We can now handle name resolution errors and get past type checking (if we're a bit lucky). This is the first step towards doing code completion for partial programs (we need error recovery in the parser and early access to save-analysis).
This PR reverts #29543 and instead implements proper support for "=*m" and "+*m" indirect output operands. This provides a framework on top of which support for plain memory operands ("m", "=m" and "+m") can be implemented.
This also fixes the liveness analysis pass not handling read/write operands correctly.
r? @nagisa
I'm going to need the `ConstVal` -> `ValueRef` translation to start removing trans/consts piece by piece. If you need anything implemented in the translation, feel free to assign an issue to me.
Fixes#26403
This adjusts the pointer, if needed, to the correct alignment by using the alignment information in the vtable.
Handling zero might not be necessary, as it shouldn't actually occur. I've left it as it's own commit so it can be removed fairly easily if people don't think it's worth doing. The way it's handled though means that there shouldn't be much impact on performance.
Instead of `ast::Ident`, bindings, paths and labels in HIR now keep a new structure called `hir::Ident` containing mtwt-renamed `name` and the original not-renamed `unhygienic_name`. `name` is supposed to be used by default, `unhygienic_name` is rarely used.
This is not ideal, but better than the status quo for two reasons:
- MTWT tables can be cleared immediately after lowering to HIR
- This is less bug-prone, because it is impossible now to forget applying `mtwt::resolve` to a name. It is still possible to use `name` instead of `unhygienic_name` by mistake, but `unhygienic_name`s are used only in few very special circumstances, so it shouldn't be a problem.
Besides name resolution `unhygienic_name` is used in some lints and debuginfo. `unhygienic_name` can be very well approximated by "reverse renaming" `token::intern(name.as_str())` or even plain string `name.as_str()`, except that it would break gensyms like `iter` in desugared `for` loops. This approximation is likely good enough for lints and debuginfo, but not for name resolution, unfortunately (see https://github.com/rust-lang/rust/issues/27639), so `unhygienic_name` has to be kept.
cc https://github.com/rust-lang/rust/issues/29782
r? @nrc
The presence of the drop flag caused the offset calculation to be
incorrect, leading to the pointer being incorrect. This has been fixed
by calculating the offset based on the field index (and not assuming
that the field is always the last one).
However, I've also stopped the drop flag from being added to the end of
unsized structs to begin with. Since it's not actually accessed for
unsized structs, and isn't actually where we would say it is, this made
more sense.
I've measured the time/memory consumption before and after - the difference is lost in statistical noise, so it's mostly a code simplification.
Sizes of `enum`s are not affected.
r? @nrc
I wonder if AST/HIR visitors could run faster if `P`s are systematically removed (except for cases where they control `enum` sizes). Theoretically they should.
Remaining unnecessary `P`s can't be easily removed because many folders accept `P<X>`s as arguments, but these folders can be converted to accept `X`s instead without loss of efficiency.
When I have a mood for some mindless refactoring again, I'll probably try to convert the folders, remove remaining `P`s and measure again.
DST fields, being of an unknown type, are not automatically aligned
properly, so a pointer to the field needs to be aligned using the
information in the vtable.
Fixes#26403 and a number of other DST-related bugs discovered while
implementing this.
This commit is the standard API stabilization commit for the 1.6 release cycle.
The list of issues and APIs below have all been through their cycle-long FCP and
the libs team decisions are listed below
Stabilized APIs
* `Read::read_exact`
* `ErrorKind::UnexpectedEof` (renamed from `UnexpectedEOF`)
* libcore -- this was a bit of a nuanced stabilization, the crate itself is now
marked as `#[stable]` and the methods appearing via traits for primitives like
`char` and `str` are now also marked as stable. Note that the extension traits
themeselves are marked as unstable as they're imported via the prelude. The
`try!` macro was also moved from the standard library into libcore to have the
same interface. Otherwise the functions all have copied stability from the
standard library now.
