ICEs demonstrated on #70453, when there are other errors (which the next commit adds)
* "typeck/wfcheck: require that explicit enum discriminants const-evaluate succesfully." - closes#70453 by picking alternative "2", i.e. erroring when a discriminant doesn't fully const-evaluate from the perspective of the `enum` definition
In the future, it might be possible to allow `enum` discriminants to actually depend on parameters, but that will likely require #68436 + some way to restrict the values so no two variants can end up with overlapping discriminants.
As this PR would close#70453, it shouldn't be merged until a decision is reached there.
r? @nikomatsakis
Decode qualifs for associated const defaults
Fixes#71734.
We encode qualifs for associated constants, but never expected to decode the qualifs for defaulted associated consts. Fix this, and test that associated const defaults have the correct qualifs cross-crate.
r? @tmandry
Implement `confusable_idents` lint.
This collects all identifier symbols into `ParseSession` and examines them within the non-ascii-idents lint.
The skeleton generation part needs to be added to `unicode-security` crate. Will update this PR when the crate is updated.
r? @petrochenkov
EDIT: also included the `concat_idents` part.
Rename `bitcode-in-rlib` option to `embed-bitcode`
This commit finishes work first pioneered in #70458 and started in #71528.
The `-C bitcode-in-rlib` option, which has not yet reached stable, is
renamed to `-C embed-bitcode` since that more accurately reflects what
it does now anyway. Various tests and such are updated along the way as
well.
This'll also need to be backported to the beta channel to ensure we
don't accidentally stabilize `-Cbitcode-in-rlib` as well.
This commit finishes work first pioneered in #70458 and started in #71528.
The `-C bitcode-in-rlib` option, which has not yet reached stable, is
renamed to `-C embed-bitcode` since that more accurately reflects what
it does now anyway. Various tests and such are updated along the way as
well.
This'll also need to be backported to the beta channel to ensure we
don't accidentally stabilize `-Cbitcode-in-rlib` as well.
Have the per-query caches store the results on arenas
This PR leverages the cache for each query to serve as storage area for the query results.
It introduces a new cache `ArenaCache`, which moves the result to an arena,
and only stores the reference in the hash map.
This allows to remove a sizeable part of the usage of the global `TyCtxt` arena.
I only migrated queries that already used arenas before.
Allow `Downcast` projections unconditionally in const-checking
`ProjectionElem::Downcast` sounds scary, but it's really just the projection we use to access a particular enum variant. They usually appear in the lowering of a `match` statement, so they have been associated with control flow in const-checking, but they don't do any control flow by themselves. We already have a HIR pass that looks for `if` and `match` (even ones that have 1 or fewer reachable branches). That pass is double-checked by a MIR pass that looks for `SwitchInt`s and `FakeRead`s for match scrutinees. In my opinion, there's no need to look for `Downcast` as well.
r? @oli-obk
MIR dump: print pointers consistently with Miri output
This makes MIR allocation dump pointer printing consistent with Miri output: both use hexadecimal offsets with a `0x` prefix. To save some space, MIR dump replaces the `alloc` prefix by `a` when necessary.
I also made AllocId/Pointer printing more consistent in their Debug/Display handling, and adjusted Display printing for Scalar a bit to avoid using decimal printing when we do not know the sign with which to interpret the value (IMO using decimal then is misleading).
Rename Unique::empty() -> Unique::dangling()
A `FIXME` comment in `src/libcore/ptr/unique.rs` suggested refactoring `Unique::empty()` to `Unique::dangling()` which this PR does.
extend NLL checker to understand `'empty` combined with universes
This PR extends the NLL region checker to understand `'empty` combined with universes. In particular, it means that the NLL region checker no longer considers `exists<R2> { forall<R1> { R1: R2 } }` to be provable. This is work towards https://github.com/rust-lang/rust/issues/59490, but we're not all the way there. One thing in particular it does not address is error messages.
The modifications to the NLL region inference code turned out to be simpler than expected. The main change is to require that if `R1: R2` then `universe(R1) <= universe(R2)`.
This constraint follows from the region lattice (shown below), because we assume then that `R2` is "at least" `empty(Universe(R2))`, and hence if `R1: R2` (i.e., `R1 >= R2` on the lattice) then `R1` must be in some universe that can name `'empty(Universe(R2))`, which requires that `Universe(R1) <= Universe(R2)`.
```
static ----------+-----...------+ (greatest)
| | |
early-bound and | |
free regions | |
| | |
scope regions | |
| | |
empty(root) placeholder(U1) |
| / |
| / placeholder(Un)
empty(U1) -- /
| /
... /
| /
empty(Un) -------- (smallest)
```
I also made what turned out to be a somewhat unrelated change to add a special region to represent `'empty(U0)`, which we use (somewhat hackily) to indicate well-formedness checks in some parts of the compiler. This fixes#68550.
I did some investigation into fixing the error message situation. That's a bit trickier: the existing "nice region error" code around placeholders relies on having better error tracing than NLL currently provides, so that it knows (e.g.) that the constraint arose from applying a trait impl and things like that. I feel like I was hoping *not* to do such fine-grained tracing in NLL, and it seems like we...largely...got away with that. I'm not sure yet if we'll have to add more tracing information or if there is some sort of alternative.
It's worth pointing out though that I've not kind of shifted my opinion on whose job it should be to enforce lifetimes: I tend to think we ought to be moving back towards *something like* the leak-check (just not the one we *had*). If we took that approach, it would actually resolve this aspect of the error message problem, because we would be resolving 'higher-ranked errors' in the trait solver itself, and hence we wouldn't have to thread as much causal information back to the region checker. I think it would also help us with removing the leak check while not breaking some of the existing crates out there.
