Auto merge of #86841 - GuillaumeGomez:reexported-macro-2-render, r=Stupremee
Fix rendering of reexported macros 2.0 and fix visibility of reexported items
So, this PR grew a bit out of focus, it does the following things:
* Fixes #86276.
* Fixes visibility display for reexported items: it now takes the visibility of the "use" statement rather than the visibility of the reexported item itself).
* Fixes the display of reexported items if "--document-private-items" option is used. Before, they were simply skipped.
* Fixes inconsistency on typedef items: they didn't display their visibility contrary to other items.
I added tests to check everything listed above.
cc `@camelid` `@ollie27` (in case one of you want to review?)
Auto merge of #86815 - FabianWolff:issue-84210, r=varkor
Improve error reporting for modifications behind `&` references
I had a look at #84210 and noticed that #85823 has effectively already fixed #84210.
However, the string matching in #85823 is _very_ crude and already breaks down when a variable name starts with `mut`. I have made this a bit more robust; further improvements could definitely be made but are complicated by the lack of information provided by an earlier pass:
https://github.com/rust-lang/rust/blob/ce331ee6ee010438d1a58c7da8ced4f26d69a20e/compiler/rustc_mir_build/src/build/matches/mod.rs#L2103-L2107
I have also fixed a missing comma in the error message.
Auto merge of #86249 - FabianWolff:issue-86238, r=varkor
Report an error if resolution of closure call functions failed
This pull request fixes #86238. The current implementation seems to assume that resolution of closure call functions (I'm not sure what the proper term is; I mean `call` of `Fn` etc.) can never fail:
https://github.com/rust-lang/rust/blob/60f1a2fc4b535ead9c85ce085fdce49b1b097531/compiler/rustc_typeck/src/check/callee.rs#L590-L595
But actually, it can, if the `fn`/`fn_mut`/`fn_once` lang items are not defined, or don't have an associated `call`/`call_mut`/`call_once` function, leading to the ICE described in #86238. I have therefore turned the `span_bug!()` into an error message, which prevents the ICE.
Auto merge of #87068 - JohnTitor:rollup-2xuisfx, r=JohnTitor
Rollup of 8 pull requests
Successful merges:
- #73936 (Rustdoc: Change all 'optflag' arguments to 'optflagmulti')
- #86926 (Update regex crates)
- #86951 ([docs] Clarify behaviour of f64 and f32::sqrt when argument is negative zero)
- #87031 (Update reference.md)
- #87037 (cleanup(rustdoc): remove unused function getObjectNameById)
- #87045 (Fix tracking issue for `bool_to_option`)
- #87049 (Account for `submodules = false` in config.toml when updating LLVM submodule)
- #87061 (Do not suggest adding a semicolon after `?`)
Rollup merge of #87061 - FabianWolff:issue-87051, r=oli-obk
Do not suggest adding a semicolon after `?`
Fixes #87051. I have only modified `report_return_mismatched_types()`, i.e. my changes only affect suggestions to add `;` for return type mismatches, but this never makes sense after `?`, because the function cannot return `()` if `?` is used (it has to return a `Result` or an `Option`), and a semicolon won't help if the expected and actual `Err` types differ, even if the expected one is `()`.
Rollup merge of #87037 - notriddle:notriddle/search-cleanup-getobjectnamefromid, r=GuillaumeGomez
cleanup(rustdoc): remove unused function getObjectNameById
This function was used in an earlier version, when idx's were used to serialize function inputs and outputs. That's not done any more, so removed the JS-side support for it.
Rollup merge of #87031 - ZuseZ4:patch-1, r=GuillaumeGomez
Update reference.md
I ran into a link to the outdated src/doc/reference.md here: https://users.rust-lang.org/t/conditional-compilation-for-debug-release/1098/6
Apparently the Rust reference has moved again, so the link gave a 404 error. This should fix it.
Rollup merge of #73936 - zachlute:rustdoc-optflagmulti, r=jyn514
Rustdoc: Change all 'optflag' arguments to 'optflagmulti'
Because specifying these flags multiple times will never be discernibly different in functionality from specifying them a single time, there is no reason to fail and report an error to the user.
This might be a slightly controversial change. it's tough to say, but it's hard to imagine a case where somebody was depending on this behavior, and doing this seem actively better for the user.
This originally came up in discussion of a fix for [Cargo #8373](https://github.com/rust-lang/cargo/issues/8373), in [Cargo PR #8422](https://github.com/rust-lang/cargo/pull/8422).
