This is a bit of an interesting upgrade to LLVM. Upstream LLVM has started using C++11 features, so they require a C++11 compiler to build. I've updated all the bots to have a C++11 compiler, and they appear to be building LLVM successfully:
* Linux bots - I added gcc/g++ 4.7 (good enough)
* Android bots - same as the linux ones
* Mac bots - I installed the most recent command line tools for Lion which gives us clang 3.2, but LLVM wouldn't build unless it was explicitly asked to link to `libc++` instead of `libstdc++`. This involved tweaking `mklldeps.py` and the `configure` script to get things to work out
* Windows bots - mingw-w64 has gcc 4.8.1 which is sufficient for building LLVM (hurray!)
* BSD bots - I updated FreeBSD to 10.0 which brought with it a relevant version of clang.
The largest fallout I've seen so far is that the test suite doesn't work at all on FreeBSD 10. We've already stopped gating on FreeBSD due to #13427 (we used to be on freebsd 9), so I don't think this puts us in too bad of a situation. I will continue to attempt to fix FreeBSD and the breakage on there.
The LLVM update brings with it all of the recently upstreamed LLVM patches. We only have one local patch now which is just an optimization, and isn't required to use upstream LLVM. I want to maintain compatibility with LLVM 3.3 and 3.4 while we can, and this upgrade is keeping us up to date with the 3.5 release. Once 3.5 is release we will in theory no longer require a bundled LLVM.
When reporting "consider removing this semicolon" hint message, the
offending semicolon may come from macro call site instead of macro
itself. Using the more appropriate span makes the hint more helpful.
Closes#13428.
This version of `is_to_be_inferred` double-checks the result from
`inferred_map` by querying the `named_region_map` and `ast_map` and
then asserts that the `inferred_map` state is consistent with its own
findings. (See issue 13261 for further discussion of the approaches).
This comes with a number of fixes to be compatible with upstream LLVM:
* Previously all monomorphizations of "mem::size_of()" would receive the same
symbol. In the past LLVM would silently rename duplicated symbols, but it
appears to now be dropping the duplicate symbols and functions now. The symbol
names of monomorphized functions are now no longer solely based on the type of
the function, but rather the type and the unique hash for the
monomorphization.
* Split stacks are no longer a global feature controlled by a flag in LLVM.
Instead, they are opt-in on a per-function basis through a function attribute.
The rust #[no_split_stack] attribute will disable this, otherwise all
functions have #[split_stack] attached to them.
* The compare and swap instruction now takes two atomic orderings, one for the
successful case and one for the failure case. LLVM internally has an
implementation of calculating the appropriate failure ordering given a
particular success ordering (previously only a success ordering was
specified), and I copied that into the intrinsic translation so the failure
ordering isn't supplied on a source level for now.
* Minor tweaks to LLVM's API in terms of debuginfo, naming, c++11 conventions,
etc.
Part of this required added an override of `fold_type_method` in the
Folder for Ctx impl; it follows the same pattern as `fold_method`.
Also, as a drive-by fix, I moved all of the calls to `folder.new_id`
in syntax::fold's no-op default traversal to really be the first
statement in each function.
* This is to uphold the invariant that `folder.new_id` is always
called first (an unfortunate requirement of the current `ast_map`
code), an invariant that we seemingly were breaking in e.g. the
previous `noop_fold_block`.
* Now it should be easier to see when adding new code that this
invariant must be upheld.
* (note that the breakage in `noop_fold_block` may not have mattered
so much previously, since the only thing that blocks can bind are
lifetimes, which I am only adding support for now.)
Before adding a variance constrant for a given early-bound param,
check if it was meant to be inferred.
To support the above, added `fn is_to_be_inferred` to
`variance::ConstraintContext`.
When instantiating trait default methods for certain implementation,
`typeck` correctly combined type parameters from trait bound with those
from method bound, but didn't do so for lifetime parameters. Applies
the same logic to lifetime parameters.
