Select obligations before processing wf obligation in `compare_method_predicate_entailment`
We need to select obligations before processing the WF obligation for the `IMPLIED_BOUNDS_ENTAILMENT` lint, since it skips over type variables.
Fixes#114783
r? `@jackh726`
TAITs do not constrain generic params
Fixes#108425
Not sure if I should rework those two failing tests. I guess `tests/ui/type-alias-impl-trait/coherence.rs` could just have the type parameter removed from it? IDK what `tests/ui/type-alias-impl-trait/coherence_generalization.rs` is even testing, though.
r? `@aliemjay`
cc `@lcnr` `@oli-obk` (when he's back from 🌴)
coverage: Store BCB counter info externally, not directly in the BCB graph
When deciding how to instrument the underlying MIR for coverage, the `InstrumentCoverage` pass builds a simplified “Basic Counter Block” graph, and then allocates coverage counters/expressions to various nodes/edges in the BCB graph as necessary. Those counters/expressions are then injected into the function's MIR.
The awkward thing here is that the code for doing this needs `&mut` access to the graph, in order to associate coverage info with individual nodes, even though it isn't making any structural changes to the graph itself. That makes it harder to understand and modify the instrumentation code.
In addition, the graph alone can't hold all the information that is needed. There ends up being an extra vector of “intermediate expressions” that needs to be passed around separately anyway.
---
This PR simplifies things by instead storing all of that temporary coverage information in a number of side-tables inside `CoverageCounters`.
This makes it easier to see all of the information produced by the make-counters step, and how it is used by the inject-into-mir step.
---
Looking at the combined changes is possible, but I recommend reviewing the commits individually, because the big changes are mostly independent of each other (despite being conceptually related).
Extract a create_wrapper_function for use in allocator shim writing
This deduplicates some logic and makes it easier to follow what wrappers are produced. In the future it may allow moving the code to determine which wrappers to create to cg_ssa.
All of them are not exported from rustc_interface and used only during
global_ctxt(). Inlining them makes it easier to follow the order of
queries and slightly reduces line count.
Also consider `mem::transmute` with the `invalid_reference_casting` lint
This PR extend the `invalid_reference_casting` lint with regard to the `std::mem::transmute` function.
```
error: casting `&T` to `&mut T` is undefined behavior, even if the reference is unused, consider instead using an `UnsafeCell`
--> $DIR/reference_casting.rs:27:16
|
LL | let _num = &mut *std::mem::transmute::<_, *mut i32>(&num);
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
```
*I encourage anyone reviewing this PR to do so [without whitespaces](https://github.blog/2011-10-21-github-secrets/#whitespace).*
rustc: Move `features` from `Session` to `GlobalCtxt`
Removes one more piece of mutable state.
Follow up to #114622.
The rule I used for passing feature in function signatures:
- if a crate already depends on `rustc_middle`, then `Session` is replaced with `TyCtxt`
- otherwise session and features are passed as a pair `sess: &Session, features: &Features`
The code in `rustc_lint` is ultimately used for implementing a trait from `rustc_expand`, so it also doesn't use tcx despite the dependency on `rustc_middle`.
normalize in `trait_ref_is_knowable` in new solver
fixes https://github.com/rust-lang/trait-system-refactor-initiative/issues/51
Alternatively we could avoid normalizing the self type and do this at the end of the `assemble_candidates_via_self_ty` stack by splitting candidates into:
- applicable without normalizing self type
- applicable for aliases, even if they can be normalized
- applicable for stuff which cannot get normalized further
I don't think this would have any significant benefits and it also seems non-trivial to avoid normalizing only the self type in `trait_ref_is_knowable`.
r? `@compiler-errors`
Storing coverage counter information in `CoverageCounters` has a few advantages
over storing it directly inside BCB graph nodes:
- The graph doesn't need to be mutable when making the counters, making it
easier to see that the graph itself is not modified during this step.
- All of the counter data is clearly visible in one place.