The issue is that Cargo will automatically add things like `--document-private-items` to binaries, because it's the only thing that makes sense there. Then some poor user comes along and adds `--document-private-items` to their `rustdoc` flags for the project and suddenly they're getting errors for specifying a flag twice and need to track down which targets to actually add it to without getting duplicates for reasons they won't understand without deep understanding of Cargo behavior.
We're apparently hesitant to inspect `rustdoc` flags provided by the user directly in Cargo, because they're supposed to be opaque, so looking to see if it's already provided before adding it is evidently a non-starter. In trying to resolve that, one suggestion I came up with was to just change `rustdoc` to support passing the flag multiple times, because the user's intent should be clear and it's not *really* an error, so maybe this is a case of 'be permissive in what you accept'.
This PR is an attempt to do that in a straightforward manner for purposes of discussion.
Auto merge of #85941 - cjgillot:qresolve, r=Aaron1011
Reduce the amount of untracked state in TyCtxt -- Take 2
Main part of #85153
The offending line (https://github.com/rust-lang/rust/pull/85153#discussion_r642866298) is replaced by a FIXME until the possible bug and the perf concern are both resolved.
Auto merge of #86995 - sexxi-goose:rewrite, r=nikomatsakis
2229: Rewrite/Refactor Closure Capture Analaysis
While handling all the differnet edge cases the code for the captur analysis got pretty compicated. Looking at the overall picture of the edge cases the rules can still be layed out simply.
Auto merge of #83918 - workingjubilee:stable-rangefrom-pat, r=joshtriplett
Stabilize "RangeFrom" patterns in 1.55
Implements a partial stabilization of #67264 and #37854.
Reference PR: https://github.com/rust-lang/reference/pull/900
# Stabilization Report
This stabilizes the `X..` pattern, shown as such, offering an exhaustive match for unsigned integers:
```rust
match x as u32 {
0 => println!("zero!"),
1.. => println!("positive number!"),
}
```
Currently if a Rust author wants to write such a match on an integer, they must use `1..={integer}::MAX` . By allowing a "RangeFrom" style pattern, this simplifies the match to not require the MAX path and thus not require specifically repeating the type inside the match, allowing for easier refactoring. This is particularly useful for instances like the above case, where different behavior on "0" vs. "1 or any positive number" is desired, and the actual MAX is unimportant.
Notably, this excepts slice patterns which include half-open ranges from stabilization, as the wisdom of those is still subject to some debate.
And I have noticed there are also a handful of "in the wild" users who have deployed it to similar effect, especially in the case of rejecting any value of a certain number or greater. It simply makes it much more ergonomic to write an irrefutable match, as done in Katholieke Universiteit Leuven's [SCALE and MAMBA project](https://github.com/KULeuven-COSIC/SCALE-MAMBA/blob/05e5db00d553573534258585651c525d0da5f83f/WebAssembly/scale_std/src/fixed_point.rs#L685-L695).
## Tests
There were already many tests in [src/test/ui/half-open-range/patterns](https://github.com/rust-lang/rust/tree/90a2e5e3fe59a254d4d707aa291517b3791ea5a6/src/test/ui/half-open-range-patterns), as well as [generic pattern tests that test the `exclusive_range_pattern` feature](https://github.com/rust-lang/rust/blob/673d0db5e393e9c64897005b470bfeb6d5aec61b/src/test/ui/pattern/usefulness/integer-ranges/reachability.rs), many dating back to the feature's introduction and remaining standing to this day. However, this stabilization comes with some additional tests to explore the... sometimes interesting behavior of interactions with other patterns. e.g. There is, at least, a mild diagnostic improvement in some edge cases, because before now, the pattern `0..=(5+1)` encounters the `half_open_range_patterns` feature gate and can thus emit the request to enable the feature flag, while also emitting the "inclusive range with no end" diagnostic. There is no intent to allow an `X..=` pattern that I am aware of, so removing the flag request is a strict improvement. The arrival of the `J | K` "or" pattern also enables some odd formations.
Some of the behavior tested for here is derived from experiments in this [Playground](https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=58777b3c715c85165ac4a70d93efeefc) example, linked at https://github.com/rust-lang/rust/issues/67264#issuecomment-812770692, which may be useful to reference to observe the current behavior more closely.
In addition tests constituting an explanation of the "slicing range patterns" syntax issue are included in this PR.
