Refactor away RBML from rustc_metadata.
RBML and `ty{en,de}code` have had their long-overdue purge. Summary of changes:
* Metadata is now a tree encoded in post-order and with relative backward references pointing to children nodes. With auto-deriving and type safety, this makes maintenance and adding new information to metadata painless and bug-free by default. It's also more compact and cache-friendly (cache misses should be proportional to the depth of the node being accessed, not the number of siblings as in EBML/RBML).
* Metadata sizes have been reduced, for `libcore` it went down 16% (`8.38MB` -> `7.05MB`) and for `libstd` 14% (`3.53MB` -> `3.03MB`), while encoding more or less the same information
* Specialization is used in the bundled `libserialize` (crates.io `rustc_serialize` remains unaffected) to customize the encoding (and more importantly, decoding) of various types, most notably those interned in the `TyCtxt`. Some of this abuses a soundness hole pending a fix (cc @aturon), but when that fix arrives, we'll move to macros 1.1 `#[derive]` and custom `TyCtxt`-aware serialization traits.
* Enumerating children of modules from other crates is now orthogonal to describing those items via `Def` - this is a step towards bridging crate-local HIR and cross-crate metadata
* `CrateNum` has been moved to `rustc` and both it and `NodeId` are now newtypes instead of `u32` aliases, for specializing their decoding. This is `[syntax-breaking]` (cc @Manishearth ).
cc @rust-lang/compiler
trans: Only instantiate #[inline] functions in codegen units referencing them
This PR changes how `#[inline]` functions are translated. Before, there was one "master instance" of the function with `external` linkage and a number of on-demand instances with `available_externally` linkage in each codegen unit that referenced the function. This had two downsides:
* Public functions marked with `#[inline]` would be present in machine code of libraries unnecessarily (see #36280 for an example)
* LLVM would crash on `i686-pc-windows-msvc` due to what I suspect to be a bug in LLVM's Win32 exception handling code, because it doesn't like `available_externally` there (#36309).
This PR changes the behavior, so that there is no master instance and only on-demand instances with `internal` linkage. The downside of this is potential code-bloat if LLVM does not completely inline away the `internal` instances because then there'd be N instances of the function instead of 1. However, this can only become a problem when using more than one codegen unit per crate.
cc @rust-lang/compiler
librustc_mir: Implement def-use chains and trivial copy propagation on MIR.
This only supports trivial cases in which there is exactly one def and
one use.
Currently, some random unrelated MIR tests are failing, probably just because they haven't been updated.
r? @eddyb
Fix overflow checking in unsigned pow()
The pow() method for unsigned integers produced 0 instead of trapping overflow for certain inputs. Calls such as 2u32.pow(1024) produced 0 when they should trap an overflow. This also adds tests for the correctly handling overflow in unsigned pow().
This was previously fixed for signed integers in #28248, but it seems unsigned integers got missed that time.
For issue number #34913
Remove data structure specialization for .zip() iterator
Go back on half the specialization, the part that changed the Zip
struct's fields themselves depending on the types of the iterators.
Previous PR: #33090
This means that the Zip iterator will always carry two usize fields,
which are sometimes unused. If a whole for loop using a .zip() iterator is
inlined, these are simply removed and have no effect.
The same improvement for Zip of for example slice iterators remain, and
they still optimize well. However, like when the specialization of zip
was merged, the compiler is still very sensistive to the exact context.
For example this code only autovectorizes if the function is used, not
if the code in zip_sum_i32 is inserted inline where it was called:
```rust
fn zip_sum_i32(xs: &[i32], ys: &[i32]) -> i32 {
let mut s = 0;
for (&x, &y) in xs.iter().zip(ys) {
s += x * y;
}
s
}
fn zipdot_i32_default_zip(b: &mut test::Bencher)
{
let xs = vec![1; 1024];
let ys = vec![1; 1024];
b.iter(|| {
zip_sum_i32(&xs, &ys)
})
}
```
Include a test that checks that `Zip<T, U>` is covariant w.r.t. T and U.
Fixes#35727
Avoid loading and parsing unconfigured non-inline modules.
For example, `#[cfg(any())] mod foo;` will always compile after this PR, even if `foo.rs` and `foo/mod.rs` do not exist or do not contain valid Rust.
Fixes#36478 and fixes#27873.
r? @nrc
trans: Let the collector find drop-glue for all vtables, not just VTableImpl.
This fixes#36260. So far, the collector has only recorded drop-glue for insertion into a vtable if the vtable was for an impl. But there's actually no reason why it shouldn't do just the same for closure vtables, afaict.
r? @eddyb
cc @rust-lang/compiler
The wording of RFC #495 enables moves out of slices. Unfortuantely, non-zeroing
moves out of slices introduce a very annoying complication: as slices can
vary in their length, indexes from the start and end may or may not overlap
depending on the slice's exact length, which prevents assigning a particular
drop flag for each individual element.
