This is an alternative to upgrading the way rvalues are handled in the
borrow check. Making rvalues handled more like lvalues in the borrow
check caused numerous problems related to double mutable borrows and
rvalue scopes. Rather than come up with more borrow check rules to try
to solve these problems, I decided to just forbid pattern bindings after
`@`. This affected fewer than 10 lines of code in the compiler and
libraries.
This breaks code like:
match x {
y @ z => { ... }
}
match a {
b @ Some(c) => { ... }
}
Change this code to use nested `match` or `let` expressions. For
example:
match x {
y => {
let z = y;
...
}
}
match a {
Some(c) => {
let b = Some(c);
...
}
}
Closes#14587.
[breaking-change]
When generating a unique symbol for things like closures or glue_drop,
we call token::gensym() to create a crate-unique Name. Recently, Name
changed its Show impl so it no longer prints as a number. This caused
symbols like glue_drop:1542 to become glue_drop:"glue_drop"(1542), or in
mangled form, glue_drop.$x22glue_drop$x22$LP$1542$RP$.
When generating a unique symbol for things like closures or glue_drop,
we call token::gensym() to create a crate-unique Name. Recently, Name
changed its Show impl so it no longer prints as a number. This caused
symbols like glue_drop:1542 to become glue_drop:"glue_drop"(1542), or in
mangled form, glue_drop.$x22glue_drop$x22$LP$1542$RP$.
Our implementation of ebml has diverged from the standard in order
to better serve the needs of the compiler, so it doesn't make much
sense to call what we have ebml anyore. Furthermore, our implementation
is pretty crufty, and should eventually be rewritten into a format
that better suits the needs of the compiler. This patch factors out
serialize::ebml into librbml, otherwise known as the Really Bad
Markup Language. This is a stopgap library that shouldn't be used
by end users, and will eventually be replaced by something better.
[breaking-change]
Currently, each time a function is monomorphized, all items within that function are translated. This is unnecessary work because the inner items already get translated when the function declaration is visited by `trans_item`. This patch adds a flag to the `FunctionContext` to prevent translation of items during monomorphization.
Remove the ability of the borrow checker to determine that repeated
dereferences of a Box<T> refer to the same memory object. This will
usually require one of two workarounds:
1) The interior of a Box<T> will sometimes need to be moved / borrowed
into a temporary before moving / borrowing individual derived paths.
2) A `ref x` pattern will have to be replaced with a `box ref x`
pattern.
Fixes#16094.
[breaking-change]
the CFG for match statements.
There were two bugs in issue #14684. One was simply that the borrow
check didn't know about the correct CFG for match statements: the
pattern must be a predecessor of the guard. This disallows the bad
behavior if there are bindings in the pattern. But it isn't enough to
prevent the memory safety problem, because of wildcards; thus, this
patch introduces a more restrictive rule, which disallows assignments
and mutable borrows inside guards outright.
I discussed this with Niko and we decided this was the best plan of
action.
This breaks code that performs mutable borrows in pattern guards. Most
commonly, the code looks like this:
impl Foo {
fn f(&mut self, ...) {}
fn g(&mut self, ...) {
match bar {
Baz if self.f(...) => { ... }
_ => { ... }
}
}
}
Change this code to not use a guard. For example:
impl Foo {
fn f(&mut self, ...) {}
fn g(&mut self, ...) {
match bar {
Baz => {
if self.f(...) {
...
} else {
...
}
}
_ => { ... }
}
}
}
Sometimes this can result in code duplication, but often it illustrates
a hidden memory safety problem.
Closes#14684.
[breaking-change]
r? @pnkfelix
Some minor changes to the compiler to expose this information. Very
inconvenient since struct fields aren't an item. Adds (yet another) table to
metadata.
Closes#15739
This commit applies stability attributes to the contents of these modules,
summarized here:
* The `unit` and `bool` modules have become #[unstable] as they are purely meant
for documentation purposes and are candidates for removal.
* The `ty` module has been deprecated, and the inner `Unsafe` type has been
renamed to `UnsafeCell` and moved to the `cell` module. The `marker1` field
has been removed as the compiler now always infers `UnsafeCell` to be
invariant. The `new` method i stable, but the `value` field, `get` and
`unwrap` methods are all unstable.
* The `tuple` module has its name as stable, the naming of the `TupleN` traits
as stable while the methods are all #[unstable]. The other impls in the module
have appropriate stability for the corresponding trait.
* The `arc` module has received the exact same treatment as the `rc` module
previously did.
* The `any` module has its name as stable. The `Any` trait is also stable, with
a new private supertrait which now contains the `get_type_id` method. This is
to make the method a private implementation detail rather than a public-facing
detail.
The two extension traits in the module are marked #[unstable] as they will not
be necessary with DST. The `is` method is #[stable], the as_{mut,ref} methods
have been renamed to downcast_{mut,ref} and are #[unstable].
The extension trait `BoxAny` has been clarified as to why it is unstable as it
will not be necessary with DST.
This is a breaking change because the `marker1` field was removed from the
`UnsafeCell` type. To deal with this change, you can simply delete the field and
only specify the value of the `data` field in static initializers.
[breaking-change]
Currently we don't emit lifetime end markers when translating the
unwinding code. I omitted that when I added the support for lifetime
intrinsics, because I initially made the mistake of just returning true
in clean_on_unwind(). That caused almost all calls to be translated as
invokes, leading to quite awful results.
