Implicit coercions from references to pointers were lowered to slightly
different Mir than explicit casts (e.g. 'foo as *mut T'). This resulted
in certain uses of self-referential structs compiling correctly when an
explicit cast was used, but not when the implicit coercion was used.
To fix this, this commit adds an outer 'Use' expr when applying a
raw-ptr-borrow adjustment. This makes the lowered Mir for coercions
identical to that of explicit coercions, allowing the original code to
compile regardless of how the raw ptr cast occurs.
Fixes#47722
Fix never-type rvalue ICE
This fixes#43061.
r? @nikomatsakis
A small post-mortem as a follow-up to our investigations in https://github.com/rust-lang/rust/pull/47291:
The problem as I understand it is that when `NeverToAny` coercions are made, the expression/statement that is coerced may be enclosed in a block. In our case, the statement `x;` was being transformed to something like: `NeverToAny( {x;} )`. Then, `NeverToAny` is transformed into an expression:
000fbbc9b8/src/librustc_mir/build/expr/into.rs (L52-L59)
Which ends up calling `ast_block_stmts` on the block `{x;}`, which triggers this condition:
000fbbc9b8/src/librustc_mir/build/block.rs (L141-L147)
In our case, there is no return expression, so `push_assign_unit` is called. But the block has already been recorded as _diverging_, meaning the result of the block will be assigned to a location of type `!`, rather than `()`. This causes the MIR error.
I'm assuming the `NeverToAny` coercion code is doing what it's supposed to (there don't seem to be any other problems), so fixing the issue simply consists of checking that the destination for the return value actually _is_ supposed to be a unit. (If no return value is given, the only other possible type for the return value is `!`, which can just be ignored, as it will be unreachable anyway.)
I checked the other cases of `push_assign_unit`, and it didn't look like they could be affected by the divergence issue (blocks are kind of special-cased in this regard as far as I can tell), so this should be sufficient to fix the issue.
Add CGU size heuristic for partitioning
This addresses the concern of #47316 by estimating CGU size based on
the size of its MIR. Looking at the size estimate differences for a
small selection of crates, this heuristic produces different orderings,
which should more accurately reflect optimisation time. (Fixes #47316.)
r? @michaelwoerister
renumber regions in generators
This fixes#47189, but I think we still have to double check various things around how to treat generators in MIR type check + borrow check (e.g., what borrows should be invalidated by a `Suspend`? What consistency properties should type check be enforcing anyway around the "interior" type?)
Also fixes#47587 thanks to @spastorino's commit.
r? @pnkfelix
Custom error when moving arg outside of its closure
When given the following code:
```rust
fn give_any<F: for<'r> FnOnce(&'r ())>(f: F) {
f(&());
}
fn main() {
let mut x = None;
give_any(|y| x = Some(y));
}
```
provide a custom error:
```
error: borrowed data cannot be moved outside of its closure
--> file.rs:7:27
|
6 | let mut x = None;
| ----- borrowed data cannot be moved into here...
7 | give_any(|y| x = Some(y));
| --- ^ cannot be moved outside of its closure
| |
| ...because it cannot outlive this closure
```
instead of the generic lifetime error:
```
error[E0495]: cannot infer an appropriate lifetime due to conflicting requirements
--> file.rs:7:27
|
7 | give_any(|y| x = Some(y));
| ^
|
note: first, the lifetime cannot outlive the anonymous lifetime #2 defined on the body at 7:14...
--> file.rs:7:14
|
7 | give_any(|y| x = Some(y));
| ^^^^^^^^^^^^^^^
note: ...so that expression is assignable (expected &(), found &())
--> file.rs:7:27
|
7 | give_any(|y| x = Some(y));
| ^
note: but, the lifetime must be valid for the block suffix following statement 0 at 6:5...
--> file.rs:6:5
|
6 | / let mut x = None;
7 | | give_any(|y| x = Some(y));
8 | | }
| |_^
note: ...so that variable is valid at time of its declaration
--> file.rs:6:9
|
6 | let mut x = None;
| ^^^^^
```
Fix#45983.
remove bogus assertion and comments
The code (incorrectly) assumed that constants could not have generics
in scope, but it's not really a problem if they do.
Fixes#47153
r? @pnkfelix
avoid double-unsizing arrays in bytestring match lowering
The match lowering code, when lowering matches against bytestrings,
works by coercing both the scrutinee and the pattern to `&[u8]` and
then comparing them using `<[u8] as Eq>::eq`.