Closes#13204
This commit changes the way move errors are reported when some value is
captured by a PatIdent. First, we collect all of the "cannot move out
of" errors before reporting them, and those errors with the same "move
source" are reported together. If the move is caused by a PatIdent (that
binds by value), we add a note indicating where it is and suggest the
user to put `ref` if they don't want the value to move. This makes the
"cannot move out of" error in match expression nicer (though the extra
note may not feel that helpful in other places :P). For example, with
the following code snippet,
```rust
enum Foo {
Foo1(~u32, ~u32),
Foo2(~u32),
Foo3,
}
fn main() {
let f = &Foo1(~1u32, ~2u32);
match *f {
Foo1(num1, num2) => (),
Foo2(num) => (),
Foo3 => ()
}
}
```
Errors before the change:
```rust
test.rs:10:9: 10:25 error: cannot move out of dereference of `&`-pointer
test.rs:10 Foo1(num1, num2) => (),
^~~~~~~~~~~~~~~~
test.rs:10:9: 10:25 error: cannot move out of dereference of `&`-pointer
test.rs:10 Foo1(num1, num2) => (),
^~~~~~~~~~~~~~~~
test.rs:11:9: 11:18 error: cannot move out of dereference of `&`-pointer
test.rs:11 Foo2(num) => (),
^~~~~~~~~
```
After:
```rust
test.rs:9:11: 9:13 error: cannot move out of dereference of `&`-pointer
test.rs:9 match *f {
^~
test.rs:10:14: 10:18 note: attempting to move value to here (to prevent the move, use `ref num1` or `ref mut num1` to capture value by reference)
test.rs:10 Foo1(num1, num2) => (),
^~~~
test.rs:10:20: 10:24 note: and here (use `ref num2` or `ref mut num2`)
test.rs:10 Foo1(num1, num2) => (),
^~~~
test.rs:11:14: 11:17 note: and here (use `ref num` or `ref mut num`)
test.rs:11 Foo2(num) => (),
^~~
```
Close#8064
When instantiating trait default methods for certain implementation,
`typeck` correctly combined type parameters from trait bound with those
from method bound, but didn't do so for lifetime parameters. Applies
the same logic to lifetime parameters.
Closes#13204
This removes the `priv` keyword from the language and removes private enum
variants as a result. The remaining use cases of private enum variants were all
updated to be a struct with one private field that is a private enum.
RFC: 0006-remove-priv
Closes#13535
Previously, if statements of the form "Foo;" or "let _ = Foo;" were encountered
where Foo had a destructor, the destructors were not run. This changes
the relevant locations in trans to check for ty::type_needs_drop and invokes
trans_to_lvalue instead of trans_into.
Closes#4734Closes#6892
Exposing ctpop, ctlz, cttz and bswap as taking signed i8/i16/... is just
exposing the internal LLVM names pointlessly (LLVM doesn't have "signed
integers" or "unsigned integers", it just has sized integer types
with (un)signed *operations*).
These operations are semantically working with raw bytes, which the
unsigned types model better.
Fixes#13507.
I haven't familiarized myself with this part of the rust compiler, so hopefully there are no mistakes (despite the simplicity of the commit). It is also 5am.
This includes a change to the way lifetime names are generated. Say we
figure that `[#0, 'a, 'b]` have to be the same lifetimes, then instead
of just generating a new lifetime `'c` like before to replace them, we
would reuse `'a`. This is done so that when the lifetime name comes
from an impl, we don't give something that's completely off, and we
don't have to do much work to figure out where the name came from. For
example, for the following code snippet:
```rust
struct Baz<'x> {
bar: &'x int
}
impl<'x> Baz<'x> {
fn baz1(&self) -> &int {
self.bar
}
}
```
`[#1, 'x]` (where `#1` is BrAnon(1) and refers to lifetime of `&int`)
have to be marked the same lifetime. With the old method, we would
generate a new lifetime `'a` and suggest `fn baz1(&self) -> &'a int`
or `fn baz1<'a>(&self) -> &'a int`, both of which are wrong.