To correctly emit the lifetime end markers, we must differentiate
between cleanup that requires unwinding and such cleanup that just wants
to emit code during unwinding.
the CFG for match statements.
There were two bugs in issue #14684. One was simply that the borrow
check didn't know about the correct CFG for match statements: the
pattern must be a predecessor of the guard. This disallows the bad
behavior if there are bindings in the pattern. But it isn't enough to
prevent the memory safety problem, because of wildcards; thus, this
patch introduces a more restrictive rule, which disallows assignments
and mutable borrows inside guards outright.
I discussed this with Niko and we decided this was the best plan of
action.
This breaks code that performs mutable borrows in pattern guards. Most
commonly, the code looks like this:
impl Foo {
fn f(&mut self, ...) {}
fn g(&mut self, ...) {
match bar {
Baz if self.f(...) => { ... }
_ => { ... }
}
}
}
Change this code to not use a guard. For example:
impl Foo {
fn f(&mut self, ...) {}
fn g(&mut self, ...) {
match bar {
Baz => {
if self.f(...) {
...
} else {
...
}
}
_ => { ... }
}
}
}
Sometimes this can result in code duplication, but often it illustrates
a hidden memory safety problem.
Closes#14684.
[breaking-change]
Currently we don't emit lifetime end markers when translating the
unwinding code. I omitted that when I added the support for lifetime
intrinsics, because I initially made the mistake of just returning true
in clean_on_unwind(). That caused almost all calls to be translated as
invokes, leading to quite awful results.
To correctly emit the lifetime end markers, we must differentiate
between cleanup that requires unwinding and such cleanup that just wants
to emit code during unwinding.
method calls are involved.
This breaks code like:
impl<T:Copy> Foo for T { ... }
fn take_param<T:Foo>(foo: &T) { ... }
fn main() {
let x = box 3i; // note no `Copy` bound
take_param(&x);
}
Change this code to not contain a type error. For example:
impl<T:Copy> Foo for T { ... }
fn take_param<T:Foo>(foo: &T) { ... }
fn main() {
let x = 3i; // satisfies `Copy` bound
take_param(&x);
}
Closes#15860.
[breaking-change]
r? @alexcrichton
method calls are involved.
This breaks code like:
impl<T:Copy> Foo for T { ... }
fn take_param<T:Foo>(foo: &T) { ... }
fn main() {
let x = box 3i; // note no `Copy` bound
take_param(&x);
}
Change this code to not contain a type error. For example:
impl<T:Copy> Foo for T { ... }
fn take_param<T:Foo>(foo: &T) { ... }
fn main() {
let x = 3i; // satisfies `Copy` bound
take_param(&x);
}
Closes#15860.
[breaking-change]
librustc: Stop desugaring `for` expressions and translate them directly.
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]
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]
This is done entirely in the libraries for functions up to 16 arguments.
A macro is used so that more arguments can be easily added if we need.
Note that I had to adjust the overloaded call algorithm to not try
calling the overloaded call operator if the callee is a built-in
function type, to prevent loops.
Closes#15448.
This eliminates the last vestige of the `~` syntax.
Instead of `~self`, write `self: Box<TypeOfSelf>`; instead of `mut
~self`, write `mut self: Box<TypeOfSelf>`, replacing `TypeOfSelf` with
the self-type parameter as specified in the implementation.
Closes#13885.
[breaking-change]
The allocas used in match expression currently don't get good lifetime
markers, in fact they only get lifetime start markers, because their
lifetimes don't match to cleanup scopes.
While the bindings themselves are bog standard and just need a matching
pair of start and end markers, they might need them twice, once for a
guard clause and once for the match body.
The __llmatch alloca OTOH needs a single lifetime start marker, but
when there's a guard clause, it needs two end markers, because its
lifetime ends either when the guard doesn't match or after the match
body.
With these intrinsics in place, LLVM can now, for example, optimize
code like this:
````rust
enum E {
A1(int),
A2(int),
A3(int),
A4(int),
}
pub fn variants(x: E) {
match x {
A1(m) => bar(&m),
A2(m) => bar(&m),
A3(m) => bar(&m),
A4(m) => bar(&m),
}
}
````
To a single call to bar, using only a single stack slot. It still fails
to eliminate some of checks.
````gas
.Ltmp5:
.cfi_def_cfa_offset 16
movb (%rdi), %al
testb %al, %al
je .LBB3_5
movzbl %al, %eax
cmpl $1, %eax
je .LBB3_5
cmpl $2, %eax
.LBB3_5:
movq 8(%rdi), %rax
movq %rax, (%rsp)
leaq (%rsp), %rdi
callq _ZN3bar20hcb7a0d8be8e17e37daaE@PLT
popq %rax
retq
````
Lifetime intrinsics help to reduce stack usage, because LLVM can apply
stack coloring to reuse the stack slots of dead allocas for new ones.
For example these functions now both use the same amount of stack, while
previous `bar()` used five times as much as `foo()`:
````rust
fn foo() {
println("{}", 5);
}
fn bar() {
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
}
````
On top of that, LLVM can also optimize out certain operations when it
knows that memory is dead after a certain point. For example, it can
sometimes remove the zeroing used to cancel the drop glue. This is
possible when the glue drop itself was already removed because the
zeroing dominated the drop glue call. For example in:
````rust
pub fn bar(x: (Box<int>, int)) -> (Box<int>, int) {
x
}
````
With optimizations, this currently results in:
````llvm
define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 {
"_ZN29_$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$17h88cf42702e5a322aE.exit":
%2 = bitcast { i64*, i64 }* %1 to i8*
%3 = bitcast { i64*, i64 }* %0 to i8*
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false)
tail call void @llvm.lifetime.end(i64 16, i8* %2)
ret void
}
````
Fixes#15665
Lifetime intrinsics help to reduce stack usage, because LLVM can apply
stack coloring to reuse the stack slots of dead allocas for new ones.
