Change to resolve and update compiler and libs for uses.
[breaking-change]
Enum variants are now in both the value and type namespaces. This means that
if you have a variant with the same name as a type in scope in a module, you
will get a name clash and thus an error. The solution is to either rename the
type or the variant.
`Box<[T]>` is created by allocating `Box<[T, ..n]>` and coercing it so
this code path is never used. It's also broken because it clamps the
capacity of the memory allocations to 4 elements and that's incompatible
with sized deallocation. This dates back to when `~[T]` was a growable
vector type implemented as:
*{ { tydesc, ref_count, prev, next }, { length, capacity, data[] } }
Since even empty vectors had to allocate, it started off the capacity of
all vectors at 4 as a heuristic. It's not possible to grow `Box<[T]>`
and there is no need for a memory allocation when it's empty, so it
would be a terrible heuristic today even if it worked.
The pointer in the slice must not be null, because enum representations
make that assumption. The `exchange_malloc` function returns a non-null
sentinel for the zero size case, and it must not be passed to the
`exchange_free` lang item.
Since the length is always equal to the true capacity, a branch on the
length is enough for most types. Slices of zero size types are
statically special cased to never attempt deallocation. This is the same
implementation as `Vec<T>`.
Closes#14395
This allows code to access the fields of tuples and tuple structs:
let x = (1i, 2i);
assert_eq!(x.1, 2);
struct Point(int, int);
let origin = Point(0, 0);
assert_eq!(origin.0, 0);
assert_eq!(origin.1, 0);
The pointer in the slice must not be null, because enum representations
make that assumption. The `exchange_malloc` function returns a non-null
sentinel for the zero size case, and it must not be passed to the
`exchange_free` lang item.
Since the length is always equal to the true capacity, a branch on the
length is enough for most types. Slices of zero size types are
statically special cased to never attempt deallocation. This is the same
implementation as `Vec<T>`.
Closes#14395
Rotate between compilation units while translating. The "worker threads"
commit added support for multiple compilation units, but only translated into
one, leaving the rest empty. With this commit, `trans` rotates between various
compilation units while translating, using a simple stragtegy: upon entering a
module, switch to translating into whichever compilation unit currently
contains the fewest LLVM instructions.
Most of the actual changes here involve getting symbol linkage right, so that
items translated into different compilation units will link together properly
at the end.
For example `let _x: &Trait = &*(box Foo as Box<Trait>);`. There was a bug where no cleanup would be scheduled by the deref.
No test because cleanup-auto-borrow-obj.rs is a test for this once we remove trait cross-borrowing (done on another branch).
[breaking-change]
1. The internal layout for traits has changed from (vtable, data) to (data, vtable). If you were relying on this in unsafe transmutes, you might get some very weird and apparently unrelated errors. You should not be doing this! Prefer not to do this at all, but if you must, you should use raw::TraitObject rather than hardcoding rustc's internal representation into your code.
2. The minimal type of reference-to-vec-literals (e.g., `&[1, 2, 3]`) is now a fixed size vec (e.g., `&[int, ..3]`) where it used to be an unsized vec (e.g., `&[int]`). If you want the unszied type, you must explicitly give the type (e.g., `let x: &[_] = &[1, 2, 3]`). Note in particular where multiple blocks must have the same type (e.g., if and else clauses, vec elements), the compiler will not coerce to the unsized type without a hint. E.g., `[&[1], &[1, 2]]` used to be a valid expression of type '[&[int]]'. It no longer type checks since the first element now has type `&[int, ..1]` and the second has type &[int, ..2]` which are incompatible.
3. The type of blocks (including functions) must be coercible to the expected type (used to be a subtype). Mostly this makes things more flexible and not less (in particular, in the case of coercing function bodies to the return type). However, in some rare cases, this is less flexible. TBH, I'm not exactly sure of the exact effects. I think the change causes us to resolve inferred type variables slightly earlier which might make us slightly more restrictive. Possibly it only affects blocks with unreachable code. E.g., `if ... { fail!(); "Hello" }` used to type check, it no longer does. The fix is to add a semicolon after the string.
