[incremental] Hash `Allocation`s
`HashSet::insert` returns `true` if the value did not exist, which is the timing we want to hash the `Allocation`.
Fixes#49595
cc @oli-obk
mem-categorization, coherence fix
make mem-categorization use adjusted type for patterns: Fixes#49631
do not propagate `Err` when determing causal info: Fixes#48728
r? @eddyb
Fix regression in defaults #49344Fixes#49344 by not checking the well-formedness wrt defaults of predicates that contain lifetimes, which is consistent with not checking generic predicates.
r? @nikomatsakis
two-phase borrows: support multiple activations in one statement
The need for this has arisen since the introduction of two-phase borrows on
method autorefs in #49348. r'ing @pnkfelix to keep things off Niko's plate so he can make this redundant, and @pnkfelix is familiar with the code.
Fixes#49635Fixes#49662
r? @pnkfelix
Fix another circular deps link args issue
It turns out that the support in #49316 wasn't enough to handle all cases
notably the example in #48661. The underlying bug was connected to panic=abort
where lang items were listed in the `missing_lang_items` sets but didn't
actually exist anywhere.
This caused the linker backend to deduce that start-group/end-group wasn't
needed because not all items were defined. Instead the missing lang items that
don't actually need to have a definition are filtered out and not considered for
the start-group/end-group arguments
Closes#48661
Allow for re-using monomorphizations in upstream crates.
Followup to #48611. This implementation is pretty much finished modulo failing tests if there are any. Not quite ready for review yet though.
### DESCRIPTION
This PR introduces a `share-generics` mode for RLIBs and Rust dylibs. When a crate is compiled in this mode, two things will happen:
- before instantiating a monomorphization in the current crate, the compiler will look for that monomorphization in all upstream crates and link to it, if possible.
- monomorphizations are not internalized during partitioning. Instead they are added to the list of symbols exported from the crate.
This results in less code being translated and LLVMed. However, there are also downsides:
- it will impede optimization somewhat, since fewer functions can be internalized, and
- Rust dylibs will have bigger symbol tables since they'll also export monomorphizations.
Consequently, this PR only enables the `shared-generics` mode for opt-levels `No`, `Less`, `Size`, and `MinSize`, and for when incremental compilation is activated. `-O2` and `-O3` will still generate generic functions per-crate.
Another thing to note is that this has a somewhat similar effect as MIR-only RLIBs, in that monomorphizations are shared, but it is less effective because it cannot share monomorphizations between sibling crates:
```
A <--- defines `fn foo<T>() { .. }`
/ \
/ \
B C <--- both call `foo<u32>()`
\ /
\ /
D <--- calls `foo<u32>()` too
```
With `share-generics`, both `B` and `C` have to instantiate `foo<u32>` and only `D` can re-use it (from either `B` or `C`). With MIR-only RLIBs, `B` and `C` would not instantiate anything, and in `D` we would then only instantiate `foo<u32>` once.
On the other hand, when there are many leaf crates in the graph (e.g. when compiling many individual test binaries) then the `share-generics` approach will often be more effective.
### TODO
- [x] Add codegen test that makes sure monomorphizations can be internalized in non-Rust binaries.
- [x] Add codegen-units test that makes sure we share generics.
- [x] Add run-make test that makes sure we don't export any monomorphizations from non-Rust binaries.
- [x] Review for reproducible-builds implications.
In intercrate mode, if we determine that a particular `T: Trait` is
unknowable, we sometimes also go and get extra causal information. An
errant `?` was causing us to propagate an error found in that process
out as if `T: Trait` was not unknowable but rather not provable. This
led to an ICE.
proc_macro: Reorganize public API
This commit is a reorganization of the `proc_macro` crate's public user-facing
API. This is the result of a number of discussions at the recent Rust All-Hands
where we're hoping to get the `proc_macro` crate into ship shape for
stabilization of a subset of its functionality in the Rust 2018 release.
The reorganization here is motivated by experiences from the `proc-macro2`,
`quote`, and `syn` crates on crates.io (and other crates which depend on them).
The main focus is future flexibility along with making a few more operations
consistent and/or fixing bugs. A summary of the changes made from today's
`proc_macro` API is:
* The `TokenNode` enum has been removed and the public fields of `TokenTree`
have also been removed. Instead the `TokenTree` type is now a public enum
(what `TokenNode` was) and each variant is an opaque struct which internally
contains `Span` information. This makes the various tokens a bit more
consistent, require fewer wrappers, and otherwise provides good
future-compatibility as opaque structs are easy to modify later on.
* `Literal` integer constructors have been expanded to be unambiguous as to what
they're doing and also allow for more future flexibility. Previously
constructors like `Literal::float` and `Literal::integer` were used to create
unsuffixed literals and the concrete methods like `Literal::i32` would create
a suffixed token. This wasn't immediately clear to all users (the
suffixed/unsuffixed aspect) and having *one* constructor for unsuffixed
literals required us to pick a largest type which may not always be true. To
fix these issues all constructors are now of the form
`Literal::i32_unsuffixed` or `Literal::i32_suffixed` (for all integral types).
This should allow future compatibility as well as being immediately clear
what's suffixed and what isn't.
* Each variant of `TokenTree` internally contains a `Span` which can also be
configured via `set_span`. For example `Literal` and `Term` now both
internally contain a `Span` rather than having it stored in an auxiliary
location.