For example these functions now both use the same amount of stack, while
previous `bar()` used five times as much as `foo()`:
````rust
fn foo() {
println("{}", 5);
}
fn bar() {
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
}
````
On top of that, LLVM can also optimize out certain operations when it
knows that memory is dead after a certain point. For example, it can
sometimes remove the zeroing used to cancel the drop glue. This is
possible when the glue drop itself was already removed because the
zeroing dominated the drop glue call. For example in:
````rust
pub fn bar(x: (Box<int>, int)) -> (Box<int>, int) {
x
}
````
With optimizations, this currently results in:
````llvm
define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 {
"_ZN29_$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
`call_visit_glue` is only ever called from trans_intrinsic, and the
block won't be unreachable there. Also, the comment doesn't make sense
anymore. When the code was introduced in 38fee9526a the function was
also responsible for the cleanup glue, which is no longer the case.
While we're at it, also fixed the debug message to output the right
function name.
This implements RFC 39. Omitted lifetimes in return values will now be
inferred to more useful defaults, and an error is reported if a lifetime
in a return type is omitted and one of the two lifetime elision rules
does not specify what it should be.
This primarily breaks two uncommon code patterns. The first is this:
unsafe fn get_foo_out_of_thin_air() -> &Foo {
...
}
This should be changed to:
unsafe fn get_foo_out_of_thin_air() -> &'static Foo {
...
}
The second pattern that needs to be changed is this:
enum MaybeBorrowed<'a> {
Borrowed(&'a str),
Owned(String),
}
fn foo() -> MaybeBorrowed {
Owned(format!("hello world"))
}
Change code like this to:
enum MaybeBorrowed<'a> {
Borrowed(&'a str),
Owned(String),
}
fn foo() -> MaybeBorrowed<'static> {
Owned(format!("hello world"))
}
Closes#15552.
[breaking-change]
r? @nick29581
This is accomplished by rewriting static expressions into equivalent patterns.
This way, patterns referencing static variables can both participate
in exhaustiveness analysis as well as be compiled down into the appropriate
branch of the decision trees that match expressions are codegened to.
Fixes#6533.
Fixes#13626.
Fixes#13731.
Fixes#14576.
Fixes#15393.
This implements RFC 39. Omitted lifetimes in return values will now be
inferred to more useful defaults, and an error is reported if a lifetime
in a return type is omitted and one of the two lifetime elision rules
does not specify what it should be.
This primarily breaks two uncommon code patterns. The first is this:
unsafe fn get_foo_out_of_thin_air() -> &Foo {
...
}
This should be changed to:
unsafe fn get_foo_out_of_thin_air() -> &'static Foo {
...
}
The second pattern that needs to be changed is this:
enum MaybeBorrowed<'a> {
Borrowed(&'a str),
Owned(String),
}
fn foo() -> MaybeBorrowed {
Owned(format!("hello world"))
}
Change code like this to:
enum MaybeBorrowed<'a> {
Borrowed(&'a str),
Owned(String),
}
fn foo() -> MaybeBorrowed<'static> {
Owned(format!("hello world"))
}
Closes#15552.
[breaking-change]
This is accomplished by rewriting static expressions into equivalent patterns.
This way, patterns referencing static variables can both participate
in exhaustiveness analysis as well as be compiled down into the appropriate
branch of the decision trees that match expressions are codegened to.
Fixes#6533.
Fixes#13626.
Fixes#13731.
Fixes#14576.
Fixes#15393.
Removed `index_to_bitset` field and `_frozen` methods.
Drive-by: Added some missing docs on the `each_bit` method.
Drive-by: Put in a regular pattern: when calling `compute_id_range`, ensure `words_per_id > 0` by either asserting it or checking and returning early. (The prior code did the latter in a few cases where necessary, but debugging is much aided by the asserts.)
Fix#15019.
`call_visit_glue` is only ever called from trans_intrinsic, and the
block won't be unreachable there. Also, the comment doesn't make sense
anymore. When the code was introduced in 38fee9526a the function was
also responsible for the cleanup glue, which is no longer the case.
While we're at it, also fixed the debug message to output the right
function name.
Importing from types was disallowed in #6462. Flag was set for paths whether it is a module or a type. Type flag was set when impl was seen. The problem is, for cross-crate situations, when reexport is involved, it is possible that impl is seen too late because metadata is loaded lazily.
Fix#15664.
This should fix issue #15541. It would be good to have an test case for this would also be nice but I haven't had the time to write one. The change is very small though and it doesn't break anything in the existing test suite, so I guess we can add it without test for now.
except where trait objects are involved.
Part of issue #15349, though I'm leaving it open for trait objects.
Cross borrowing for trait objects remains because it is needed until we
have DST.