- It becomes possible to use a representation that doesn't correspond 1:1 to
graph nodes, e.g. storing all the edge counters in a single hashmap instead of
several.
remove builtin `Copy` and `Clone` impl for float and int infer
it's only change is whether `{integer}: Copy` is ambiguous, this has the following properties
- these goals get proven earlier, potentially resulting in slightly better perf
- it causes inconsistent behavior and ICE if there do not exist impls for all integers, causing issues when using `#[no_core]`
- it means `Clone` has user-facing differences from other traits from `core` with the new solver because it can potentially guide inference there
- it's just very sus™ to have a builtin impl which applies during type inference but not afterwards
Rollup of 7 pull requests
Successful merges:
- #94455 (Partially stabilize `int_roundings`)
- #114132 (Better Debug for Vars and VarsOs)
- #114584 (E0277 nolonger points at phantom `.await`)
- #114667 (Record binder for bare trait object in LifetimeCollectVisitor)
- #114692 (downgrade `internal_features` to warn)
- #114703 (Cover ParamConst in smir)
- #114734 (Mark oli as "on vacation")
r? `@ghost`
`@rustbot` modify labels: rollup
Respect `#[expect]` the same way `#[allow]` is with the `dead_code` lint
This PR makes the `#[expect]` attribute being respected in the same way the `#[allow]` attribute is with the `dead_code` lint.
The fix is much more involved than I would have liked (and it's not because I didn't tried!), because the implementation took advantage of the fact that firing a lint in a allow context is a nop (for the user, as the lint is suppressed) to not fire-it at all.
And will it's fine for `#[allow]`, it definitively isn't for `#[expect]`, as the presence and absence of the lint is significant. So a big part of the PR is just adding the context information of whenever an item is on the worklist because of an `[allow]`/`#[expect]` or not.
Fixes https://github.com/rust-lang/rust/issues/114557
downgrade `internal_features` to warn
Not sure if this requires an FCP or whatever. By having the lint as deny I need to modify test cases when testing them outside of the test suite as the test suite implicitly allows the lint. This takes maybe 10 to 20 seconds per test, but given just how frequently I end up copying tests to different repos it's a significant annoyance.
r? `@Nilstrieb`
make `typeid::typeid_itanium_cxx_abi::transform_ty` evaluate length in array types
the ICE in https://github.com/rust-lang/rust/issues/114275 was caused by `transform_ty`
in compiler/rustc_symbol_mangling/src/typeid/typeid_itanium_cxx_abi.rs encountering an unevaluated const, while expecting it to already be evaluated.
Rollup of 7 pull requests
Successful merges:
- #114599 (Add impl trait declarations to SMIR)
- #114622 (rustc: Move `crate_types` and `stable_crate_id` from `Session` to `GlobalCtxt`)
- #114662 (Unlock trailing where-clauses for lazy type aliases)
- #114693 (Remove myself from the review rotation)
- #114694 (make the provisional cache slightly less broken)
- #114705 (Add spastorino to mailmap)
- #114712 (Fix a couple of bad comments)
r? `@ghost`
`@rustbot` modify labels: rollup
Fix a couple of bad comments
A couple of nits I saw. Sorry, this really should be folded into some other PR of mine, but I will literally forget if I don't put these up now.
make the provisional cache slightly less broken
It is still broken for the following cycles:
```mermaid
graph LR
R["R: coinductive"] --> A["A: inductive"]
R --> B["B: coinductive"]
A --> B
B --> R
```
the `R -> A -> B -> R` cycle should be considered to not hold, as it is mixed, but because we first put `B` into the cache from the `R -> B -> R` cycle which is coinductive, it does hold.
This issue will also affect our new coinduction approach. Longterm cycles are coinductive as long as one step goes through an impl where-clause, see f4fc5bae36/crates/formality-prove/src/prove/prove_wc.rs (L51-L62). Here we would first have a fully inductive cycle `R -> B -> R` which is then entered by a cycle with a coinductive step `R -> A -coinductive-> B -> R`.
I don't know how to soundly implement a provisional cache for goals not on the stack without tracking all cycles the goal was involved in and whether they were inductive or not. We could then only use goals from the cache if the *inductivity?* of every cycle remained the same. This is a mess to implement. I therefore want to rip out the provisional cache entirely, but will wait with this until I talked about it with `@nikomatsakis.`
r? `@compiler-errors`
Unlock trailing where-clauses for lazy type aliases
Allows trailing where-clauses on lazy type aliases and forbids[^1] leading ones.
Completes #89122 (see section *Top-level type aliases*).
`@rustbot` label F-lazy_type_alias
r? `@oli-obk`
[^1]: This is absolutely fine since lazy type aliases are only meant to be stabilized as part of a new edition.
coverage: Don't convert filename/symbol strings to `CString` for FFI
LLVM APIs are usually perfectly happy to accept pointer/length strings, as long as we supply a suitable length value when creating a `StringRef` or `std::string`.