Regardless, I think it's worth landing this change, because it was relatively simple and it aligns the set of programs that NLL accepts with those that are accepted by the main region checker, and hence should at least *help* us in migration (though I guess we still also have to resolve the existing crates that rely on leak check for coherence).
r? @matthewjasper
Miri: better document and fix dynamic const pattern soundness checks
https://github.com/rust-lang/const-eval/issues/42 got me thinking about soundness for consts being used in patterns, and I found a hole in our existing dynamic checks: a const referring to a mutable static *in a different crate* was not caught. This PR fixes that. It also adds some comments that explain which invariants are crucial for soundness of const-patterns.
Curiously, trying to weaponize this soundness hole failed: pattern matching compilation ICEd when encountering the cross-crate static, saying "expected allocation ID alloc0 to point to memory". I don't know why that would happen, statics *should* be entirely normal memory for pattern matching to access.
r? @oli-obk
Cc @rust-lang/wg-const-eval
Suggest deref when coercing `ty::Ref` to `ty::RawPtr`
Fixes#32122
Currently we do autoderef when casting `ty::Ref` ->`ty::Ref`, but we don't autoderef when casting `ty::Ref` -> `ty::RawPtr`. This PR make the compiler suggests deref when coercing `ty::Ref` to `ty::RawPtr`
rustc: fix check_attr() for methods, closures and foreign functions
This fixes an issue that previously turned up for methods in https://github.com/rust-lang/rust/pull/69274, but also exists for closures and foreign function: `check_attr` does not call `codegen_fn_attrs()` for these types when it should, meaning that incorrectly used function attributes are not diagnosed without codegen.
The issue affects our UI tests, as they run with `--emit=metadata` by default, but as it turns out, this is not the only case: Function attributes are not checked on any dead code without this fix!
This makes the fix a **breaking change**. The following very silly Rust programs compiles fine on stable Rust when it should not, which is fixed by this PR.
```rust
fn main() {
#[target_feature(enable = "sse2")]
|| {};
}
```
I assume any real-world program which may trigger this issue would at least emit a dead code warning, but of course that is no guarantee that such code does not exist...
Fixes#70307
Remove -Z no-landing-pads flag
Since #67502, `-Z no-landing-pads` will cause all attempted unwinds to abort since we don't generate a `try` / `catch`. This previously worked because `__rust_try` was located in libpanic_unwind which is always compiled with `-C panic=unwind`, but `__rust_try` is now directly inline into the crate that uses `catch_unwind`.
As such, `-Z no-landing-pads` is now mostly useless and people should use `-C panic=abort` instead.
Store LLVM bitcode in object files, not compressed
This commit is an attempted resurrection of #70458 where LLVM bitcode
emitted by rustc into rlibs is stored into object file sections rather
than in a separate file. The main rationale for doing this is that when
rustc emits bitcode it will no longer use a custom compression scheme
which makes it both easier to interoperate with existing tools and also
cuts down on compile time since this compression isn't happening.
The blocker for this in #70458 turned out to be that native linkers
didn't handle the new sections well, causing the sections to either
trigger bugs in the linker or actually end up in the final linked
artifact. This commit attempts to address these issues by ensuring that
native linkers ignore the new sections by inserting custom flags with
module-level inline assembly.
Note that this does not currently change the API of the compiler at all.
The pre-existing `-C bitcode-in-rlib` flag is co-opted to indicate
whether the bitcode should be present in the object file or not.
Finally, note that an important consequence of this commit, which is also
one of its primary purposes, is to enable rustc's `-Clto` bitcode
loading to load rlibs produced with `-Clinker-plugin-lto`. The goal here
is that when you're building with LTO Cargo will tell rustc to skip
codegen of all intermediate crates and only generate LLVM IR. Today
rustc will generate both object code and LLVM IR, but the object code is
later simply thrown away, wastefully.
This commit is an attempted resurrection of #70458 where LLVM bitcode
emitted by rustc into rlibs is stored into object file sections rather
than in a separate file. The main rationale for doing this is that when
rustc emits bitcode it will no longer use a custom compression scheme
which makes it both easier to interoperate with existing tools and also
cuts down on compile time since this compression isn't happening.
The blocker for this in #70458 turned out to be that native linkers
didn't handle the new sections well, causing the sections to either
trigger bugs in the linker or actually end up in the final linked
artifact. This commit attempts to address these issues by ensuring that
native linkers ignore the new sections by inserting custom flags with
module-level inline assembly.
Note that this does not currently change the API of the compiler at all.
The pre-existing `-C bitcode-in-rlib` flag is co-opted to indicate
whether the bitcode should be present in the object file or not.
Finally, note that an important consequence of this commit, which is also
one of its primary purposes, is to enable rustc's `-Clto` bitcode
loading to load rlibs produced with `-Clinker-plugin-lto`. The goal here
is that when you're building with LTO Cargo will tell rustc to skip
codegen of all intermediate crates and only generate LLVM IR. Today
rustc will generate both object code and LLVM IR, but the object code is
later simply thrown away, wastefully.
Suggest `into` instead of `try_into` if possible with int types
If it is possible to convert an integer type into another using `into`, don't suggest `try_into`. This commit changes the suggested method to convert from one integer type to another for the following cases:
- u{n} -> i{m} where n < m
- u8 -> isize
- i{n} -> isize where n <= 16
- u{n} -> usize where n <= 16
Fixes#71580