This will break code like:
fn foo(x: &int) { ... }
let a = box 3i;
foo(a);
Change this code to:
fn foo(x: &int) { ... }
let a = box 3i;
foo(&*a);
[breaking-change]
This makes two changes to region inference: (1) it allows region
inference to relate early-bound regions; and (2) it allows regions to be
related before variance runs. The former is needed because there is no
relation between the two regions before region substitution happens,
while the latter is needed because type collection has to run before
variance. We assume that, before variance is inferred, that lifetimes
are invariant. This is a conservative overapproximation.
This relates to #13885. This does not remove `~self` from the language
yet, however.
[breaking-change]
Disabling the redzone is required in x86-64's kernel mode to avoid interrupts trashing the stack.
I'm not sure if decl_fn is the right place to tag all functions with noredzone. It might have interactions with external functions when linking with bitcode built without -C no-redzone although I see no reason to do that.
I'm not sure how to write a test inspecting the bitcode output for noredzone attributes on all functions either.
This patch applies the excellent suggestion of @pnkfelix to group the helper methods for method field access into a Trait, making the code much more readable, and much more similar to the way it was before.
Closes#15525
The important bit of this are the changes from line 445 in mem_categorization.rs. Most of the other changes are about adding an Implicit PointerKind, and this is only necessary for getting a decent error message :-s An alternative would have been to add an implciti/explicit flag to cat_deref, which could be mostly ignored and so would mean much fewer changes. However, the implicit state would only be valid if the PointerKind was BorrowedPtr, so it felt like it ought to be another kind of PointerKind. I still don't know which is the better design.
To verify that a type can satisfy Send
`check_struct_safe_for_destructor` attempts to construct a new `ty::t`
an empty substitution list.
Previously the function would verify that the function has no type
parameters before attempting this. Unfortunately this check would not
catch functions with only regions parameters. In this case, the type
would eventually find its way to the substition engine which would
attempt to perform a substitution on the region parameters. As the
constructed substitution list is empty, this would fail, leading to a
compiler crash.
We fix this by verifying that types have both no type and region
parameters.
Previously this was an Option::unwrap() which failed for me.
Unfortunately I've since inadvertently worked around the bug and have
been unable to reproduce it. With this patch hopefully the next person
to encounter this will be in a slightly better position to debug it.
Per @pnkfelix 's suggestion, using a trait to make these
field accesses more readable (and vastly more similar
to the original code.
oops fix new ast_map fix
Use one or more of the following `-Z` flag options to tell the
graphviz renderer to include the corresponding dataflow sets (after
the iterative constraint propagation reaches a fixed-point solution):
* `-Z flowgraph-print-loans` : loans computed via middle::borrowck
* `-Z flowgraph-print-moves` : moves computed via middle::borrowck::move_data
* `-Z flowgraph-print-assigns` : assignments, via middle::borrowck::move_data
* `-Z flowgraph-print-all` : all of the available sets are included.
Fix#15016.
Use one or more of the following `-Z` flag options to tell the
graphviz renderer to include the corresponding dataflow sets (after
the iterative constraint propagation reaches a fixed-point solution):
* `-Z flowgraph-print-loans` : loans computed via middle::borrowck
* `-Z flowgraph-print-moves` : moves computed via middle::borrowck::move_data
* `-Z flowgraph-print-assigns` : assignments, via middle::borrowck::move_data
* `-Z flowgraph-print-all` : all of the available sets are included.
Fix#15016.
----
This also adds a module, `syntax::ast_map::blocks`, that captures a
common abstraction shared amongst code blocks and procedure-like
things. As part of this, moved `ast_map.rs` to subdir
`ast_map/mod.rs`, to follow our directory layout conventions.
(incorporated review feedback from huon, acrichto.)
This change propagates to many locations, but because of the
Macro Exterminator (or, more properly, the invariant that it
protects), macro invocations can't occur downstream of expansion.
This means that in librustc and librustdoc, extracting the
desired field can simply assume that it can't be a macro
invocation. Functions in ast_util abstract over this check.
This is a continuation of @brson's work from https://github.com/rust-lang/rust/pull/12144.
This implements the minimal scaffolding that allows mapping diagnostic messages to alpha-numeric codes, which could improve the searchability of errors. In addition, there's a new compiler option, `--explain {code}` which takes an error code and prints out a somewhat detailed explanation of the error. Example:
```rust
fn f(x: Option<bool>) {
match x {
Some(true) | Some(false) => (),
None => (),
Some(true) => ()
}
}
```
```shell
[~/rust]$ ./build/x86_64-apple-darwin/stage2/bin/rustc ./diagnostics.rs --crate-type dylib
diagnostics.rs:5:3: 5:13 error: unreachable pattern [E0001] (pass `--explain E0001` to see a detailed explanation)
diagnostics.rs:5 Some(true) => ()
^~~~~~~~~~
error: aborting due to previous error
[~/rust]$ ./build/x86_64-apple-darwin/stage2/bin/rustc --explain E0001
This error suggests that the expression arm corresponding to the noted pattern
will never be reached as for all possible values of the expression being matched,
one of the preceeding patterns will match.
This means that perhaps some of the preceeding patterns are too general, this
one is too specific or the ordering is incorrect.
```
I've refrained from migrating many errors to actually use the new macros as it can be done in an incremental fashion but if we're happy with the approach, it'd be good to do all of them sooner rather than later.
Originally, I was going to make libdiagnostics a separate crate but that's posing some interesting challenges with semi-circular dependencies. In particular, librustc would have a plugin-phase dependency on libdiagnostics, which itself depends on librustc. Per my conversation with @alexcrichton, it seems like the snapshotting process would also have to change. So for now the relevant modules from libdiagnostics are included using `#[path = ...] mod`.