If the scrutinee is already of type `&[u8]`, then unsizing it is both
unneccessary and a trait error caught by the new and updated MIR typeck,
so this PR changes lowering to avoid doing that (match lowering tried to
avoid that before, but that attempt was quite broken).
Fixes#46920.
r? @eddyb
This addresses the concern of #47316 by estimating CGU size based on
the size of its MIR. Looking at the size estimate differences for a
small selection of crates, this heuristic produces different orderings,
which should more accurately reflect optimisation time.
Fixes#47316.
remove noop landing pads in cleanup shims
No-op landing pads are already removed in the normal optimization pipeline - so also removing them on the shim pipeline should slightly improve codegen performance, as these cleanup blocks are known to hurt LLVM.
This un-regresses and is therefore a fix for #47442. However, the reporter of that issue should try using `-C panic=abort` instead of carefully avoiding panics.
r? @eddyb
Add a default directory for -Zmir-dump-dir
The current behaviour of dumping in the current directory is rarely
desirable: a sensible default directory for dumping is much more
convenient. This makes sets the default value for `-Zmir-dump-dir`
to `mir_dump/`.
r? @eddyb
The match lowering code, when lowering matches against bytestrings,
works by coercing both the scrutinee and the pattern to `&[u8]` and
then comparing them using `<[u8] as Eq>::eq`.
If the scrutinee is already of type `&[u8]`, then unsizing it is both
unneccessary and a trait error caught by the new and updated MIR typeck,
so this PR changes lowering to avoid doing that (match lowering tried to
avoid that before, but that attempt was quite broken).
Fixes#46920.
These are already removed in the normal optimization pipeline - so this
should slightly improve codegen performance, as these cleanup blocks are
known to hurt LLVM.
This un-regresses and is therefore a fix for #47442. However, the
reporter of that issue should try using `-C panic=abort` instead of
carefully avoiding panics.
No longer parse it.
Remove AutoTrait variant from AST and HIR.
Remove backwards compatibility lint.
Remove coherence checks, they make no sense for the new syntax.
Remove from rustdoc.
Use the new fs_read_write functions in rustc internals
Uses `fs::read` and `fs::write` (added by #45837) where appropriate, to simplify code and dog-food these new APIs. This also improves performance, when combined with #47324.
Shorten names of some compiler generated artifacts.
This PR makes the compiler mangle codegen unit names by default. The name of every codegen unit name will now be a random string of 16 characters. It also makes the file extensions of some intermediate compiler products shorter. Hopefully, these changes will reduce the pressure on tools with path length restrictions like buildbot. The change should also solve problems with case-insensitive file system.
cc #47186 and #47222
r? @alexcrichton
Make normalize_and_test_predicates into a query
From #44891.
I'm not real solid on how `dep_graph` stuff works, but if a node is going to have a key (again, not sure how important that is), then the key needs to be `Copy`. So since `normalize_and_test_predicates` only had one out-of-module use, I changed that call site to use a new function, `substitute_normalize_and_test_predicates` which is the query and enables having the arguments be `Copy`. Hopefully this makes sense.
r? @nikomatsakis
and/or @michaelwoerister
rustc: use {U,I}size instead of {U,I}s shorthands.
`Us`/`Is` come from a time when `us` and `is` were the literal suffixes that are now `usize` / `isize`.
r? @nikomatsakis
Bump to 1.25.0
* Bump the release version to 1.25
* Bump the bootstrap compiler to the recent beta
* Allow using unstable rustdoc features on beta - this fix has been applied to
the beta branch but needed to go to the master branch as well.
Reword reason for move note
On move errors, when encountering an enum variant, be more ambiguous and do not refer to the type on the cause note, to avoid referring to `(maybe as std::prelude::v1::Some).0`, and instead refer to `the value`.
Sidesteps part of the problem with #41962:
```
error[E0382]: use of partially moved value: `maybe`
--> file.rs:5:30
|
5 | if let Some(thing) = maybe {
| ----- ^^^^^ value used here after move
| |
| value moved here
= note: move occurs because the value has type `std::vec::Vec<bool>`, which does not implement the `Copy` trait
error[E0382]: use of moved value: `(maybe as std::prelude::v1::Some).0`
--> file.rs:5:21
|
5 | if let Some(thing) = maybe {
| ^^^^^ value moved here in previous iteration of loop
= note: move occurs because the value has type `std::vec::Vec<bool>`, which does not implement the `Copy` trait
error: aborting due to 2 previous errors
```
Previous discussion: #44360
r? @arielb1
NLL fixes
First, introduce pre-statement effects to dataflow to fix#46875. Edge dataflow effects might make that redundant, but I'm not sure of the best way to integrate them with liveness etc., and if this is a hack, this is one of the cleanest hacks I've seen.