This patch primarily does two things: (1) it prevents lifetimes from
leaking out of unboxed closures; (2) it allows unboxed closure type
notation, call notation, and construction notation to construct closures
matching any of the three traits.
This breaks code that looked like:
let mut f;
{
let x = &5i;
f = |&mut:| *x + 10;
}
Change this code to avoid having a reference escape. For example:
{
let x = &5i;
let mut f; // <-- move here to avoid dangling reference
f = |&mut:| *x + 10;
}
I believe this is enough to consider unboxed closures essentially
implemented. Further issues (for example, higher-rank lifetimes) should
be filed as followups.
Closes#14449.
[breaking-change]
by-reference upvars.
This partially implements RFC 38. A snapshot will be needed to turn this
on, because stage0 cannot yet parse the keyword.
Part of #12381.
This makes edge cases in which the `Iterator` trait was not in scope
and/or `Option` or its variants were not in scope work properly.
This breaks code that looks like:
struct MyStruct { ... }
impl MyStruct {
fn next(&mut self) -> Option<int> { ... }
}
for x in MyStruct { ... } { ... }
Change ad-hoc `next` methods like the above to implementations of the
`Iterator` trait. For example:
impl Iterator<int> for MyStruct {
fn next(&mut self) -> Option<int> { ... }
}
Closes#15392.
[breaking-change]
Lifetime intrinsics help to reduce stack usage, because LLVM can apply
stack coloring to reuse the stack slots of dead allocas for new ones.
For example these functions now both use the same amount of stack, while
previous `bar()` used five times as much as `foo()`:
````rust
fn foo() {
println("{}", 5);
}
fn bar() {
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
}
````
On top of that, LLVM can also optimize out certain operations when it
knows that memory is dead after a certain point. For example, it can
sometimes remove the zeroing used to cancel the drop glue. This is
possible when the glue drop itself was already removed because the
zeroing dominated the drop glue call. For example in:
````rust
pub fn bar(x: (Box<int>, int)) -> (Box<int>, int) {
x
}
````
With optimizations, this currently results in:
````llvm
define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 {
"_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit":
%2 = bitcast { i64*, i64 }* %1 to i8*
%3 = bitcast { i64*, i64 }* %0 to i8*
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false)
tail call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 16, i32 8, i1 false)
ret void
}
````
But with lifetime intrinsics we get:
````llvm
define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 {
"_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit":
%2 = bitcast { i64*, i64 }* %1 to i8*
%3 = bitcast { i64*, i64 }* %0 to i8*
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false)
tail call void @llvm.lifetime.end(i64 16, i8* %2)
ret void
}
````
Fixes#15665
This will break code that used the old `Index` trait. Change this code
to use the new `Index` traits. For reference, here are their signatures:
pub trait Index<Index,Result> {
fn index<'a>(&'a self, index: &Index) -> &'a Result;
}
pub trait IndexMut<Index,Result> {
fn index_mut<'a>(&'a mut self, index: &Index) -> &'a mut Result;
}
Closes#6515.
[breaking-change]
LLVM doesn't really like types with a bit-width that isn't a multiple of
8 and disable various optimizations if it encounters such types used
with loads/stores. OTOH, booleans must be represented as i1 when used as
SSA values. To get the best results, we must use i1 for SSA values, and
i8 when storing the value to memory.
By using range asserts on loads, LLVM can eliminate the required
zero-extend and truncate operations.
Fixes#15203
We currently compiled bools to i8 values, because there was a bug in
LLVM that sometimes caused miscompilations when using i1 in, for
example, structs.
Using i8 means a lot of unnecessary zero-extend and truncate operations
though, since we have to convert the value from and to i1 when using for
example icmp or br instructions. Besides the unnecessary overhead caused
by this, it also sometimes made LLVM miss some optimizations.
Fixes#8106.