## Desiderata
The exclusive range patterns and half-open range patterns are fairly strongly requested by many authors, as they make some patterns much more natural to write, but there is disagreement regarding the "closed" exclusive range pattern or the "RangeTo" pattern, especially where it creates "off by one" gaps in the presence of a "catch-all" wildcard case. Also, there are obviously no range analyses in place that will force diagnostics for e.g. highly overlapping matches. I believe these should be warned on, ideally, and I think it would be reasonable to consider such a blocker to stabilizing this feature, but there is no technical issue with the feature as-is from the purely syntactic perspective as such overlapping or missed matches can already be generated today with such a catch-all case. And part of the "point" of the feature, at least from my view, is to make it easier to omit wildcard matches: a pattern with such an "open" match produces an irrefutable match and does not need the wild card case, making it easier to benefit from exhaustiveness checking.
## History
- Implemented:
- Partially via exclusive ranges: https://github.com/rust-lang/rust/pull/35712
- Fully with half-open ranges: https://github.com/rust-lang/rust/pull/67258
- Unresolved Questions:
- The precedence concerns of https://github.com/rust-lang/rust/pull/48501 were considered as likely requiring adjustment but probably wanting a uniform consistent change across all pattern styles, given https://github.com/rust-lang/rust/issues/67264#issuecomment-720711656, but it is still unknown what changes might be desired
- How we want to handle slice patterns in ranges seems to be an open question still, as witnessed in the discussion of this PR!
I checked but I couldn't actually find an RFC for this, and given "approved provisionally by lang team without an RFC", I believe this might require an RFC before it can land? Unsure of procedure here, on account of this being stabilizing a subset of a feature of syntax.
Auto merge of #86965 - sexxi-goose:rfc2229-improve-lint, r=nikomatsakis,lqd
Improves migrations lint for RFC2229
This PR improves the current disjoint capture migration lint by providing more information on why drop order or auto trait implementation for a closure is impacted by the use of the new feature.
The drop order migration lint will now look something like this:
```
error: changes to closure capture in Rust 2021 will affect drop order
--> $DIR/significant_drop.rs:163:21
|
LL | let c = || {
| ^^
...
LL | tuple.0;
| ------- in Rust 2018, closure captures all of `tuple`, but in Rust 2021, it only captures `tuple.0`
...
LL | }
| - in Rust 2018, `tuple` would be dropped here, but in Rust 2021, only `tuple.0` would be dropped here alongside the closure
```
The auto trait migration lint will now look something like this:
```
error: changes to closure capture in Rust 2021 will affect `Send` trait implementation for closure
--> $DIR/auto_traits.rs:14:19
|
LL | thread::spawn(move || unsafe {
| ^^^^^^^^^^^^^^ in Rust 2018, this closure would implement `Send` as `fptr` implements `Send`, but in Rust 2021, this closure would no longer implement `Send` as `fptr.0` does not implement `Send`
...
LL | *fptr.0 = 20;
| ------- in Rust 2018, closure captures all of `fptr`, but in Rust 2021, it only captures `fptr.0`
```
Auto merge of #86416 - Amanieu:asm_clobber_only, r=nagisa
Add clobber-only register classes for asm!
These are needed to properly express a function call ABI using a clobber
list, even though we don't support passing actual values into/out of
these registers.
Auto merge of #85953 - inquisitivecrystal:weak-linkat-in-fs-hardlink, r=joshtriplett
Fix linker error
Currently, `fs::hard_link` determines whether platforms have `linkat` based on the OS, and uses `link` if they don't. However, this heuristic does not work well if a platform provides `linkat` on newer versions but not on older ones. On old MacOS, this currently causes a linking error.
This commit fixes `fs::hard_link` by telling it to use `weak!` on macOS. This means that, on that operating system, we now check for `linkat` at runtime and use `link` if it is not available.
rustc_expand: Simplify span quoting in proc macro server
- The `Rustc::expn_id` field kept redundant information
- `SyntaxContext` is no longer thrown away before `save_proc_macro_span` because it's thrown away during metadata encoding anyway
Because specifying these flags multiple times will never be discernibly different in functionality from specifying them a single time, there is no reason to fail and report an error to the user.
Michael Howell [Sat, 10 Jul 2021 20:16:16 +0000 (13:16 -0700)]
cleanup(rustdoc): remove unused function getObjectNameById
This function was used in an earlier version, when idx's were used
to serialize function inputs and outputs. That's not done any more,
so removed the JS-side support for it.
On old macos systems, `fs::hard_link()` will follow symlinks.
This changes the test `symlink_hard_link` to exit early on
these systems, so that tests can pass.
`weak!` is needed in a test in another module. With macros
1.0, importing `weak!` would require reordering module
declarations in `std/src/lib.rs`, which is a bit too
evil.