And I want a small fix to avoid the torrent of bug reports.
Second, fix linking of projections to fix#46974
r? @pnkfelix
Do not expand a derive invocation when derive is not allowed
Closes#46655.
The first commit is what actually closes#46655. The second one is just a refactoring I have done while waiting on a test.
Implements RFC 1937: `?` in `main`
This is the first part of the RFC 1937 that supports new
`Termination` trait in the rust `main` function.
Thanks @nikomatsakis, @arielb1 and all other people in the gitter channel for all your help!
The support for doctest and `#[test]` is still missing, bu as @nikomatsakis said, smaller pull requests are better :)
[MIR Borrowck] Moveck inline asm statements
Closes#45695
New behavior:
* Input operands to `asm!` are moved, direct output operands are initialized.
* Direct, non-read-write outputs match the assignment changes in #46752 (Shallow writes, end borrows).
Prevent unwinding past FFI boundaries
Second attempt to write a patch to solve this.
r? @nikomatsakis
~~So, my biggest issue with this patch is the way the patch determines *what* functions should have an abort landing pad (in `construct_fn`). I would ideally have this code match [src/librustc_trans/callee.rs::get_fn](https://github.com/rust-lang/rust/blob/master/src/librustc_trans/callee.rs#L107-L115) but couldn't find an id that returns true for `is_foreign_item`. Also tried `tcx.has_attr("unwind")` with no luck.~~ FIXED
Other issues:
* llvm.trap is an SIGILL on amd64. Ideally we could use panic-abort's version of aborting which is nicer but we don't want to depend on that library...
* ~~Mir inlining is a stub currently.~~ FIXED (no-op)
Also, when reviewing please take into account that I'm new to the code and only partially know what I'm doing... and that I've mostly made made matches on `TerminatorKind::Abort` match either `TerminatorKind::Resume` or `TerminatorKind::Unreachable` based on what looked best.
Issue #46589 - Kill borrows on a local variable whenever we assign ov…
…er this variable
This is a first patch for the issue, handling the simple case while I figure out the data structures involved in the more complex cases.
rustc: Sort CGUs before merging
This commit fixes some nondeterminism in compilation when using multiple codegen
units. The algorithm for splitting codegen units currently takes the
otherwise-would-be-for-incremental partitioning and then continuously merges the
two smallest codegen units until the desired number of codegen units are
reached.
We want to be sure to merge the same codegen units each time a compilation is
run but there's some subtle reorderings amongst all the items which was causing
this step to be slightly buggy. Notably this step involves sorting codegen units
by size, but if two codegen units had the same size they would appear in
different locations in the list each time.
This commit fixes this issue by sorting codegen units by name before doing the
loop to merge the two smallest. This means that we've got a deterministic
order going in and since we're using a stable sort this should mean that we're
always now getting a deterministic merging of codegen units.
Closes#46846
Generics refactoring (groundwork for const generics)
These changes were suggested by @eddyb.
After this change, the `Generics` contain one `Vec` of an enum for the generic parameters, rather than two separate `Vec`s for lifetime and type parameters. Type params and const params will need to be in a shared `Vec` to preserve their ordering, and moving lifetimes into the same `Vec` should simplify the code that processes `Generics`.
This commit fixes some nondeterminism in compilation when using multiple codegen
units. The algorithm for splitting codegen units currently takes the
otherwise-would-be-for-incremental partitioning and then continuously merges the
two smallest codegen units until the desired number of codegen units are
reached.
We want to be sure to merge the same codegen units each time a compilation is
run but there's some subtle reorderings amongst all the items which was causing
this step to be slightly buggy. Notably this step involves sorting codegen units
by size, but if two codegen units had the same size they would appear in
different locations in the list each time.
This commit fixes this issue by sorting codegen units by name before doing the
loop to merge the two smallest. This means that we've got a deterministic
order going in and since we're using a stable sort this should mean that we're
always now getting a deterministic merging of codegen units.
Closes#46846
MIR: terminate unreachable blocks in construct_const
Fixes#46843.