A mismatched type with more type parameters than the expected one causes
`typeck` looking up out of the bound of type parameter vector, which
leads to ICE.
Closes#13466
The current error message is misleading, it asks users to add `#[feature(..)]` which ends up being treated as an outer attribute, which then has no error unless `attribute_usage` lint is enforced. The code will still fail and the user might not understand why.
This patch fixes issue #13186.
When generating constant expression for enum, it is possible that
alignment of expression may be not equal to alignment of type. In that
case space after last struct field must be padded to match size of value
and size of struct. This commit adds that padding.
See detailed explanation in src/test/run-pass/trans-tag-static-padding.rs
A mismatched type with more type parameters than the expected one causes
`typeck` looking up out of the bound of type parameter vector, which
leads to ICE.
Closes#13466
Cleans up some remnants of the old mutability system and only allows vector/trait mutability in `VstoreSlice` (`&mut [T]`) and `RegionTraitStore` (`&mut Trait`).
libstd: Implement `StrBuf`, a new string buffer type like `Vec`, and port all code over to use it.
Rebased & tests-fixed version of https://github.com/mozilla/rust/pull/13269
Previously, a private use statement would shadow a public use statement, all of
a sudden publicly exporting the privately used item. The correct behavior here
is to only shadow the use for the module in question, but for now it just
reverts the entire name to private so the pub use doesn't have much effect.
The behavior isn't exactly what we want, but this no longer has backwards
compatibility hazards.
Previously resolve was checking the "import resolution" for whether an import
had succeeded or not, but this was the same structure filled in by a previous
import if a name is shadowed. Instead, this alters resolve to consult the local
resolve state (as opposed to the shared one) to test whether an import succeeded
or not.
Closes#13404
Resolve is currently erroneously allowing imports through private `use`
statements in some circumstances, even across module boundaries. For example,
this code compiles successfully today:
use std::c_str;
mod test {
use c_str::CString;
}
This should not be allowed because it was explicitly decided that private `use`
statements are purely bringing local names into scope, they are not
participating further in name resolution.
As a consequence of this patch, this code, while valid today, is now invalid:
mod test {
use std::c_str;
unsafe fn foo() {
::test::c_str::CString::new(0 as *u8, false);
}
}
While plausibly acceptable, I found it to be more consistent if private imports
were only considered candidates to resolve the first component in a path, and no
others.
Closes#12612
This commit makes sure that code inlined from other functions isn't assigned the source position of the call site, since this leads to undesired behavior when setting line breakpoints (issue #12886)
This commit changes the way move errors are reported when some value is
captured by a PatIdent. First, we collect all of the "cannot move out
of" errors before reporting them, and those errors with the same "move
source" are reported together. If the move is caused by a PatIdent (that
binds by value), we add a note indicating where it is and suggest the
user to put `ref` if they don't want the value to move. This makes the
"cannot move out of" error in match expression nicer (though the extra
note may not feel that helpful in other places :P). For example, with
the following code snippet,
```rust
enum Foo {
Foo1(~u32, ~u32),
Foo2(~u32),
Foo3,
}
fn main() {
let f = &Foo1(~1u32, ~2u32);
match *f {
Foo1(num1, num2) => (),
Foo2(num) => (),
Foo3 => ()
}
}
```
Errors before the change:
```rust
test.rs:10:9: 10:25 error: cannot move out of dereference of `&`-pointer
test.rs:10 Foo1(num1, num2) => (),
^~~~~~~~~~~~~~~~
test.rs:10:9: 10:25 error: cannot move out of dereference of `&`-pointer
test.rs:10 Foo1(num1, num2) => (),
^~~~~~~~~~~~~~~~
test.rs:11:9: 11:18 error: cannot move out of dereference of `&`-pointer
test.rs:11 Foo2(num) => (),
^~~~~~~~~
```
After:
```rust
test.rs:9:11: 9:13 error: cannot move out of dereference of `&`-pointer
test.rs:9 match *f {
^~
test.rs:10:14: 10:18 note: attempting to move value to here (to prevent the move, you can use `ref num1` to capture value by reference)
test.rs:10 Foo1(num1, num2) => (),
^~~~
test.rs:10:20: 10:24 note: and here (use `ref num2`)
test.rs:10 Foo1(num1, num2) => (),
^~~~
test.rs:11:14: 11:17 note: and here (use `ref num`)
test.rs:11 Foo2(num) => (),
^~~
```
Close#8064
This fixes the categorization of the upvars of procs (represented internally
as once fns) to consider usage to require a loan. In doing so, upvars are no
longer allowed to be moved out of repeatedly in loops and such.