This removes a bunch of token types. Tokens now store the original, unaltered
numeric literal (that is still checked for correctness), which is parsed into
an actual number later, as needed, when creating the AST.
This can change how syntax extensions work, but otherwise poses no visible
changes.
[breaking-change]
formerly, the self identifier was being discarded during parsing, which
stymies hygiene. The best fix here seems to be to attach a self identifier
to ExplicitSelf_, a change that rippled through the rest of the compiler,
but without any obvious damage.
This updates https://github.com/rust-lang/rust/pull/15075.
Rename `ToStr::to_str` to `ToString::to_string`. The naive renaming ends up with two `to_string` functions defined on strings in the prelude (the other defined via `collections::str::StrAllocating`). To remedy this I removed `StrAllocating::to_string`, making all conversions from `&str` to `String` go through `Show`. This has a measurable impact on the speed of this conversion, but the sense I get from others is that it's best to go ahead and unify `to_string` and address performance for all `to_string` conversions in `core::fmt`. `String::from_str(...)` still works as a manual fast-path.
Note that the patch was done with a script, and ended up renaming a number of other `*_to_str` functions, particularly inside of rustc. All the ones I saw looked correct, and I didn't notice any additional API breakage.
Closes#15046.
closes#13367
[breaking-change] Use `Sized?` to indicate a dynamically sized type parameter or trait (used to be `type`). E.g.,
```
trait Tr for Sized? {}
fn foo<Sized? X: Share>(x: X) {}
```
This will break code that looks like:
struct Foo {
...
}
mod Foo {
...
}
Change this code to:
struct Foo {
...
}
impl Foo {
...
}
Or rename the module.
Closes#15205.
[breaking-change]
r? @nick29581
Extend the null ptr optimization to work with slices, closures, procs, & trait objects by using the internal pointers as the discriminant.
This decreases the size of `Option<&[int]>` (and similar) by one word.
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]
r? @nick29581
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]
This will break code that looks like:
struct Foo {
...
}
mod Foo {
...
}
Change this code to:
struct Foo {
...
}
impl Foo {
...
}
Or rename the module.
Closes#15205.
[breaking-change]
LLVM doesn't handle i1 value in allocas/memory very well and skips a number of optimizations if it hits it. So we have to do the same thing that Clang does, using i1 for SSA values, but storing i8 in memory.
Fixes#15203.
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
This is an implementation of [RFC 35](https://github.com/rust-lang/rfcs/blob/master/active/0035-remove-crate-id.md).
The summary for this PR is the same as that of the RFC, with one addendum:
* Removes the `#[crate_id]` attribute and knowledge of versions from rustc.
* Added a `#[crate_name]` attribute similar to the old `#[crate_id]` attribute
* Output filenames no longer have versions or hashes
* Symbols no longer have versions (they still have hashes)
* A new flag, `--extern`, is used to override searching for external crates
* A new flag, `-C metadata=foo`, used when hashing symbols
* [added] An old flag, `--crate-name`, was re purposed to specify the crate name from the command line.
I tried to maintain backwards compatibility wherever possible (with warnings being printed). If I missed anywhere, however, please let me know!
[breaking-change]
Closes#14468Closes#14469Closes#14470Closes#14471
This comit implements a new flag, --extern, which is used to specify where a
crate is located. The purpose of this flag is to bypass the normal crate
loading/matching of the compiler to point it directly at the right file.
This flag takes the form `--extern foo=bar` where `foo` is the name of a crate
and `bar` is the location at which to find the crate. Multiple `--extern`
directives are allowed with the same crate name to specify the rlib/dylib pair
for a crate. It is invalid to specify more than one rlib or more than one dylib,
and it's required that the crates are valid rust crates.
I have also added some extensive documentation to metadata::loader about how
crate loading should work.
RFC: 0035-remove-crate-id
This commit removes all support in the compiler for the #[crate_id] attribute
and all of its derivative infrastructure. A list of the functionality removed is:
* The #[crate_id] attribute no longer exists
* There is no longer the concept of a version of a crate
* Version numbers are no longer appended to symbol names
* The --crate-id command line option has been removed
To migrate forward, rename #[crate_id] to #[crate_name] and only the name of the
crate itself should be mentioned. The version/path of the old crate id should be
removed.
For a transitionary state, the #[crate_id] attribute is still accepted if
the #[crate_name] is not present, but it is warned about if it is the only
identifier present.
RFC: 0035-remove-crate-id
[breaking-change]
In my informal measurements, this brings the peak memory usage when
building librustc from 1662M down to 1502M. Since 1662 - 1502 = 160,
this may not recover the entirety of the observed memory regression
(250M) from PR #14604. (However, according to my local measurements,
the regression when building librustc was more like 209M, so perhaps
this will still recover the lions share of the lost memory.)
In my informal measurements, this brings the peak memory usage when
building librustc from 1662M down to 1502M. Since 1662 - 1502 = 160,
this may not recover the entirety of the observed memory regression
(250M) from PR #14604. (However, according to my local measurements,
the regression when building librustc was more like 209M, so perhaps
this will still recover the lions share of the lost memory.)
This basically meant changing the interface so that no borrowed `&Vec`
is exposed, by hiding `fn get_vec` and `fn get_mut_vec` and revising
`fn all_vecs`.