This lets us avoid quite a few intermediate `CString` copies during coverage codegen. It also lets us use an `IndexSet<Symbol>` (instead of an `IndexSet<CString>`) when building the deduplicated filename table.
Remove redundant calls to `resolve_vars_with_obligations`
I've been auditing the calls to `resolve_vars_with_obligations` for the new solver, and have found a few that have no effect on diagnostics. Let's just remove 'em.
Also remove a redundant `resolve_vars_with_obligations_and_mutate_fulfillment` call.
r? ``@lcnr``
`Expr::can_have_side_effects()` is incorrect for struct/enum/array/tuple literals
It would return 'false' unless *all* sub-expressions had side effects. This would easily allow side effects to slip through, and also wrongly label empty literals as having side effects. Add some tests for the last point
The function is only used for simple lints and error messages, so not a serious bug.
this ICE was caused by `transform_ty`
in compiler/rustc_symbol_mangling/src/typeid/typeid_itanium_cxx_abi.rs
encountering an unevaluated const, while expecting it to already be evaluated.
add a regression test
Update tests/ui/sanitize/issue-114275-cfi-const-expr-in-arry-len.rs
Co-authored-by: Michael Goulet <michael@errs.io>
Update tests/ui/sanitize/issue-114275-cfi-const-expr-in-arry-len.rs
Co-authored-by: Michael Goulet <michael@errs.io>
fix test compiling for targets with -crt-static and failing
this was causign https://github.com/rust-lang/rust/pull/114686 to fail
Rollup of 6 pull requests
Successful merges:
- #110435 (rustdoc-json: Add test for field ordering.)
- #111891 (feat: `riscv-interrupt-{m,s}` calling conventions)
- #114377 (test_get_dbpath_for_term(): handle non-utf8 paths (fix FIXME))
- #114469 (Detect method not found on arbitrary self type with different mutability)
- #114587 (Convert Const to Allocation in smir)
- #114670 (Don't use `type_of` to determine if item has intrinsic shim)
Failed merges:
- #114599 (Add impl trait declarations to SMIR)
r? `@ghost`
`@rustbot` modify labels: rollup
Don't use `type_of` to determine if item has intrinsic shim
When we're calling `resolve_instance` on an inline const, we were previously looking at the `type_of` for that const, seeing that it was an `extern "intrinsic"` fn def, and treating it as if we were computing the instance of that intrinsic itself. This is incorrect.
Instead, we should be using the def-id of the item we're computing to determine if it's an intrinsic.
Fixes#114660
Detect method not found on arbitrary self type with different mutability
```
error[E0599]: no method named `x` found for struct `Pin<&S>` in the current scope
--> $DIR/arbitrary_self_type_mut_difference.rs:11:18
|
LL | Pin::new(&S).x();
| ^ help: there is a method with a similar name: `y`
|
note: method is available for `Pin<&mut S>`
--> $DIR/arbitrary_self_type_mut_difference.rs:6:5
|
LL | fn x(self: Pin<&mut Self>) {}
| ^^^^^^^^^^^^^^^^^^^^^^^^^^
```
Related to #57994, as one of the presented cases can lead to code like this.
feat: `riscv-interrupt-{m,s}` calling conventions
Similar to prior support added for the mips430, avr, and x86 targets this change implements the rough equivalent of clang's [`__attribute__((interrupt))`][clang-attr] for riscv targets, enabling e.g.
```rust
static mut CNT: usize = 0;
pub extern "riscv-interrupt-m" fn isr_m() {
unsafe {
CNT += 1;
}
}
```
to produce highly effective assembly like:
```asm
pub extern "riscv-interrupt-m" fn isr_m() {
420003a0: 1141 addi sp,sp,-16
unsafe {
CNT += 1;
420003a2: c62a sw a0,12(sp)
420003a4: c42e sw a1,8(sp)
420003a6: 3fc80537 lui a0,0x3fc80
420003aa: 63c52583 lw a1,1596(a0) # 3fc8063c <_ZN12esp_riscv_rt3CNT17hcec3e3a214887d53E.0>
420003ae: 0585 addi a1,a1,1
420003b0: 62b52e23 sw a1,1596(a0)
}
}
420003b4: 4532 lw a0,12(sp)
420003b6: 45a2 lw a1,8(sp)
420003b8: 0141 addi sp,sp,16
420003ba: 30200073 mret
```
(disassembly via `riscv64-unknown-elf-objdump -C -S --disassemble ./esp32c3-hal/target/riscv32imc-unknown-none-elf/release/examples/gpio_interrupt`)
This outcome is superior to hand-coded interrupt routines which, lacking visibility into any non-assembly body of the interrupt handler, have to be very conservative and save the [entire CPU state to the stack frame][full-frame-save]. By instead asking LLVM to only save the registers that it uses, we defer the decision to the tool with the best context: it can more accurately account for the cost of spills if it knows that every additional register used is already at the cost of an implicit spill.