Closes#10398Closes#12041Closes#12127
This commit removes the compiler support for floating point modulus operations,
as well as from the language. An implementation for this operator is now
required to be provided by libraries.
Floating point modulus is rarely used, doesn't exist in C, and is always lowered
to an fmod library call by LLVM, and LLVM is considering removing support
entirely.
Closes#12278
Previously, if statements of the form "Foo;" or "let _ = Foo;" were encountered
where Foo had a destructor, the destructors were not run. This changes
the relevant locations in trans to check for ty::type_needs_drop and invokes
trans_to_lvalue instead of trans_into.
Closes#4734Closes#6892
`Reader`, `Writer`, `MemReader`, `MemWriter`, and `MultiWriter` now work with `Vec<u8>` instead of `~[u8]`. This does introduce some extra copies since `from_utf8_owned` isn't usable anymore, but I think that can't be helped until `~str`'s representation changes.
In the error message for when a private field is used, include the name of the struct, or if it's a struct-like enum variant, the names of the variant and the enum.
This fixes#13341.
This can be a frustrating error message, ideally we should print the signature mismatch, but hinting that it's a trait incompatibility helps tracking root cause. Also beefed up the testcases for this.
Ideally we would print the signature mismatch in the error helper?
rustc: move the check_loop pass earlier.
This pass is purely AST based, and by running it earlier we emit more
useful error messages, e.g. type inference fails in the case of
`let r = break;` with few constraints on `r`, but it's more useful to be told that
the `break` is outside the loop (rather than a type error) when it is.
Closes#13292.
This pass is purely AST based, and by running it earlier we emit more
useful error messages, e.g. type inference fails in the case of `let r =
break;` with few constraints on `r`, but its more useful to be told that
the `break` is outside a loop (rather than a type error) when it is.
Closes#13292.
`RefCell::get` can be a bit surprising, because it actually clones the wrapped value. This removes `RefCell::get` and replaces all the users with `RefCell::borrow()` when it can, and `RefCell::borrow().clone()` when it can't. It removes `RefCell::set` for consistency. This closes#13182.
It also fixes an infinite loop in a test when debugging is on.
This was missed when dropping the null-termination from our string
types. An explicit null byte can still be placed anywhere in a string if
desired, but there's no reason to stick one at the end of every string
constant.
It's surprising that `RefCell::get()` is implicitly doing a clone
on a value. This patch removes it and replaces all users with
either `.borrow()` when we can autoderef, or `.borrow().clone()`
when we cannot.
This was missed when dropping the null-termination from our string
types. An explicit null byte can still be placed anywhere in a string if
desired, but there's no reason to stick one at the end of every string
constant.
This change removes the AbiSet from the AST, converting all usage to have just
one Abi value. The current scheme selects a relevant ABI given a list of ABIs
based on the target architecture and how relevant each ABI is to that
architecture.
Instead of this mildly complicated scheme, only one ABI will be allowed in abi
strings, and pseudo-abis will be created for special cases as necessary. For
example the "system" abi exists for stdcall on win32 and C on win64.
Closes#10049
This commit tightens up the restriction on types used to index slices to require
exactly `uint` indices. Previously any integral type was accepted, but this
leads to a few subtle problems:
* 64-bit indices don't make much sense on 32-bit systems
* Signed indices for slices used as negative indexing isn't implemented
This was discussed at the recent work week, and also has some discussion on
issue #10453.