Instead, clients should use one of the other methods; `get_slice`,
`pop`, `truncate`, `replace`, `push_all`, or `is_empty_in`, which
should work for any case currently used in rustc.
This pull request adds hygiene for 3 kinds of argument bindings:
- arguments to item fns,
- arguments to `ExprFnBlock`s, and
- arguments to `ExprProc`s
It also adds a bunch of unit tests, fixes a few macro uses to be non-capturing, and has a few cleanup items.
local `make check` succeeds.
parameters.
This can break code that mistakenly used type parameters in place of
`Self`. For example, this will break:
trait Foo {
fn bar<X>(u: X) -> Self {
u
}
}
Change this code to not contain a type error. For example:
trait Foo {
fn bar<X>(_: X) -> Self {
self
}
}
Closes#15172.
[breaking-change]
r? @alexcrichton
Closes#15276 (Guide: if)
Closes#15280 (std::os - Add join_paths, make setenv non-utf8 capable)
Closes#15314 (Guide: functions)
Closes#15327 (Simplify PatIdent to contain an Ident rather than a Path)
Closes#15340 (Guide: add mutable binding section)
Closes#15342 (Fix ICE with nested macro_rules!-style macros)
Closes#15350 (Remove duplicated slash in install script path)
Closes#15351 (correct a few spelling mistakes in the tutorial)
Closes#15352 (librustc: Have the kind checker check sub-bounds in trait casts.)
Closes#15359 (Fix spelling errors.)
Closes#15361 (Rename set_broadast() to set_broadcast().)
Closes#15366 (Simplify creating a parser from a token tree)
Closes#15367 (Add examples for StrVector methods)
Closes#15372 (Vec::grow should use reserve_additional, Vec::reserve should check against capacity)
Closes#15373 (Fix minor issues in the documentation of libtime.)
This can break code that looked like:
struct S<T> {
val: T,
}
trait Gettable<T> {
...
}
impl<T: Copy> Gettable<T> for S<T> {
...
}
let t: Box<S<String>> = box S {
val: "one".to_string(),
};
let a = t as Box<Gettable<String>>;
// ^ note no `Copy` bound
Change this code to:
impl<T> Gettable<T> for S<T> {
// ^ remove `Copy` bound
...
}
Closes#14061.
[breaking-change]
Rationale: for what appear to be historical reasons only, the PatIdent contains
a Path rather than an Ident. This means that there are many places in the code
where an ident is artificially promoted to a path, and---much more problematically---
a bunch of elements from a path are simply thrown away, which seems like an invitation
to some really nasty bugs.
This commit replaces the Path in a PatIdent with a SpannedIdent, which just contains an ident
and a span.
This was causing lots of ICEs in cargo. I sadly wasn't ever able to reduce the
test case down, but I presume that's because it has to do with node id
collisions which are pretty difficult to turn up...
So far, type names generated for debuginfo where a bit sketchy. It was not clearly defined when a name should be fully qualified and when not, if region parameters should be shown or not, and other things like that.
This commit makes the debuginfo module responsible for creating type names instead of using `ppaux::ty_to_str()` and brings type names (as they show up in the DWARF information) in line with GCC and Clang:
* The name of the type being described is unqualified. It's path is defined by its position in the namespace hierarchy.
* Type arguments are always fully qualified, no matter if they would actually be in scope at the type definition location.
Care is also taken to make type names consistent across crate boundaries. That is, the code now tries make the type name the same, regardless if the type is in the local crate or reconstructed from metadata. Otherwise LLVM will complain about violating the one-definition-rule when using link-time-optimization.
This commit also removes all source location information from type descriptions because these cannot be reconstructed for types instantiated from metadata. Again, with LTO enabled, this can lead to two versions of the debuginfo type description, one with and one without source location information, which then triggers the LLVM ODR assertion.
Fortunately, source location information about types is rarely used, so this has little impact. Once source location information is preserved in metadata (#1972) it can also be re-enabled for type descriptions.
`RUSTFLAGS=-g make check` no works again for me locally, including the LTO test cases (note that I've taken care of #15156 by reverting the change in LLVM that @luqmana identified as the culprit for that issue).
with the corresponding trait parameter bounds.
This is a version of the patch in PR #12611 by Florian Hahn, modified to
address Niko's feedback.
It does not address the issue of duplicate type parameter bounds, nor
does it address the issue of implementation-defined methods that contain
*fewer* bounds than the trait, because Niko's review indicates that this
should not be necessary (and indeed I believe it is not). A test has
been added to ensure that this works.
This will break code like:
trait Foo {
fn bar<T:Baz>();
}
impl Foo for Boo {
fn bar<T:Baz + Quux>() { ... }
// ^~~~ ERROR
}
This will be rejected because the implementation requires *more* bounds
than the trait. It can be fixed by either adding the missing bound to
the trait:
trait Foo {
fn bar<T:Baz + Quux>();
// ^~~~
}
impl Foo for Boo {
fn bar<T:Baz + Quux>() { ... } // OK
}
Or by removing the bound from the impl:
trait Foo {
fn bar<T:Baz>();
}
impl Foo for Boo {
fn bar<T:Baz>() { ... } // OK
// ^ remove Quux
}
This patch imports the relevant tests from #2687, as well as the test
case in #5886, which is fixed as well by this patch.
Closes#2687.
Closes#5886.
[breaking-change]
r? @pnkfelix
parameters.