* The `#![no_std]` attribute
* `fs::DirBuilder`
* `fs::DirBuilder::new`
* `fs::DirBuilder::recursive`
* `fs::DirBuilder::create`
* `os::unix::fs::DirBuilderExt`
* `os::unix::fs::DirBuilderExt::mode`
* `vec::Drain`
* `vec::Vec::drain`
* `string::Drain`
* `string::String::drain`
* `vec_deque::Drain`
* `vec_deque::VecDeque::drain`
* `collections::hash_map::Drain`
* `collections::hash_map::HashMap::drain`
* `collections::hash_set::Drain`
* `collections::hash_set::HashSet::drain`
* `collections::binary_heap::Drain`
* `collections::binary_heap::BinaryHeap::drain`
* `Vec::extend_from_slice` (renamed from `push_all`)
* `Mutex::get_mut`
* `Mutex::into_inner`
* `RwLock::get_mut`
* `RwLock::into_inner`
* `Iterator::min_by_key` (renamed from `min_by`)
* `Iterator::max_by_key` (renamed from `max_by`)
Deprecated APIs
* `ErrorKind::UnexpectedEOF` (renamed to `UnexpectedEof`)
* `OsString::from_bytes`
* `OsStr::to_cstring`
* `OsStr::to_bytes`
* `fs::walk_dir` and `fs::WalkDir`
* `path::Components::peek`
* `slice::bytes::MutableByteVector`
* `slice::bytes::copy_memory`
* `Vec::push_all` (renamed to `extend_from_slice`)
* `Duration::span`
* `IpAddr`
* `SocketAddr::ip`
* `Read::tee`
* `io::Tee`
* `Write::broadcast`
* `io::Broadcast`
* `Iterator::min_by` (renamed to `min_by_key`)
* `Iterator::max_by` (renamed to `max_by_key`)
* `net::lookup_addr`
New APIs (still unstable)
* `<[T]>::sort_by_key` (added to mirror `min_by_key`)
Closes#27585Closes#27704Closes#27707Closes#27710Closes#27711Closes#27727Closes#27740Closes#27744Closes#27799Closes#27801
cc #27801 (doesn't close as `Chars` is still unstable)
Closes#28968
This PR allows the constant evaluation of index operations on constant arrays and repeat expressions. This allows index expressions to appear in the expression path of the length expression of a repeat expression or an array type.
An example is
```rust
const ARR: [usize; 5] = [1, 2, 3, 4, 5];
const ARR2: [usize; ARR[1]] = [42, 99];
```
In most other locations llvm's const evaluator figures it out already. This is not specific to index expressions and could be remedied in the future.
Fixes https://github.com/rust-lang/rust/issues/28692
Fixes https://github.com/rust-lang/rust/issues/28992
Fixes some other similar issues (see the tests)
[breaking-change], needs crater run (cc @brson or @alexcrichton )
The pattern with parens `UnitVariant(..)` for unit variants seems to be popular in rustc (see the second commit), but mostly used by one person (@nikomatsakis), according to git blame. If it causes breakage on crates.io I'll add an exceptional case for it.
Since fat pointers do not qualify as structural types, they got copied
using load_ty and store_ty, which means that we load an FCA and use
extractvalue to get the components of the fat pointer. This breaks
certain optimizations in LLVM.
Found via apasel422/ref_count#13
encounter each module. This is somewhat different than how it used to
work; it should ensure a more equitable distribution of work than
before. The reason is that, before, when we rotated, we would rotate
before we had seen the full contents of the current module. So e.g. if
we have `mod a { mod b { .. } .. }`, then we rotate when we encounter
`b`, but we haven't processed the remainder of `a` yet. Unclear if this
makes any difference in practice, but it seemed suboptimal. Also, this
structure (with an outer walk over modules) is closer to what we will
want for an incremental setting.
noteworthy because trans got mildly simpler, since it doesn't have to
ensure that we walk the contents of all things just to find all the
hidden items.
Introduce a `SwitchInt` and restructure pattern matching to collect integers and characters into one master switch. This is aimed at #29227, but is not a complete fix. Whereas before we generated an if-else-if chain and, at least on my machine, just failed to compile, we now spend ~9sec compiling `rustc_abuse`. AFAICT this is basically just due to a need for more micro-optimization of the matching process: perf shows a fair amount of time just spent iterating over the candidate list. Still, it seemed worth opening a PR with this step alone, since it's a big step forward.
Function arguments that are small aggregates get passed as integer types
instead. To correctly handle that, we need to use store_ty instead of
plain Store.
Note: for now, this change only affects `-windows-gnu` builds.
So why was this `libgcc` dylib dependency needed in the first place?
The stack unwinder needs to know about locations of unwind tables of all the modules loaded in the current process. The easiest portable way of achieving this is to have each module register itself with the unwinder when loaded into the process. All modules compiled by GCC do this by calling the __register_frame_info() in their startup code (that's `crtbegin.o` and `crtend.o`, which are automatically linked into any gcc output).
Another important piece is that there should be only one copy of the unwinder (and thus unwind tables registry) in the process. This pretty much means that the unwinder must be in a shared library (unless everything is statically linked).
Now, Rust compiler tries very hard to make sure that any given Rust crate appears in the final output just once. So if we link the unwinder statically to one of Rust's crates, everything should be fine.