Closes#10453
This commit deals with the fallout of the previous change by making tuples
structs have public fields where necessary (now that the fields are private by
default).
This is a continuation of the work done in #13184 to make struct fields private
by default. This commit finishes RFC 4 by making all tuple structs have private
fields by default. Note that enum variants are not affected.
A tuple struct having a private field means that it cannot be matched on in a
pattern match (both refutable and irrefutable), and it also cannot have a value
specified to be constructed. Similarly to private fields, switching the type of
a private field in a tuple struct should be able to be done in a backwards
compatible way.
The one snag that I ran into which wasn't mentioned in the RFC is that this
commit also forbids taking the value of a tuple struct constructor. For example,
this code now fails to compile:
mod a {
pub struct A(int);
}
let a: fn(int) -> a::A = a::A; //~ ERROR: first field is private
Although no fields are bound in this example, it exposes implementation details
through the type itself. For this reason, taking the value of a struct
constructor with private fields is forbidden (outside the containing module).
RFC: 0004-private-fields
This commit switches privacy's checking of fields to have *all* fields be
private by default. This does not yet change tuple structs, this only affects
structs with named fields. The fallout of this change will follow shortly.
RFC: 0004-private-fields
cc #8122Closes#11809
`collections::list::List` was decided in a [team meeting](https://github.com/mozilla/rust/wiki/Meeting-weekly-2014-03-25) that it was unnecessary, so this PR removes it. Additionally, it removes an old and redundant purity test and fixes some warnings.
Only supports crate level statics. No debug info is generated for function level statics. Closes#9227.
As discussed at the end of the comments for #9227, I took an initial stab at adding support for function level statics and decided it would be enough work to warrant being split into a separate issue.
See #13144 for the new issue describing the need to add support for function level static variables.
r? @nikomatsakis
Fix#13140
Includes two fixes, and a semi-thorough regression test.
(There is another set of tests that I linked from #5121, but those are sort of all over the place, while the ones I've included here are more directly focused on the issues at hand.)
Summary:
So far, we've used the term POD "Plain Old Data" to refer to types that
can be safely copied. However, this term is not consistent with the
other built-in bounds that use verbs instead. This patch renames the Pod
kind into Copy.
RFC: 0003-opt-in-builtin-traits
Test Plan: make check
Reviewers: cmr
Differential Revision: http://phabricator.octayn.net/D3
This commit removes implicitly adding the Send bound to ~Trait objects and
procedure types. It will now be manually required to specify that a procedure
or trait must be send-able.
Closes#10296
The `_match.rs` takes advantage of passes prior to `trans` and
aggressively prunes the sub-match tree based on exact equality. When it
comes to literal or range, the strategy may lead to wrong result if
there's guard function or multiple patterns inside tuple.
Closes#12582.
Closes#13027.
This change prepares `rustc` to accept private fields by default. These changes will have to go through a snapshot before the rest of the changes can happen.
This change is in preparation for #8122. Nothing is currently done with these
visibility qualifiers, they are just parsed and accepted by the compiler.
RFC: 0004-private-fields
The problem was that we need to apply the substitution, so that the
formal lifetime parameters get replaced with (unifiable)
free-lifetimes that can actually be fed into the constraint solver.
Also, refactor code os that substitution for `check_item` and
`check_method`, moving both down the control flow into `check_bare_fn`.
----
Finally, there was another (similar) spot where we needed to
substitute early-bound lifetimes when invoking an object method of a
trait.
It was possible to borrow unsafe static items in static initializers.
This patch implements a small `Visitor` that walks static initializer's
expressions and checks borrows aliasability.
Fixes#13005
cc @nikomatsakis r?
Summary:
It was possible to borrow unsafe static items in static initializers.
This patch implements a small `Visitor` that walks static initializer's
expressions and checks borrows aliasability.
Fixes#13005
Test Plan: make check
Differential Revision: http://phabricator.octayn.net/D2
This needs to be removed as part of removing `~[T]`. Partial type hints
are now allowed, and will remove the need to add a version of this
method for `Vec<T>`. For now, this involves a few workarounds for
partial type hints not completely working.