This can break code that mistakenly used type parameters in place of
`Self`. For example, this will break:
trait Foo {
fn bar<X>(u: X) -> Self {
u
}
}
Change this code to not contain a type error. For example:
trait Foo {
fn bar<X>(_: X) -> Self {
self
}
}
Closes#15172.
[breaking-change]
Slice patterns are different from the rest in that a single slice pattern
does not have a distinct constructor if it contains a variable-length subslice
pattern. For example, the pattern [a, b, ..tail] can match a slice of length 2, 3, 4
and so on.
As a result, the decision tree for exhaustiveness and redundancy analysis should
explore each of those constructors separately to determine if the pattern could be useful
when specialized for any of them.
So far, type names generated for debuginfo where a bit sketchy. It was not clearly defined when a name should be fully qualified and when not, if region parameters should be shown or not, and other things like that.
This commit makes the debuginfo module responsible for creating type names instead of using ppaux::ty_to_str() and brings type names, as they show up in the DWARF information, in line with GCC and Clang:
* The name of the type being described is unqualified. It's path is defined by its position in the namespace hierarchy.
* Type arguments are always fully qualified, no matter if they would actually be in scope at the type definition location.
Care is also taken to reliably make type names consistent across crate boundaries. That is, the code now tries make the type name the same, regardless if the type is in the local crate or reconstructed from metadata. Otherwise LLVM will complain about violating the one-definition-rule when using link-time-optimization.
This commit also removes all source location information from type descriptions because these cannot be reconstructed for types instantiated from metadata. Again, with LTO enabled, this can lead to two versions of the debuginfo type description, one with and one without source location information, which then triggers the LLVM ODR assertion.
Fortunately, source location information about types is rarely used, so this has little impact. Once source location information is preserved in metadata (#1972) it can also be reenabled for type descriptions.
Being able to index into the bytes of a string encourages
poor UTF-8 hygiene. To get a view of `&[u8]` from either
a `String` or `&str` slice, use the `as_bytes()` method.
Closes#12710.
[breaking-change]
If the diffstat is any indication this shouldn't have a huge impact but it will have some. Most changes in the `str` and `path` module. A lot of the existing usages were in tests where ascii is expected. There are a number of other legit uses where the characters are known to be ascii.
with the corresponding trait parameter bounds.
This is a version of the patch in PR #12611 by Florian Hahn, modified to
address Niko's feedback.
It does not address the issue of duplicate type parameter bounds, nor
does it address the issue of implementation-defined methods that contain
*fewer* bounds than the trait, because Niko's review indicates that this
should not be necessary (and indeed I believe it is not). A test has
been added to ensure that this works.
This will break code like:
trait Foo {
fn bar<T:Baz>();
}
impl Foo for Boo {
fn bar<T:Baz + Quux>() { ... }
// ^~~~ ERROR
}
This will be rejected because the implementation requires *more* bounds
than the trait. It can be fixed by either adding the missing bound to
the trait:
trait Foo {
fn bar<T:Baz + Quux>();
// ^~~~
}
impl Foo for Boo {
fn bar<T:Baz + Quux>() { ... } // OK
}
Or by removing the bound from the impl:
trait Foo {
fn bar<T:Baz>();
}
impl Foo for Boo {
fn bar<T:Baz>() { ... } // OK
// ^ remove Quux
}
This patch imports the relevant tests from #2687, as well as the test
case in #5886, which is fixed as well by this patch.
Closes#2687.
Closes#5886.
[breaking-change]
Being able to index into the bytes of a string encourages
poor UTF-8 hygiene. To get a view of `&[u8]` from either
a `String` or `&str` slice, use the `as_bytes()` method.
Closes#12710.
[breaking-change]
This commit hooks rustdoc into the stability index infrastructure in two
ways:
1. It looks up stability levels via the index, rather than by manual
attributes.
2. It adds stability level information throughout rustdoc output, rather
than just at the top header. In particular, a stability color (with
mouseover text) appears next to essentially every item that appears
in rustdoc's HTML output.
Along the way, the stability index code has been lightly refactored.
floating point numbers for real.
This will break code that looks like:
let mut x = 0;
while ... {
x += 1;
}
println!("{}", x);
Change that code to:
let mut x = 0i;
while ... {
x += 1;
}
println!("{}", x);
Closes#15201.
[breaking-change]
This change registers new snapshots, allowing `*T` to be removed from the language. This is a large breaking change, and it is recommended that if compiler errors are seen that any FFI calls are audited to determine whether they should be actually taking `*mut T`.
This can break code that looked like:
impl Foo for Box<Any> {
fn f(&self) { ... }
}
let x: Box<Any + Send> = ...;
x.f();
Change such code to:
impl Foo for Box<Any> {
fn f(&self) { ... }
}
let x: Box<Any> = ...;
x.f();
That is, upcast before calling methods.
This is a conservative solution to #5781. A more proper treatment (see
the xfail'd `trait-contravariant-self.rs`) would take variance into
account. This change fixes the soundness hole.
Some library changes had to be made to make this work. In particular,
`Box<Any>` is no longer showable, and only `Box<Any+Send>` is showable.
Eventually, this restriction can be lifted; for now, it does not prove
too onerous, because `Any` is only used for propagating the result of
task failure.
This patch also adds a test for the variance inference work in #12828,
which accidentally landed as part of DST.
Closes#5781.
[breaking-change]
Since procs do not have lifetime bounds, we must do this to maintain
safety.