At the LLVM level, this is apparently [implemented by] marking every register as "[callee-save]," matching the semantics of an interrupt handler nicely (it has to leave the CPU state just as it found it after its `{m|s}ret`).
This approach is not suitable for every interrupt handler, as it makes no attempt to e.g. save the state in a user-accessible stack frame. For a full discussion of those challenges and tradeoffs, please refer to [the interrupt calling conventions RFC][rfc].
Inside rustc, this implementation differs from prior art because LLVM does not expose the "all-saved" function flavor as a calling convention directly, instead preferring to use an attribute that allows for differentiating between "machine-mode" and "superivsor-mode" interrupts.
Finally, some effort has been made to guide those who may not yet be aware of the differences between machine-mode and supervisor-mode interrupts as to why no `riscv-interrupt` calling convention is exposed through rustc, and similarly for why `riscv-interrupt-u` makes no appearance (as it would complicate future LLVM upgrades).
[clang-attr]: https://clang.llvm.org/docs/AttributeReference.html#interrupt-risc-v
[full-frame-save]: 9281af2ecf/src/lib.rs (L440-L469)
[implemented by]: b7fb2a3fec/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp (L61-L67)
[callee-save]: 973f1fe7a8/llvm/lib/Target/RISCV/RISCVCallingConv.td (L30-L37)
[rfc]: https://github.com/rust-lang/rfcs/pull/3246
fix: not insert missing lifetime for `ConstParamTy`
Fixes#113462
We should ignore the missing lifetime, as it's illegal to include a lifetime in a const param.
r? ``@compiler-errors``
These new interrupt calling conventions are not themselves stabilized,
but there are other unstable calling conventions present in the SMIR
mapping (e.g. AVR interrupts) and the mapping appears to be "complete"
so far, with no obvious way to represent unstable conventions separately
from the stable ones.
Similar to prior support added for the mips430, avr, and x86 targets
this change implements the rough equivalent of clang's
[`__attribute__((interrupt))`][clang-attr] for riscv targets, enabling
e.g.
```rust
static mut CNT: usize = 0;
pub extern "riscv-interrupt-m" fn isr_m() {
unsafe {
CNT += 1;
}
}
```
to produce highly effective assembly like:
```asm
pub extern "riscv-interrupt-m" fn isr_m() {
420003a0: 1141 addi sp,sp,-16
unsafe {
CNT += 1;
420003a2: c62a sw a0,12(sp)
420003a4: c42e sw a1,8(sp)
420003a6: 3fc80537 lui a0,0x3fc80
420003aa: 63c52583 lw a1,1596(a0) # 3fc8063c <_ZN12esp_riscv_rt3CNT17hcec3e3a214887d53E.0>
420003ae: 0585 addi a1,a1,1
420003b0: 62b52e23 sw a1,1596(a0)
}
}
420003b4: 4532 lw a0,12(sp)
420003b6: 45a2 lw a1,8(sp)
420003b8: 0141 addi sp,sp,16
420003ba: 30200073 mret
```
(disassembly via `riscv64-unknown-elf-objdump -C -S --disassemble ./esp32c3-hal/target/riscv32imc-unknown-none-elf/release/examples/gpio_interrupt`)
This outcome is superior to hand-coded interrupt routines which, lacking
visibility into any non-assembly body of the interrupt handler, have to
be very conservative and save the [entire CPU state to the stack
frame][full-frame-save]. By instead asking LLVM to only save the
registers that it uses, we defer the decision to the tool with the best
context: it can more accurately account for the cost of spills if it
knows that every additional register used is already at the cost of an
implicit spill.