#45821 added unreachable blocks in matches, which were terminated in
construct_fn but not in construct_const, causing a panic due to "no
terminator on block" when constants involved matching on enums.
The "unimplemented expression type" error may go away in the future, the
key is that we see the E0015 about using a non-const function and then
don't ICE.
The Generics now contain one Vec of an enum for the generic parameters,
rather than two separate Vec's for lifetime and type parameters.
Additionally, places that previously used Vec<LifetimeDef> now use
Vec<GenericParam> instead.
Ensure separate activations only occur for assignments to locals
Ensure separate activations only occur for assignments to locals, not projections.
Fix#46746.
(I didn't make a regression test because we do not really have a good way to directly express the case that we are trying to catch, because we cannot write MIR directly.)
In particular, -Znll might as well imply -Zborrowck=mir by default,
just like `#![feature(nll)]` does.
Also, if NLL is in use, no reason to emit end regions. The NLL pass
just strips them out anyway.
Fixes#46843.
#45821 added unreachable blocks in matches, which were terminated in
construct_fn but not in construct_const, causing a panic due to "no
terminator on block" when constants involved matching on enums.
The "unimplemented expression type" error may go away in the future, the
key is that we see the E0015 about using a non-const function and then
don't ICE.
Split PlaceContext::Store into Store & AsmOutput
Outputs in InlineAsm can be read-write, so splitting it out is useful for things like Store-Store folding, as that's unsound for a Store-AsmOutput.
This PR is intended to make no changes, just be the mechanical split of the enum. Future changes can use the split, like a MIR pass I'm working on and perhaps two-phase borrows (see this FIXME: https://github.com/rust-lang/rust/pull/46852/files#diff-74dcd7740ab2104cd2b9a3b68dd4f208R543)
nll part 5
Next round of changes from the nll-master branch.
Extensions:
- we now propagate ty-region-outlives constraints out of closures and into their creator when necessary
- we fix a few ICEs that can occur by doing liveness analysis (and the resulting normalization) during type-checking
- we handle the implicit region bound that assumes that each type `T` outlives the fn body
- we handle normalization of inputs/outputs in fn signatures
Not included in this PR (will come next):
- handling `impl Trait`
- tracking causal information
- extended errors
r? @arielb1
Outputs in InlineAsm can be read-write, so splitting it out is useful for things like Store-Store folding, as it cannot be done for a Store-AsmOutput.
This PR is intended to make no changes, just be the mechanical split of the enum. Future changes can use the split, like a MIR pass I'm working on and perhaps two-phase borrows.
Point at var in short lived borrows instead of drop location
For RLS' sake, point at the borrow location as primary span for short lived borrows, instead of the borrow drop location.
Fix#39268.
The input/output types found in `UniversalRegions` are not normalized.
The old code used to assign them directly into the MIR, which would
lead to errors when there was a projection in a argument or return
type. This also led to some special cases in the `renumber` code.
We now renumber uniformly but then pass the input/output types into
the MIR type-checker, which equates them with the types found in MIR.
This allows us to normalize at the same time.
This allows us to re-use the `normalize` method on `TypeCheck`, which
is important since normalization may create fresh region
variables. This is not an ideal solution, though, since the current
representation of "liveness constraints" (a vector of (region, point)
pairs) is rather inefficient. Could do somewhat better by converting
to indices, but it'd still be less good than the older code. Unclear
how important this is.
Before, we would always have a `Some` ClosureRegionRequirements if we
were inferring values for a closure. Now we only do is it has a
non-empty set of outlives requirements.
[MIR-borrowck] Two phase borrows
This adds limited support for two-phase borrows as described in
http://smallcultfollowing.com/babysteps/blog/2017/03/01/nested-method-calls-via-two-phase-borrowing/
The support is off by default; you opt into it via the flag `-Z two-phase-borrows`
I have written "*limited* support" above because there are simple variants of the simple `v.push(v.len())` example that one would think should work but currently do not, such as the one documented in the test compile-fail/borrowck/two-phase-reservation-sharing-interference-2.rs
(To be clear, that test is not describing something that is unsound. It is just providing an explicit example of a limitation in the implementation given in this PR. I have ideas on how to fix, but I want to land the work that is in this PR first, so that I can stop repeatedly rebasing this branch.)
Instead, filter out (non-)conflicts of activiations with themselves in
the same manner that we filter out non-conflict between an activation
and its reservation.
