The old code created a flat listing of "HIR -> WorkProduct" edges.
While perfectly general, this could lead to a lot of repetition if the
same HIR nodes affect many work-products. This is set to be a problem
when we start to skip typeck, since we will be adding a lot more
"work-product"-like nodes.
The newer code uses an alternative strategy: it "reduces" the graph
instead. Basically we walk the dep-graph and convert it to a DAG, where
we only keep intermediate nodes if they are used by multiple
work-products.
This DAG does not contain the same set of nodes as the original graph,
but it is guaranteed that (a) every output node is included in the graph
and (b) the set of input nodes that can reach each output node is
unchanged.
(Input nodes are basically HIR nodes and foreign metadata; output nodes
are nodes that have assocaited state which we will persist to disk in
some way. These are assumed to be disjoint sets.)
travis: move IBM backwards in time
Using Ubuntu's cross-toolchains for powerpc* and s390x meant they were
depending on glibc symbols from Ubuntu 16.04. And if that host is ever
updated to a new release, the toolchains would raise the bar too.
This switches powerpc, powerpc64, and s390x to use crosstool-ng
toolchains, configured approximately like RHEL6 with kernel 2.6.32 and
glibc 2.12. This ABI level should also be compatible with Debian 7
(wheezy) and Ubuntu 12.04 (precise).
For powerpc64le, the challenge was that only glibc-2.19 officially added
support, but RHEL7 backported those changes to glibc-2.17. The backport
patches are complex and numerous, so instead of trying to push those
into crosstool-ng, this just uses glibc binaries directly from CentOS 7
and builds the toolchain manually.
This is ported from rust-lang/rust-buildbot#149.
r? @alexcrichton
Remove dead recursive partial eq impl
Its nowhere used (if it had been used used, the rust stack would have overflown
due to the recursion). Its presence was confusing for mrustc.
cc @thepowersgang
Using Ubuntu's cross-toolchains for powerpc* and s390x meant they were
depending on glibc symbols from Ubuntu 16.04. And if that host is ever
updated to a new release, the toolchains would raise the bar too.
This switches powerpc, powerpc64, and s390x to use crosstool-ng
toolchains, configured approximately like RHEL6 with kernel 2.6.32 and
glibc 2.12. This ABI level should also be compatible with Debian 7
(wheezy) and Ubuntu 12.04 (precise).
For powerpc64le, the challenge was that only glibc-2.19 officially added
support, but RHEL7 backported those changes to glibc-2.17. The backport
patches are complex and numerous, so instead of trying to push those
into crosstool-ng, this just uses glibc binaries directly from CentOS 7
and builds the toolchain manually.
This is ported from rust-lang/rust-buildbot#149.
r? @alexcrichton
Make backtraces work on Windows GNU targets again.
This is done by adding a function that can return a filename
to pass to backtrace_create_state. The filename is obtained in
a safe way by first getting the filename, locking the file so it can't
be moved, and then getting the filename again and making sure it's the same.
See: https://github.com/rust-lang/rust/pull/37359#issuecomment-260123399
Issue: #33985
Note though that this isn't that pretty...
I had to implement a `WideCharToMultiByte` wrapper function to convert to the ANSI code page. This will work better than only allowing ASCII provided that the ANSI code page is set to the user's local language, which is often the case.
Also, please make sure that I didn't break the Unix build.
Perform lifetime elision (more) syntactically, before type-checking.
The *initial* goal of this patch was to remove the (contextual) `&RegionScope` argument passed around `rustc_typeck::astconv` and allow converting arbitrary (syntactic) `hir::Ty` to (semantic) `Ty`.
I've tried to closely match the existing behavior while moving the logic to the earlier `resolve_lifetime` pass, and [the crater report](https://gist.github.com/eddyb/4ac5b8516f87c1bfa2de528ed2b7779a) suggests none of the changes broke real code, but I will try to list everything:
There are few cases in lifetime elision that could trip users up due to "hidden knowledge":
```rust
type StaticStr = &'static str; // hides 'static
trait WithLifetime<'a> {
type Output; // can hide 'a
}
// This worked because the type of the first argument contains
// 'static, although StaticStr doesn't even have parameters.
fn foo(x: StaticStr) -> &str { x }
// This worked because the compiler resolved the argument type
// to <T as WithLifetime<'a>>::Output which has the hidden 'a.
fn bar<'a, T: WithLifetime<'a>>(_: T::Output) -> &str { "baz" }
```
In the two examples above, elision wasn't using lifetimes that were in the source, not even *needed* by paths in the source, but rather *happened* to be part of the semantic representation of the types.
To me, this suggests they should have never worked through elision (and they don't with this PR).
Next we have an actual rule with a strange result, that is, the return type here elides to `&'x str`:
```rust
impl<'a, 'b> Trait for Foo<'a, 'b> {
fn method<'x, 'y>(self: &'x Foo<'a, 'b>, _: Bar<'y>) -> &str {
&self.name
}
}
```
All 3 of `'a`, `'b` and `'y` are being ignored, because the `&self` elision rule only cares that the first argument is "`self` by reference". Due implementation considerations (elision running before typeck), I've limited it in this PR to a reference to a primitive/`struct`/`enum`/`union`, but not other types, but I am doing another crater run to assess the impact of limiting it to literally `&self` and `self: &Self` (they're identical in HIR).
It's probably ideal to keep an "implicit `Self` for `self`" type around and *only* apply the rule to `&self` itself, but that would result in more bikeshed, and #21400 suggests some people expect otherwise.
Another decent option is treating `self: X, ... -> Y` like `X -> Y` (one unique lifetime in `X` used for `Y`).
The remaining changes have to do with "object lifetime defaults" (see RFCs [599](https://github.com/rust-lang/rfcs/blob/master/text/0599-default-object-bound.md) and [1156](https://github.com/rust-lang/rfcs/blob/master/text/1156-adjust-default-object-bounds.md)):
```rust
trait Trait {}
struct Ref2<'a, 'b, T: 'a+'b>(&'a T, &'b T);
// These apply specifically within a (fn) body,
// which allows type and lifetime inference:
fn main() {
// Used to be &'a mut (Trait+'a) - where 'a is one
// inference variable - &'a mut (Trait+'b) in this PR.
let _: &mut Trait;
// Used to be an ambiguity error, but in this PR it's
// Ref2<'a, 'b, Trait+'c> (3 inference variables).
let _: Ref2<Trait>;
}
```
What's happening here is that inference variables are created on the fly by typeck whenever a lifetime has no resolution attached to it - while it would be possible to alter the implementation to reuse inference variables based on decisions made early by `resolve_lifetime`, not doing that is more flexible and works better - it can compile all testcases from #38624 by not ending up with `&'static mut (Trait+'static)`.
The ambiguity specifically cannot be an early error, because this is only the "default" (typeck can still pick something better based on the definition of `Trait` and whether it has any lifetime bounds), and having an error at all doesn't help anyone, as we can perfectly infer an appropriate lifetime inside the `fn` body.
**TODO**: write tests for the user-visible changes.
cc @nikomatsakis @arielb1
Use __SIZEOF_INT128__ to test __int128 presence
Previously we tested whether a handful of preprocessor variables indicating certain 64 bit
platforms, but this does not work for other 64 bit targets which have support for __int128 in C
compiler.
Use the `__SIZEOF__INT128__` preprocessor variable instead. This variable gets set to 16 by gcc and
clang for every target where __int128 is supported.