Unfortunately, GCC startup objects are built under assumption that `libgcc` is the one true place for the unwind info registry, so I couldn't find any better way than to replace them. So out go `crtbegin`/`crtend`, in come `rsbegin`/`rsend`!
A side benefit of this change is that rustc is now more in control of the command line that goes to the linker, so we could stop using `gcc` as the linker driver and just invoke `ld` directly.
Motivation:
- It is not actually a pattern
- It is not actually needed, except for...
Drawback:
- Slice patterns like `[a, _.., b]` are pretty-printed as `[a, .., b]`. Great loss :(
plugin-[breaking-change], as always
Similarly to the simd intrinsics. I believe this is a better solution than #29288, and I could implement it as well for overflowing_add/sub/mul. Also rename from udiv/sdiv to div, and same for rem.
when evaluating a recursive type, the `type_of` of the interior could be
still in progress, so trying to get its size would cause an ICE.
Fixes#19001
r? @eddyb
this has the funky side-effect of also allowing constant evaluation of function calls to functions that are not `const fn` as long as `check_const` didn't mark that function `NOT_CONST`
It's still not possible to call a normal function from a `const fn`, but let statements' initialization value can get const evaluated (this caused the fallout in the overflowing tests)
we can now do this:
```rust
const fn add(x: usize, y: usize) -> usize { x + y }
const ARR: [i32; add(1, 2)] = [5, 6, 7];
```
also added a test for destructuring in const fn args
```rust
const fn i((a, b): (u32, u32)) -> u32 { a + b } //~ ERROR: E0022
```
This is a **[breaking change]**, since it turns some runtime panics into compile-time errors. This statement is true for ANY improvement to the const evaluator.
I could have added a check for explicit recursion, as irregular types
tend to cause selection errors, but I am not sufficiently sure that
cannot be bypassed.
Fixes#22919Fixes#25639Fixes#26548
PR for #28157. At the moment, `rustc` emits a warning when a bare semicolon is encountered (could also be a fail, but I think this is a backwards incompatible change).
Also I am not sure where the best place for a test for that warning would be. Seems run-pass tests do not check warnings.
Rather than injecting a local `_Unwind_Resume` into the current translation unit,
just replace `resume` instruction with a direct call the the `eh_unwind_resume` lang item.
This is likely to be more robust in the face of future LLVM changes, and also allows us to delegate
work back to libgcc's `_Unwind_Resume`.
This PR turns statically known erroneous code (e.g. numeric overflow) into a warning and continues normal code-generation to emit the same code that would have been generated without `check_const` detecting that the result can be computed at compile-time.
<del>It's not done yet, as I don't know how to properly emit a lint from trans. I can't seem to extract the real lint level of the item the erroneous expression is in.</del> It's an unconditional warning now.
r? @pnkfelix
cc @nikomatsakis
* [RFC 1229 text](https://github.com/rust-lang/rfcs/blob/master/text/1229-compile-time-asserts.md)
* RFC PR: rust-lang/rfcs#1229
* tracking issue: https://github.com/rust-lang/rust/issues/28238
Fixes#28676.
There doesn't seem to be a good way to add a test for this, but I tested the repro in #28676 and confirmed it now yields the correct result.
This is needed as item types are allowed to be unnormalized.
Fixes an ICE that occurs when foreign function signatures contained
an associated type.
Fixes#28983
This is needed as item types are allowed to be unnormalized.
Fixes an ICE that occurs when foreign function signatures contained
an associated type.
Fixes#28983
For enum variants, the default alignment for a specific variant might be
lower than the alignment of the enum type itself. In such cases we, for
example, generate memcpy calls with an alignment that's higher than the
alignment of the constant we copy from.
To avoid that, we need to explicitly set the required alignment on
constants.
Fixes#28912.
For enum variants, the default alignment for a specific variant might be
lower than the alignment of the enum type itself. In such cases we, for
example, generate memcpy calls with an alignment that's higher than the
alignment of the constant we copy from.
To avoid that, we need to explicitly set the required alignment on
constants.
Fixes#28912.
This turned up as part of #3170. When constructing an `undef` value to
return in the error case, we were trying to get the element type of the
Rust-level value being indexed instead of the underlying array; when
indexing a slice, that's not an array and the LLVM assertion failure
reflects this.
The regression test is a lightly altered copy of `const-array-oob.rs`.
libcore.rlib reduced from 19121 kiB to 15934 kiB - 20% win.