* Constructors of all tokens are called `new` now (aka `Term::intern` is gone)
and most do not take spans. Manufactured tokens typically don't have a fresh
span to go with them and the span is purely used for error-reporting
**except** the span for `Term`, which currently affects hygiene. The default
spans for all these constructed tokens is `Span::call_site()` for now.
The `Term` type's constructor explicitly requires passing in a `Span` to
provide future-proofing against possible hygiene changes. It's intended that a
first pass of stabilization will likely only stabilize `Span::call_site()`
which is an explicit opt-in for "I would like no hygiene here please". The
intention here is to make this explicit in procedural macros to be
forwards-compatible with a hygiene-specifying solution.
* Some of the conversions for `TokenStream` have been simplified a little.
* The `TokenTreeIter` iterator was renamed to `token_stream::IntoIter`.
Overall the hope is that this is the "final pass" at the API of `TokenStream`
and most of `TokenTree` before stabilization. Explicitly left out here is any
changes to `Span`'s API which will likely need to be re-evaluated before
stabilization.
All changes in this PR have already been reflected to the [`proc-macro2`],
`quote`, and `syn` crates. New versions of all these crates have also been
published to crates.io.
Once this lands in nightly I plan on making an internals post again summarizing
the changes made here and also calling on all macro authors to give the APIs a
spin and see how they work. Hopefully pending no major issues we can then have
an FCP to stabilize later this cycle!
[`proc-macro2`]: https://docs.rs/proc-macro2/0.3.1/proc_macro2/Closes#49596
Expand macros in `extern {}` blocks
This permits macro and proc-macro and attribute invocations (the latter only with the `proc_macro` feature of course) in `extern {}` blocks, gated behind a new `macros_in_extern` feature.
A tracking issue is now open at #49476closes#48747
Rollup of 9 pull requests
Successful merges:
- #48658 (Add a generic CAS loop to std::sync::Atomic*)
- #49253 (Take the original extra-filename passed to a crate into account when resolving it as a dependency)
- #49345 (RFC 2008: Finishing Touches)
- #49432 (Flush executables to disk after linkage)
- #49496 (Add more vec![... ; n] optimizations)
- #49563 (add a dist builder to build rust-std components for the THUMB targets)
- #49654 (Host compiler documentation: Include private items)
- #49667 (Add more features to rust_2018_preview)
- #49674 (ci: Remove x86_64-gnu-incremental builder)
Failed merges:
It turns out that the support in #49316 wasn't enough to handle all cases
notably the example in #48661. The underlying bug was connected to panic=abort
where lang items were listed in the `missing_lang_items` sets but didn't
actually exist anywhere.
This caused the linker backend to deduce that start-group/end-group wasn't
needed because not all items were defined. Instead the missing lang items that
don't actually need to have a definition are filtered out and not considered for
the start-group/end-group arguments
Closes#48661
Rollup of 25 pull requests
Successful merges:
- #49179 (Handle future deprecation annotations )
- #49512 (Add support for variant and types fields for intra links)
- #49515 (fix targetted value background)
- #49516 (Add missing anchor for union type fields)
- #49532 (Add test for rustdoc ignore test)
- #49533 (Add #[must_use] to a few standard library methods)
- #49540 (Fix miri Discriminant() for non-ADT)
- #49559 (Introduce Vec::resize_with method (see #41758))
- #49570 (avoid IdxSets containing garbage above the universe length)
- #49577 (Stabilize String::replace_range)
- #49599 (Fix typo)
- #49603 (Fix url for intra link provided method)
- #49607 (Stabilize iterator methods in 1.27)
- #49609 (run-pass/attr-stmt-expr: expand test cases)
- #49612 (Fix "since" version for getpid feature.)
- #49618 (Fix build error when compiling libcore for 16bit targets)
- #49619 (tweak core::fmt docs)
- #49637 (Stabilize parent_id())
- #49639 (Update Cargo)
- #49628 (Re-write the documentation index)
- #49594 (Add some performance guidance to std::fs and std::io docs)
- #49625 (miri: add public alloc_kind accessor)
- #49634 (Add a test for the fix to issue #43058)
- #49641 (Regression test for #46314)
- #49547 (Unignore borrowck test)
Failed merges:
Better document the implementors of Clone and Copy
There are two parts to this change. The first part is a change to the compiler and to the standard library (specifically, libcore) to allow implementations of `Clone` and `Copy` to be written for a subset of builtin types. By adding these implementations to libcore, they now show up in the documentation. This is a [breaking-change] for users of `#![no_core]`, because they will now have to supply their own copy of the implementations of `Clone` and `Copy` that were added in libcore.
The second part is purely a documentation change to document the other implementors of `Clone` and `Copy` that cannot be described in Rust code (yet) and are thus provided by the compiler.
Fixes#25893
Remove adjacent all-const match arm hack.
An old fix for moves-in-guards had a hack for adjacent all-const match arms.
The hack was explained in a comment, which you can see here:
https://github.com/rust-lang/rust/pull/22580/files#diff-402a0fa4b3c6755c5650027c6d4cf1efR497
But hack was incomplete (and thus unsound), as pointed out here:
https://github.com/rust-lang/rust/issues/47295#issuecomment-357108458
Plus, it is likely to be at least tricky to reimplement this hack in
the new NLL borrowck.
So rather than try to preserve the hack, we want to try to just remove
it outright. (At least to see the results of a crater run.)
[breaking-change]
This is a breaking-change, but our hope is that no one is actually
relying on such an extreme special case. (We hypothesize the hack was
originally added to accommodate a file in our own test suite, not code
in the wild.)