For example, in the code
```Rust
fn foo<T>(a: Box<[Box<[T]>]>, c: bool) -> T {
match (a, c) {
(box [box [t, ..], ..], true) => t,
(box [.., box [.., t]], false) => t,
_ => panic!()
}
}
```
If the condition is false, we have to drop the first element
of `a`, unless `a` has size 1 in which case we drop all the elements
of it but the last.
If someone comes with a nice way of handling it, we can always re-allow
moves out of slices.
This is a [breaking-change], but it is behind the `slice_patterns` feature
gate and was not allowed until recently.
Be more specific when type parameter shadows primitive type
When a type parameter shadows a primitive type, the error message
was non obvious. For example, given the file `file.rs`:
```rust
trait Parser<T> {
fn parse(text: &str) -> Option<T>;
}
impl<bool> Parser<bool> for bool {
fn parse(text: &str) -> Option<bool> {
Some(true)
}
}
fn main() {
println!("{}", bool::parse("ok").unwrap_or(false));
}
```
The output was:
```bash
% rustc file.rs
error[E0308]: mismatched types
--> file.rs:7:14
|
7 | Some(true)
| ^^^^ expected type parameter, found bool a
|
= note: expected type `bool`
= note: found type `bool`
error: aborting due to previous error
```
We now show extra information about the type:
```bash
% rustc file.rs
error[E0308]: mismatched types
--> file.rs:7:14
|
7 | Some(true)
| ^^^^ expected type parameter, found bool a
|
= note: expected type `bool` (type parameter)
= note: found type `bool` (bool)
error: aborting due to previous error
```
Fixes#35030
When a type parameter shadows a primitive type, the error message
was non obvious. For example, given the file `file.rs`:
```rust
trait Parser<T> {
fn parse(text: &str) -> Option<T>;
}
impl<bool> Parser<bool> for bool {
fn parse(text: &str) -> Option<bool> {
Some(true)
}
}
fn main() {
println!("{}", bool::parse("ok").unwrap_or(false));
}
```
The output was:
```bash
% rustc file.rs
error[E0308]: mismatched types
--> file.rs:7:14
|
7 | Some(true)
| ^^^^ expected type parameter, found bool
|
= note: expected type `bool`
= note: found type `bool`
error: aborting due to previous error
```
We now show extra information about the type:
```bash
% rustc file.rs
error[E0308]: mismatched types
--> file.rs:7:14
|
7 | Some(true)
| ^^^^ expected type parameter, found bool
|
= note: expected type `bool` (type parameter)
= note: found type `bool` (bool)
error: aborting due to previous error
```
Fixes#35030
clear obligations-added flag with nested fulfillcx
This flag is a debugging measure designed to detect cases where we start
a snapshot, create type variables, register obligations involving those
type variables in the fulfillment cx, and then have to unroll the
snapshot, leaving "dangling type variables" behind. HOWEVER, in some
cases the flag is wrong. In particular, we sometimes create a
"mini-fulfilment-cx" in which we enroll obligations. As long as this
fulfillment cx is fully drained before we return, this is not a problem,
as there won't be any escaping obligations in the main cx. So we add a
fn to save/restore the flag.
Fixes#36053.
r? @arielb1
invoke drop glue with a ptr to (data, meta)
This is done by creating a little space on the stack. Hokey, but it's the simplest fix I can see, and I am in "kill regressions" mode right now.
Fixes#35546
r? @eddyb
Use primitive indexing in slice's Index/IndexMut
[T]'s Index implementation is normally not used for indexing, instead
the compiler supplied indexing is used.
Use the compiler supplied version in Index/IndexMut.
This removes an inconsistency:
Compiler supplied bound check failures look like this:
thread 'main' panicked at 'index out of bounds: the len is 3 but the index is 4'
If you convince Rust to use the Index impl for slices, bounds check
failure looks like this instead:
thread 'main' panicked at 'assertion failed: index < self.len()'
The latter is used if you for example use Index generically:
```rust
use std::ops::Index;
fn foo<T: ?Sized>(x: &T) where T: Index<usize> { &x[4]; }
foo(&[1, 2, 3][..])
```
Assign node ids during macro expansion
After this PR,
- The `ExtCtxt` can access `resolve`'s `Resolver` through the trait object `ext::base::Resolver`.
- The `Resolver` trait object can load macros and replaces today's `MacroLoader` trait object.
- The macro expander uses the `Resolver` trait object to resolve macro invocations.
- The macro expander assigns node ids and builds the `Resolver`'s `macros_at_scope` map.
- This is groundwork for merging import resolution and expansion.
- Performance of expansion together with node id assignment improves by ~5%.
**EDIT:** Since Github is reordering the commits, here is `git log`:
- b54e1e3997: Differentiate between monotonic and non-monotonic expansion and only assign node ids during monotonic expansion.