This can break code that incorrectly captured references in procedure
types. Change such code to not do this, perhaps with a trait object
instead.
A better solution would be to add higher-rank lifetime support to procs.
However, this would be a lot of work for a feature we want to remove in
favor of unboxed closures. The corresponding "real fix" is #15067.
Closes#14036.
[breaking-change]
This will break code like:
fn f(x: &mut int) {}
let mut a = box 1i;
f(a);
Change it to:
fn f(x: &mut int) {}
let mut a = box 1i;
f(&mut *a);
RFC 33; issue #10504.
[breaking-change]
r? @brson
vector-reference-to-unsafe-pointer-to-element cast if the type to be
casted to is not fully specified.
This is a conservative change to fix the user-visible symptoms of the
issue. A more flexible treatment would delay cast checks to after
function typechecking.
This can break code that did:
let x: *u8 = &([0, 0]) as *_;
Change this code to:
let x: *u8 = &([0, 0]) as *u8;
Closes#14893.
[breaking-change]
This will break code like:
fn f(x: &mut int) {}
let mut a = box 1i;
f(a);
Change it to:
fn f(x: &mut int) {}
let mut a = box 1i;
f(&mut *a);
RFC 33; issue #10504.
[breaking-change]
The f128 type has very little support in the compiler and the feature is
basically unusable today. Supporting half-baked features in the compiler can be
detrimental to the long-term development of the compiler, and hence this feature
is being removed.
bindings.
This will break code that incorrectly did things like:
fn f(a @ box b: Box<String>) {}
Fix such code to not rely on undefined behavior.
Closes#12534.
[breaking-change]
This breaks a fair amount of code. The typical patterns are:
* `for _ in range(0, 10)`: change to `for _ in range(0u, 10)`;
* `println!("{}", 3)`: change to `println!("{}", 3i)`;
* `[1, 2, 3].len()`: change to `[1i, 2, 3].len()`.
RFC #30. Closes#6023.
[breaking-change]
The f128 type has very little support in the compiler and the feature is
basically unusable today. Supporting half-baked features in the compiler can be
detrimental to the long-term development of the compiler, and hence this feature
is being removed.
Also change some code formatting.
lint::builtin becomes a sibling of lint::context in order to ensure that lints
implemented there use the same public API as lint plugins.
We're going to have more modules under lint, and the paths get unwieldy. We
also plan to have lints run at multiple points in the compilation pipeline.
The aim of these changes is not working out a generic bi-endianness architectures support but to allow people develop for little endian MIPS machines (issue #7190).
This is just a cleanup of the code. Doesn't really change anything deep about the way we operate. This is a prelude to implementing a good solution for one-way matching for #5527.
r? @pnkfelix (we were just crawling about this code, after all)
In other words, Late-bound regions that occur non-free should be
skipped.
Fix#10846 (specifically the ICE, not the weakness in the current type inference).
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.
First, we have to fix some bugs concerning the handling of
attributes in foreign function declarations and calls. These
are required because the i1 type needs the ZExt attribute when
used as a function parameter or return type.
Then we have to update LLVM to get a bugfix without which LLVM
sometimes generates broken code when using i1.
And then, finally, we can switch bools over to i1.
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.
To fix#8106, we need an LLVM version that contains r211082 aka 0dee6756
which fixes a bug that blocks that issue.
There have been some tiny API changes in LLVM, and cmpxchg changed its
return type. The i1 part of the new return type is only interesting when
using the new weak cmpxchg, which we don't do.
When calling a foreign function, some arguments and/or return value
attributes are required to conform to the foreign ABI. Currently those
attributes are only added to the declaration of foreign functions. With
direct calls, this is no problem, because LLVM can see that those
attributes apply to the call. But with an indirect call, LLVM cannot do
that and the attribute is missing.
To fix that, we have to add those attribute to the calls to foreign
functions as well.
This also allows to remove the special handling of the SRet attribute,
which is ABI-dependent and will be set via the `attr` field of the
return type's `ArgType`.
The ArgType type gives us a generic way to specify an attribute for a
type to ensure ABI conformance for foreign functions. But the code that
actually sets the argument attributes in the function declaration
only sets the attribute for the return type when the type is indirect.
Since LLVMAddAttribute() doesn't allow to set attributes on the return
type, we have to use LLVMAddFunctionAttribute() instead.
This didn't cause problems yet, because currently only some indirect
types require attributes to be set.
This commit makes several changes to the stability index infrastructure:
* Stability levels are now inherited lexically, i.e., each item's
stability level becomes the default for any nested items.
* The computed stability level for an item is stored as part of the
metadata. When using an item from an external crate, this data is
looked up and cached.
* The stability lint works from the computed stability level, rather
than manual stability attribute annotations. However, the lint still
checks only a limited set of item uses (e.g., it does not check every
component of a path on import). This will be addressed in a later PR,
as part of issue #8962.
* The stability lint only applies to items originating from external
crates, since the stability index is intended as a promise to
downstream crates.
* The "experimental" lint is now _allow_ by default. This is because
almost all existing crates have been marked "experimental", pending
library stabilization. With inheritance in place, this would generate
a massive explosion of warnings for every Rust program.
The lint should be changed back to deny-by-default after library
stabilization is complete.
* The "deprecated" lint still warns by default.
The net result: we can begin tracking stability index for the standard
libraries as we stabilize, without impacting most clients.
Closes#13540.