The librustc encoded AST is 9013500 bytes long - for the record, librustc consists of about 2254126 characters. Might be worth looking at.
r? @eddyb
paths, and construct paths for all definitions. Also, stop rewriting
DefIds for closures, and instead just load the closure data from
the original def-id, which may be in another crate.
this simplifies the code while reducing the size of libcore.rlib by
3.3 MiB (~1M of which is bloat a separate patch of mine removes
too), while reducing rustc memory usage on small crates by 18MiB.
This also simplifies the code considerably.
I have measured a small, but possibly insignificant, bootstrap performance improvement, and the memory savings grow to about 30M for larger crates (but that is still less as a percentage).
r? @eddyb
this simplifies the code while reducing the size of libcore.rlib by
3.3 MiB (~1M of which is bloat a separate patch of mine removes
too), while reducing rustc memory usage on small crates by 18MiB.
This also simplifies the code considerably.
By putting an "unreachable" instruction into the default arm of a switch
instruction we can let LLVM know that the match is exhaustive, allowing
for better optimizations.
For example, this match:
```rust
pub enum Enum {
One,
Two,
Three,
}
impl Enum {
pub fn get_disc(self) -> u8 {
match self {
Enum::One => 0,
Enum::Two => 1,
Enum::Three => 2,
}
}
}
```
Currently compiles to this on x86_64:
```asm
.cfi_startproc
movzbl %dil, %ecx
cmpl $1, %ecx
setne %al
testb %cl, %cl
je .LBB0_2
incb %al
movb %al, %dil
.LBB0_2:
movb %dil, %al
retq
.Lfunc_end0:
```
But with this change we get:
```asm
.cfi_startproc
movb %dil, %al
retq
.Lfunc_end0:
```
As discussed in the referenced issues, this PR makes rustc emit `__imp_<symbol>` stubs for all public static data to ensure smooth linking in on `-windows-msvc` targets.
Resolves#26591, cc #27438
This PR removes random remaining `Ident`s outside of libsyntax and performs general cleanup
In particular, interfaces of `Name` and `Ident` are tidied up, `Name`s and `Ident`s being small `Copy` aggregates are always passed to functions by value, and `Ident`s are never used as keys in maps, because `Ident` comparisons are tricky.
Although this PR closes https://github.com/rust-lang/rust/issues/6993 there's still work related to it:
- `Name` can be made `NonZero` to compress numerous `Option<Name>`s and `Option<Ident>`s but it requires const unsafe functions.
- Implementation of `PartialEq` on `Ident` should be eliminated and replaced with explicit hygienic, non-hygienic or member-wise comparisons.
- Finally, large parts of AST can potentially be converted to `Name`s in the same way as HIR to clearly separate identifiers used in hygienic and non-hygienic contexts.
r? @nrc
By putting an "unreachable" instruction into the default arm of a switch
instruction we can let LLVM know that the match is exhaustive, allowing
for better optimizations.
For example, this match:
```rust
pub enum Enum {
One,
Two,
Three,
}
impl Enum {
pub fn get_disc(self) -> u8 {
match self {
Enum::One => 0,
Enum::Two => 1,
Enum::Three => 2,
}
}
}
```
Currently compiles to this on x86_64:
```asm
.cfi_startproc
movzbl %dil, %ecx
cmpl $1, %ecx
setne %al
testb %cl, %cl
je .LBB0_2
incb %al
movb %al, %dil
.LBB0_2:
movb %dil, %al
retq
.Lfunc_end0:
```
But with this change we get:
```asm
.cfi_startproc
movb %dil, %al
retq
.Lfunc_end0:
```
Make sure Name, SyntaxContext and Ident are passed by value
Make sure Idents don't serve as keys (or parts of keys) in maps, Ident comparison is not well defined
Unwinding across an FFI boundary is undefined behaviour, so we can mark
all external function as nounwind. The obvious exception are those
functions that actually perform the unwinding.
This also involved adding `[TYPE;N]` syntax and aggregate indexing
support to the generator script: it's the only way to be able to have a
parameterised intrinsic that returns an aggregate, since one can't refer
to previous elements of the current aggregate (and that was harder to
implement).
This adds a new Python script (compatible with 2.7 and 3.x) that will consume some JSON files that define a platform's intrinsics. It can output a file that defines the intrinsics in the compiler, or an `extern` block that will import them.
The complexity of the generator is to be DRY: platforms (especially ARM and AArch64) have a lot of repetition with their intrinsics, for different versions with different types, so being able to write it once is nice.
This adds support for flattened intrinsics, which are called in Rust
with tuples but in LLVM without them (e.g. `foo((a, b))` becomes `foo(a,
b)`). Unflattened ones could be supported, but are not yet.
