This fixes#27254.
On a 64-bit Linux machine, for example, `configure --libdir=/usr/local/lib64` was creating both `x86_64-unknown-linux-gnu/stage0/lib/rustlib` and `x86_64-unknown-linux-gnu/stage0/lib64/rustlib`. Crates from the stage0 snapshot, like `libcore`, are extracted to `x86_64-unknown-linux-gnu/stage0/lib/rustlib`, but the stage0 compiler was attempting to find them in `x86_64-unknown-linux-gnu/stage0/lib64/rustlib`, which has the highest priority on a 64-bit system.
The issue can be fixed by creating only `x86_64-unknown-linux-gnu/stage0/lib/rustlib`, since this is the only rustlib directory needed for stage0 anyways.
This PR for #29789 uses `rustc::repr::mir::Constant` in `ExprKind::Repeat`, which seems to fit quite nicely. Is there a reason for not re-using that type?
Tuple and unit variants from other crates weren't put into type namespace.
Now variant namespacing is aligned with struct namespacing and is not affected by the variant's crate of origin (struct -> type, tuple/unit -> type/value).
Additionally, struct variants from other crates are put into value namespace (struct variants from local crate were already in it). This is not a necessity, but a future proofing measure.
This fix can result in some new shadowing errors in cross-crate scenarios, crater reports [three regressions](https://github.com/rust-lang/rust/pull/30882#issuecomment-172369883).
[breaking-change]
Use raw pointers to avoid aliasing in str::split_at_mut
Introduce private function from_raw_parts_mut for str to factor out the logic.
We want to use raw pointers here instead of duplicating a &mut str, to
be on safer ground w.r.t rust aliasing rules.
This has already been fixed for slices in PR #27358, issue #27357
was the major use-case, and to update the dep-graph. Other kinds of
predicates are now excluded from the cache because there is no easy way
to make a good dep-graph node for them, and because they are not
believed to be that useful. :)
Fixes#30741. (However, the test still gives wrong result for trans,
for an independent reason which is fixed in the next commit.)
Introduce private function from_raw_parts_mut for str to factor out the logic.
We want to use raw pointers here instead of duplicating a &mut str, to
be on safer ground w.r.t rust aliasing rules.
Use the fallback impl for memrchr on non-linux
The memrchr code was never used(!). This brings the memrchr improvements to
non-linux platforms (LineWriter / buffered stdout benefits).
Previous PR #30381
All structs and their constructors are defined as `DefStruct`.
`DefTy` is splitted into `DefEnum` and `DefTyAlias`.
Ad hoc flag `bool is_structure` is removed from `DefVariant`, it was required in one place in resolve and could be obtained by other means.
Flag `bool is_ctor` is removed from `DefFn`, it wasn't really used for constructors outside of metadata decoding.
Observable effects:
More specific error messages are selected in some cases.
Two name resolution bugs fixed (https://github.com/rust-lang/rust/issues/30992 and FIXME in compile-fail/empty-struct-braces-expr.rs).
Fixes https://github.com/rust-lang/rust/issues/30992
Closes https://github.com/rust-lang/rust/issues/30361
As an attempt to make loop body destination be optional, author implemented a pretty self contained
change and deemed it to be (much) uglier than the alternative of just keeping the unit temporary.
Having the temporary created lazily also has a nice property of not figuring in the MIR of
functions which do not use loops of any sort.
r? @nikomatsakis
We've been seeing a lot of timeouts in tests on the bots and investigation ended
pointing to jemalloc/jemalloc#315 as the culprit. Unfortunately it looks like
that doesn't seem to have a fix on the way soon, so let's temporarily downgrade
back to the previous version of jemalloc we were using (where #30434 was the
most recent upgrade)
The protocol for `serialize::{En,De}code` doesn't allow for two
integers to be serialized next to each other. This switches the
protocol to serializing `Span`s as a struct. rbml structs don't
have any overhead, so the metadata shouldn't increase in size,
but it allows the json format to be properly generated, albeit
slightly more heavy than when it was just serializing a span as
a u64.
Closes#31025.
s