- 78c0039878: Expand generated test harnesses and macro registries.
- f3c2dca353: Remove scope placeholders from the crate root.
- c86c8d41a2: Perform node id assignment and `macros_at_scope` construction during the `InvocationCollector` and `PlaceholderExpander` folds.
- 72a636975f: Move macro resolution into `librustc_resolve`.
- 20b43b2323: Rewrite the unit tests in `ext/expand.rs` as a `compile-fail` test.
- a9821e1658: Refactor `ExtCtxt` to use a `Resolver` instead of a `MacroLoader`.
- 60440b226d: Refactor `noop_fold_stmt_kind` out of `noop_fold_stmt`.
- 50f94f6c95: Avoid needless reexpansions.
r? @nrc
Fix indeterminism in ty::TraitObject representation.
Make sure that projection bounds in `ty::TraitObject` are sorted in a way that is stable across compilation sessions and crate boundaries.
This PR
+ moves `DefPathHashes` up into `librustc` so it can be used there to create a stable sort key for `DefId`s,
+ changes `PolyExistentialProjection::sort_key()` to take advantage of the above,
+ and removes the unused `PolyProjectionPredicate::sort_key()` and `ProjectionTy::sort_key()` methods.
Fixes#36155
Improve shallow `Clone` deriving
`Copy` unions now support `#[derive(Clone)]`.
Less code is generated for `#[derive(Clone, Copy)]`.
+
Unions now support `#[derive(Eq)]`.
Less code is generated for `#[derive(Eq)]`.
---
Example of code reduction:
```
enum E {
A { a: u8, b: u16 },
B { c: [u8; 100] },
}
```
Before:
```
fn clone(&self) -> E {
match (&*self,) {
(&E::A { a: ref __self_0, b: ref __self_1 },) => {
::std::clone::assert_receiver_is_clone(&(*__self_0));
::std::clone::assert_receiver_is_clone(&(*__self_1));
*self
}
(&E::B { c: ref __self_0 },) => {
::std::clone::assert_receiver_is_clone(&(*__self_0));
*self
}
}
}
```
After:
```
fn clone(&self) -> E {
{
let _: ::std::clone::AssertParamIsClone<u8>;
let _: ::std::clone::AssertParamIsClone<u16>;
let _: ::std::clone::AssertParamIsClone<[u8; 100]>;
*self
}
}
```
All the matches are removed, bound assertions are more lightweight.
`let _: Checker<CheckMe>;`, unlike `checker(&check_me);`, doesn't have to be translated by rustc_trans and then inlined by LLVM, it doesn't even exist in MIR, this means faster compilation.
---
Union impls are generated like this:
```
union U {
a: u8,
b: u16,
c: [u8; 100],
}
```
```
fn clone(&self) -> U {
{
let _: ::std::clone::AssertParamIsCopy<Self>;
*self
}
}
```
Fixes https://github.com/rust-lang/rust/issues/36043
cc @durka
r? @alexcrichton
Go back on half the specialization, the part that changed the Zip
struct's fields themselves depending on the types of the iterators.
This means that the Zip iterator will always carry two usize fields,
which are unused. If a whole for loop using a .zip() iterator is
inlined, these are simply removed and have no effect.
The same improvement for Zip of for example slice iterators remain, and
they still optimize well. However, like when the specialization of zip
was merged, the compiler is still very sensistive to the exact context.
For example this code only autovectorizes if the function is used, not
if the code in zip_sum_i32 is inserted inline it was called:
```
fn zip_sum_i32(xs: &[i32], ys: &[i32]) -> i32 {
let mut s = 0;
for (&x, &y) in xs.iter().zip(ys) {
s += x * y;
}
s
}
fn zipdot_i32_default_zip(b: &mut test::Bencher)
{
let xs = vec![1; 1024];
let ys = vec![1; 1024];
b.iter(|| {
zip_sum_i32(&xs, &ys)
})
}
```
Include a test that checks that Zip<T, U> is covariant w.r.t. T and U.
Inherit overflow checks for sum and product
We have previously documented the fact that these will panic on overflow, but I think this behavior is what people actually want/expect. `#[rustc_inherit_overflow_checks]` didn't exist when we discussed these for stabilization.
r? @alexcrichton
Closes#35807
[T]'s Index implementation is normally not used for indexing, instead
the compiler supplied indexing is used.
Use the compiler supplied version in Index/IndexMut.
This removes an inconsistency:
Compiler supplied bound check failures look like this:
thread 'main' panicked at 'index out of bounds: the len is 3 but the index is 4'
If you convince Rust to use the Index impl for slices, bounds check
failure looks like this instead:
thread 'main' panicked at 'assertion failed: index < self.len()'
The latter is used if you for example use Index generically::
use std::ops::Index;
fn foo<T: ?Sized>(x: &T) where T: Index<usize> { &x[4]; }
foo(&[1, 2, 3][..])
