Make `i*::signum` a `const fn`.
Ticks a box in #53718.
This uses a well-known branchless implementation of `signum`: `(n > 0) as i32 - (n < 0) as i32`.
Here's a [playground](https://play.rust-lang.org/?version=nightly&mode=release&edition=2018&gist=747cf191c4974bf66c9d75e509ae6e6e) comparing the two techniques. On x86 in release mode, the branchless implementation is able to replace a `mov` and `cmov` with a `sar` and `add`, so this should be a bit faster as well.
~~This is marked as a draft since I think I'll need to add `#[rustc_const_unstable]` somewhere. Perhaps the reviewer can point me in the right direction.~~
syntax: Treat error literals in more principled way
Free them from their character literal origins.
I actually tried to remove `LitKind::Err` entirely (by converting it into `ExprKind::Err` immediately), and it caused no diagnostic regressions in the test suite.
However, I'd still want to use error literals as general purpose error tokens some day, so I kept them.
The downside of having `LitKind::Err` in addition to `ExprKind::Err` is that every time you want to do something with `ExprKind::Err` you need to make sure that `ExprKind::Lit(LitKind::Err)` is treated in the same way.
Fortunately, this usually happens automatically because both literals and errors are "leaf" expressions, however this PR does fix a couple of inconsistencies between them.
Addresses https://github.com/rust-lang/rust/pull/60679#discussion_r282640663 in a way
[const-prop] Support Rvalue::{Ref,Len} and Deref
Also fixes an ICE I found in testing.
r? @oli-obk
~~The final commit is just for a perf run. I'll remove it after that is completed.~~
Remove asterisk suggestion for move errors in borrowck
As per the decision in #54985 completely removes the suggestion to add an asterisk when checking move errors. I believe I've preserved the correct behavior with the "consider borrowing here" branch of the original match arm, but I'm not positive on that.
This is my first PR to rustc so any feedback is greatly appreciated. Thanks.
Make tuple constructors real const fns
Mir construction special cases `Ctor(...)` to be lowered as `Ctor { 0: ... }`, which means this doesn't come up much in practice, but it seems inconsistent not to allow this.
r? @oli-obk
Emit StorageDead along unwind paths for generators
Completion of the work done in #60840. That PR made a change to implicitly consider a local `StorageDead` after Drop, but that was incorrect for DropAndReplace (see also #61060 which tried to fix this in a different way).
This finally enables the optimization implemented in #60187.
r? @eddyb
cc @Zoxc @cramertj @RalfJung
Implementation of RFC 2289 (associated_type_bounds)
This PR implements the [`asociated_type_bounds` feature](https://github.com/rust-lang/rfcs/blob/master/text/2289-associated-type-bounds.md).
Associated type bounds are implemented in:
- function/method arguments and return types
- structs, enums, unions
- associated items in traits
- type aliases
- type parameter defaults
- trait objects
- let bindings
CC @nikomatsakis @centril
Rollup of 5 pull requests
Successful merges:
- #61503 (Fix cfg(test) build for x86_64-fortanix-unknown-sgx)
- #61534 (Edit docs of ExitStatus)
- #61536 (Don't allow using const fn arguments as "args_required_const")
- #61538 (Don't use GNU noexec stack note)
- #61546 (azure: Fix some minor issues which have broken our configuration )
Failed merges:
r? @ghost
Add "type_name" support in emulate_intrinsic()
I did some dumb Git things and deleted my original fork repo semi-accidentally (but probably for the best as I'd messed up the history.)
This is the same issue as #61399, which was obviously auto-closed, to be clear.
Fix NLL typeck ICEs
* Don't ICE when a type containing a region is constrained by nothing
* Don't ICE trying to normalize a type in a `ParamEnv` containing global bounds.
To explain what was happening in the `issue-61311-normalize.rs` case:
* When borrow checking the `the_fn` in the last `impl` we would try to normalize `Self::Proj` (`<Unit as HasProjFn>::Proj`).
* We would find the `impl` that we're checking and and check its `where` clause.
* This would need us to check `<Box<dyn Obj + 'static> as HasProj>::Proj: Bound`
* We find two possible implementations, the blanket impl and the bound in our `ParamEnv`.
* The bound in our `ParamEnv` was canonicalized, so we don't see it as a global bound. As such we prefer it to the `impl`.
* This means that we cannot normalize `<Box<dyn Obj + 'static> as HasProj>::Proj` to `Unit`.
* The `<Box<dyn Obj + 'static> as HasProj>::Proj: Bound` bound, which looks like it should be in our `ParamEnv` has been normalized to `Unit: Bound`.
* We fail to prove `<Box<dyn Obj + 'static> as HasProj>::Proj: Bound`.
* We ICE, since we believe typeck have errored.
Closes#61311Closes#61315Closes#61320
r? @pnkfelix
cc @nikomatsakis