Replace syntax::opt_vec with syntax::owned_slice
The `owned_slice::OwnedSlice` is `(*T, uint)` (i.e. a direct equivalent to DSTs `~[T]`).
This shaves two words off the old OptVec type; and also makes substituting in other implementations easy, by removing all the mutation methods. (And also everything that's very rarely/never used.)
The compiler will no longer inject libgreen as the default runtime for rust
programs, this commit switches it over to libnative by default. Now that
libnative has baked for some time, it is ready enough to start getting more
serious usage as the default runtime for rustc generated binaries.
We've found that there isn't really a correct decision in choosing a 1:1 or M:N
runtime as a default for all applications, but it seems that a larger number of
programs today would work more reasonable with a native default rather than a
green default.
With this commit come a number of bugfixes:
* The main native task is now named "<main>"
* The main native task has the stack bounds set up properly
* #[no_uv] was renamed to #[no_start]
* The core-run-destroy test was rewritten for both libnative and libgreen and
one of the tests was modified to be more robust.
* The process-detach test was locked to libgreen because it uses signal handling
This is the first step to replacing OptVec with a new representation:
remove all mutability. Any mutations have to go via `Vec` and then make
to `OptVec`.
Many of the uses of OptVec are unnecessary now that Vec has no-alloc
emptiness (and have been converted to Vec): the only ones that really
need it are the AST and sty's (and so on) where there are a *lot* of
instances of them, and they're (mostly) immutable.
In Rust, the strongest guarantee that `&mut` provides is that the memory
pointed to is *not aliased*, whereas `&`'s guarantees are much weaker:
that the value can be aliased, and may be mutated under proper precautions
(interior mutability).
Our atomics though use `&mut` for mutation even while creating multiple
aliases, so this changes them to use 'interior mutability', mutating
through immutable references.
Most of the standard distribution is still using ~[] instead of Vec, so this
lint is essentially useless currently. When the standard distribution has been
ported to not use ~[], then we can turn the lint back on.
`Share` implies that all *reachable* content is *threadsafe*.
Threadsafe is defined as "exposing no operation that permits a data race if multiple threads have access to a &T pointer simultaneously". (NB: the type system should guarantee that if you have access to memory via a &T pointer, the only other way to gain access to that memory is through another &T pointer)...
Fixes#11781
cc #12577
What this PR will do
================
- [x] Add Share kind and
- [x] Replace usages of Freeze with Share in bounds.
- [x] Add Unsafe<T> #12577
- [x] Forbid taking the address of a immutable static item with `Unsafe<T>` interior
What's left to do in a separate PR (after the snapshot)?
===========================================
- Remove `Freeze` completely
These variants occur rarely but inflate the whole enum for the other variants, leaving a lot of wasted space. In total this reduces `ty::sty` from 160 bytes to 96 (on a 64-bit platform).
After this, `ty_struct` and `ty_enum` are the largest variants, with the 80-byte `substs` being the major contributor.
This fixes struct passing abi on x86 ffi: Structs are now passed
indirectly with byval attribute (as clang does).
Empty structs are also explicitly ignored rather than directly passed.
Fixes#5744Fixes#11198Fixes#11343
This reduces the size of sty from 112 to 96; like with the ty_trait
variant, this variant of sty occurs rarely (~1%) so the benefits are
large and the costs small.
This reduces ty::sty from 160 bytes to just 112, and some measurements
eddyb made suggest that the ty_trait variant occurs very
rarely (e.g. ~1% of all sty instances) hence this will result in a large
memory saving, and the cost of the indirection is unlikely to be an
issue.
Remove the linker_private and linker_private_weak linkage attributes,
they have been superseded by private and private_weak and have been
removed in upstream LLVM in commit r203866.
The pretty printer constitues an enormous amount of code, there's no reason for
it to be generic. This just least to a huge amount of metadata which isn't
necessary. Instead, this change migrates the pretty printer to using a trait
object instead.
Closes#12985