This is necessary to reflect the ARM APIs accurately, since some
functions explicitly take an unsigned parameter and a signed one, of the
same integer shape, so the no-duplicates check will fail unless we
distinguish.
The major change here is in the tiny commit at the end and makes it so that we no longer emit lifetime intrinsics for allocas for function arguments. They are live for the whole function anyway, so the intrinsics add no value. This makes the resulting IR more clear, and reduces the peak memory usage and LLVM times by about 1-4%, depending on the crate.
The remaining changes are just preparatory cleanups and fixes for missing lifetime intrinsics.
Function arguments are live for the whole function scope, so adding
lifetime intrinsics around them adds no value. The same is true for drop
hint allocas and everything else that goes directly through
lvalue_scratch_datum. So the easiest fix is to emit lifetime intrinsics
only for lvalue datums that are created in to_lvalue_datum_in_scope().
The reduces peak memory usage and LLVM times by about 1-4%, depending on
the crate.
Combining them seemed like a good idea at the time, but turns out that
handling lifetimes separately makes it somewhat easier to handle cases
where we don't want the intrinsics, and let's you see more easily where
the start/end pairs are.
The issues that the comments referred to were fixed before the PR even
landed but we never got around to remove the hack of skipping the
lifetime start.
The functions is useful for all kinds of fat pointers, but get_len()
just feels so wrong for trait object fat pointers. Let's use get_meta()
because that's rather neutral.
This increases regionck performance greatly - type-checking on
librustc decreased from 9.1s to 8.1s. Because of Amdahl's law,
total performance is improved only by about 1.5% (LLVM wizards,
this is your opportunity to shine!).
before:
576.91user 4.26system 7:42.36elapsed 125%CPU (0avgtext+0avgdata 1142192maxresident)k
after:
566.50user 4.84system 7:36.84elapsed 125%CPU (0avgtext+0avgdata 1124304maxresident)k
I am somewhat worried really need to find out why we have this Red Queen's
Race going on here. Originally I suspected it may be a problem from RFC1214's
warnings, but it seems to be an effect from other changes.
However, the increase seems to be mostly in LLVM's time, so I guess
it's the LLVM wizards' problem.
We're currently possibly introducing an unneeded temporary, make use of
InsertValue which is said to kick us off of FastISel and we generate
loads/stores of first class aggregates, which is bad as well. Let's not
do all these things.
We're currently possibly introducing an unneeded temporary, make use of
InsertValue which is said to kick us off of FastISel and we generate
loads/stores of first class aggregates, which is bad as well. Let's not
do all these things.
Currently `f32 % f32` will generate a link error on 32-bit MSVC because LLVM
will lower the operation to a call to the nonexistent function `fmodf`. Work
around in this in the backend by lowering to a call to `fmod` instead with
necessary extension/truncation between floats/doubles.
Closes#27859
Currently `f32 % f32` will generate a link error on 32-bit MSVC because LLVM
will lower the operation to a call to the nonexistent function `fmodf`. Work
around in this in the backend by lowering to a call to `fmod` instead with
necessary extension/truncation between floats/doubles.
Closes#27859
This is purposely separate to the "rust-intrinsic" ABI, because these
intrinsics are theoretically going to become stable, and should be fine
to be independent of the compiler/language internals since they're
intimately to the platform.
When using a generic enum type that was defined in an external crate,
our debuginfo currently claims that the concrete type (e.g. Option<i32>)
was defined in the current crate, where it was first used.
This means that if there are multiple crates that all use, for example,
Option<i32> values, they'll have conflicting debuginfo, each crate
claiming to have defined that type. This doesn't cause problems in
regular builds, but with LTO enabled, LLVM complains because it tries to
merge the debuginfo for those types and sees the ODR violations.
Since I couldn't find a way to get the file info for the external crate
that actually defined the enum, I'm working around the issue by using
"<unknown>" as the file for enum types. We'll want to re-visit and fix
this later, but this at least this fixes the ICE. And with the file
being unknown instead of wrong, the debuginfo isn't really worse than
before either.
Fixes#26447
Just a little code cleanup I was doing as part of another refactoring (which may turn out not to be needed). The main thrust of this is to cleanup the interface to `tydecode.rs` to be less ridiculously repetitive. I also purged the generic "def-id conversion" parameter in favor of a trait object, just to reduce code duplication a bit and make the signatures a bit less messy. I measured the bootstrapping time to build stage2 with these changes, it was identical. (But it'd be easy enough to restore the unboxed closure if we wanted it.)
If you had previously tried to get the ValueRef associated with an
intrinsic that hadn't been described in
`trans::context::declare_intrinsic()`, the compile would panic with
an empty message.
Now we print out details about the error in the panic message.