At the LLVM level, this is apparently [implemented by] marking every
register as "[callee-save]," matching the semantics of an interrupt
handler nicely (it has to leave the CPU state just as it found it after
its `{m|s}ret`).
This approach is not suitable for every interrupt handler, as it makes
no attempt to e.g. save the state in a user-accessible stack frame. For
a full discussion of those challenges and tradeoffs, please refer to
[the interrupt calling conventions RFC][rfc].
Inside rustc, this implementation differs from prior art because LLVM
does not expose the "all-saved" function flavor as a calling convention
directly, instead preferring to use an attribute that allows for
differentiating between "machine-mode" and "superivsor-mode" interrupts.
Finally, some effort has been made to guide those who may not yet be
aware of the differences between machine-mode and supervisor-mode
interrupts as to why no `riscv-interrupt` calling convention is exposed
through rustc, and similarly for why `riscv-interrupt-u` makes no
appearance (as it would complicate future LLVM upgrades).
[clang-attr]: https://clang.llvm.org/docs/AttributeReference.html#interrupt-risc-v
[full-frame-save]: 9281af2ecf/src/lib.rs (L440-L469)
[implemented by]: b7fb2a3fec/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp (L61-L67)
[callee-save]: 973f1fe7a8/llvm/lib/Target/RISCV/RISCVCallingConv.td (L30-L37)
[rfc]: https://github.com/rust-lang/rfcs/pull/3246
Restrict linker version script of proc-macro crates to just its two symbols
Restrict linker version script of proc-macro crates to just the two symbols of each proc-macro crate.
The main known effect of doing this is to stop including `#[no_mangle]` symbols in the linker version script.
Background:
The combination of a proc-macro crate with an import of another crate that itself exports a no_mangle function was broken for a period of time, because:
* In PR #99944 we stopped exporting no_mangle symbols from proc-macro crates; proc-macro crates have a very limited interface and are meant to be treated as a blackbox to everything except rustc itself. However: he constructed linker version script still referred to them, but resolving that discrepancy was left as a FIXME in the code, tagged with issue #99978.
* In PR #108017 we started telling the linker to check (via the`--no-undefined-version` linker invocation flag) that every symbol referenced in the "linker version script" is provided as linker input. So the unresolved discrepancy from #99978 started surfacing as a compile-time error (e.g. #111888).
Fix#111888Fix#99978.
Allowing re-implementation of mir_drops_elaborated query
For our use case of the rust compiler interface (a rust verifier called [Prusti](https://github.com/viperproject/prusti-dev/)), it would be extremely useful if we were able to "copy" the implementation of the `mir_drops_elaborated_and_const_checked` query to override it. This would mean that the following items would need to be made public:
>6d55184d05/compiler/rustc_mir_transform/src/lib.rs (L434)
>6d55184d05/compiler/rustc_mir_transform/src/inline.rs (L32)
(for the latter its module needs to be public or it needs to be re-exported)
To explain why (we think) this is necessary: I am currently working on a new feature, where we try to modify the generated executables by inserting certain additional checks, and potentially perform some optimizations based on verification results.
We are using the rust compiler interface and most of our goals can be achieved by overriding queries, in our case this is currently `mir_drops_elaborated_and_const_checked`.
However, at the moment this approach is somewhat limited. When overriding queries, we can call and steal the base-query and then modify the results before allocating and returning those.
The problem is that the verification works with a copy of `mir_promoted`. For the modifications we want to make to the mir, we would often want to rely on results of the verifier that refer to Locations in the `mir_promoted`. We can not modify the `mir_promoted` query using these results, because to run the verification we also need the results of `mir_borrowck()`, which means `mir_promoted` will already be constructed and cached.
The Locations we get from the verifier are also no longer usable to modify `mir_drops_elaborated_and_const_checked`, because the MIR obviously changes between those 2 phases. Tracking all Locations between the two seems to be pretty much unfeasible, and would also be extremely unstable.
By being able to override the query with its original implementation, we could modify the MIR before drop elaboration and the various other passes are performed.
I have spent quite a bit of time investigating other solutions, and didn't find any other way solving this problem. If I still missed something I would of course be happy to hear any suggestions that do not require exposing more internal compiler functionality. However, I think being able to re-implement certain queries could also benefit other use cases in the future, for example in PR #108328 one of the approaches discussed involved doing the same thing for `mir_promoted`.