Validate miri against the HIR const evaluator
r? @eddyb
cc @alexcrichton @arielb1 @RalfJung
The interesting parts are the last few functions in `librustc_const_eval/eval.rs`
* We warn if miri produces an error while HIR const eval does not.
* We warn if miri produces a value that does not match the value produced by HIR const eval
* if miri succeeds and HIR const eval fails, nothing is emitted, but we still return the HIR error
* if both error, nothing is emitted and the HIR const eval error is returned
So there are no actual changes, except that miri is forced to produce the same values as the old const eval.
* This does **not** touch the const evaluator in trans at all. That will come in a future PR.
* This does **not** cause any code to compile that didn't compile before. That will also come in the future
It would be great if someone could start a crater run if travis passes
Instead we are "just" careful to invoke it (which sets up a bunch of kill bits)
before we go into the code that sets up the gen bits.
That way, when the gen bits are set up, they will override any
previously set kill-bits for those reservations or activations.
Since we are now checking activation points, I removed one of the
checks at the reservation point. (You can see the effect this had on
two-phase-reservation-sharing-interference-2.rs)
Also, since we now have checks at both the reservation point and the
activation point, we sometimes would observe duplicate errors (since
either one independently interferes with another mutable borrow). To
deal with this, I used a similar strategy to one used as discussed on
issue #45360: keep a set of errors reported (in this case for
reservations), and then avoid doing the checks for the corresponding
activations. (This does mean that some errors could get masked, namely
for conflicting borrows that start after the reservation but still
conflict with the activation, which is unchecked when there was an
error for the reservation. But this seems like a reasonable price to
pay.)
High-level picture: The old `Borrows` analysis is now called
`Reservations` (implemented as a newtype wrapper around `Borrows`);
this continues to compute whether a `Rvalue::Ref` can reach a
statement without an intervening `EndRegion`. In addition, we also
track what `Place` each such `Rvalue::Ref` was immediately assigned
to in a given borrow (yay for MIR-structural properties!).
The new `ActiveBorrows` analysis then tracks the initial use of any of
those assigned `Places` for a given borrow. I.e. a borrow becomes
"active" immediately after it starts being "used" in some way. (This
is conservative in the sense that we will treat a copy `x = y;` as a
use of `y`; in principle one might further delay activation in such
cases.)
The new `ActiveBorrows` analysis needs to take the `Reservations`
results as an initial input, because the reservation state influences
the gen/kill sets for `ActiveBorrows`. In particular, a use of `a`
activates a borrow `a = &b` if and only if there exists a path (in the
control flow graph) from the borrow to that use. So we need to know if
the borrow reaches a given use to know if it really gets a gen-bit or
not.
* Incorporating the output from one dataflow analysis into the input
of another required more changes to the infrastructure than I had
expected, and even after those changes, the resulting code is still
a bit subtle.
* In particular, Since we need to know the intrablock reservation
state, we need to dynamically update a bitvector for the
reservations as we are also trying to compute the gen/kills
bitvector for the active borrows.
* The way I ended up deciding to do this (after also toying with at
least two other designs) is to put both the reservation state and
the active borrow state into a single bitvector. That is why we now
have separate (but related) `BorrowIndex` and
`ReserveOrActivateIndex`: each borrow index maps to a pair of
neighboring reservation and activation indexes.
As noted above, these changes are solely adding the active borrows
dataflow analysis (and updating the existing code to cope with the
switch from `Borrows` to `Reservations`). The code to process the
bitvector in the borrow checker currently just skips over all of the
active borrow bits.
But atop this commit, one *can* observe the analysis results by
looking at the graphviz output, e.g. via
```rust
#[rustc_mir(borrowck_graphviz_preflow="pre_two_phase.dot",
borrowck_graphviz_postflow="post_two_phase.dot")]
```
Includes doc for `FindPlaceUses`, as well as `Reservations` and
`ActiveBorrows` structs, which are wrappers are the `Borrows` struct
that dictate which flow analysis should be performed.
This is meant to ease development of multi-stage dataflow analyses
where the output from one analysis is used to initialize the state
for the next; in such a context, you cannot start with `bottom_value`
for all the bits.
Converting a `RegionElementIndex` to a `Location` is O(n) though could
trivially be O(log n), but we don't do it that much anyhow -- just on
error and debugging.
If the gen/kill bits are set there, the effects of `start_block_effects`
will not be seen when using `FlowAtLocation` etc. to go over the MIR.
EverInitializedLvals is the only pass that got this wrong, but this
fixes the footgun for everyone.