Auto merge of #86873 - nikic:opaque-ptrs, r=nagisa
Improve opaque pointers support
Opaque pointers are coming, and rustc is not ready.
This adds partial support by passing an explicit load type to LLVM. Two issues I've encountered:
* The necessary type was not available at the point where non-temporal copies were generated. I've pushed the code for that upwards out of the memcpy implementation and moved the position of a cast to make do with the types we have available. (I'm not sure that cast is needed at all, but have retained it in the interest of conservativeness.)
* The `PlaceRef::project_deref()` function used during debuginfo generation seems to be buggy in some way -- though I haven't figured out specifically what it does wrong. Replacing it with `load_operand().deref()` did the trick, but I don't really know what I'm doing here.
~~[The docs](https://doc.rust-lang.org/stable/nightly-rustc/rustc_mir/transform/inline/struct.Inliner.html#method.check_codegen_attributes) seem to indicate rust thinks this function is generic over the lifetime `'satic`~~ This is because of `in_band_lifetimes`
Rollup merge of #87020 - RalfJung:const_raw_ptr_to_usize_cast, r=oli-obk
remove const_raw_ptr_to_usize_cast feature
This feature currently has the strange status of "const-only `unsafe`", which was an experiment that we no longer think is a good idea. We need to find better ways to enable things like "messing with the low bits of a pointer" during CTFE.
Rollup merge of #87013 - FabianWolff:issue-83921, r=estebank
Fix several ICEs related to malformed `#[repr(...)]` attributes
This PR fixes #83921. #83921 actually contains two related but distinct issues (one of them incorrectly reported as a duplicate in https://github.com/rust-lang/rust/issues/83921#issuecomment-814640734):
In the first, a call to `delay_span_bug` leads to an ICE when compiling with `-Zunpretty=everybody_loops` (and some other pretty-printing modes), because the corresponding error is emitted in a later pass, which does not run when only pretty-printing is requested.
The second issue is about parsing `#[repr(...)]` attributes. Currently, all of the following cause an ICE when applied to a struct/enum:
```rust
#[repr(packed())]
#[repr(align)]
#[repr(align(2, 4))]
#[repr(align())]
#[repr(i8())]
#[repr(u32(42))]
#[repr(i64 = 2)]
```
I have fixed this by expanding the well-formedness checks in `find_repr_attrs()`.
Amanieu d'Antras [Thu, 17 Jun 2021 20:00:52 +0000 (21:00 +0100)]
Add clobber-only register classes for asm!
These are needed to properly express a function call ABI using a clobber
list, even though we don't support passing actual values into/out of
these registers.
Auto merge of #81360 - Aaron1011:trait-caller-loc, r=nagisa
Support forwarding caller location through trait object method call
Since PR #69251, the `#[track_caller]` attribute has been supported on
traits. However, it only has an effect on direct (monomorphized) method
calls. Calling a `#[track_caller]` method on a trait object will *not*
propagate caller location information - instead, `Location::caller()` will
return the location of the method definition.
This PR forwards caller location information when `#[track_caller]` is
present on the method definition in the trait. This is possible because
`#[track_caller]` in this position is 'inherited' by any impls of that
trait, so all implementations will have the same ABI.
This PR does *not* change the behavior in the case where
`#[track_caller]` is present only on the impl of a trait.
While all implementations of the method might have an explicit
`#[track_caller]`, we cannot know this at codegen time, since other
crates may have impls of the trait. Therefore, we keep the current
behavior of not forwarding the caller location, ensuring that all
implementations of the trait will have the correct ABI.
See the modified test for examples of how this works
Auto merge of #86987 - lcnr:const-default-eval-bound, r=oli-obk
only check cg defaults wf once instantiated
the previous fixmes here didn't make too much sense as I didn't yet fully understand the code further below.
That code only runs if the predicates using our generic param default are fully concrete after substituting our default, which never happens if our default is generic.
This code is the source of #57569. The problem is subtle, so let me point it out. This code makes the mistake of assuming that all of the macros in `krate.exported_macros` are exported.
...Yeah. For some historical reason, all `macro` macros are marked as exported, regardless of whether they actually are, which is dreadfully confusing. It would be more accurate to say that `exported_macros` currently contains only macros that have paths.
This PR renames `exported_macros` to `importable_macros`, since these macros can be imported with `use` while others cannot. It also fixes the code above to no longer lint on private `macro` macros, since the `missing_docs` lint should only appear on exported items.
Auto merge of #86419 - ricobbe:raw-dylib-stdcall, r=petrochenkov
Add support for raw-dylib with stdcall, fastcall functions
Next stage of work for #58713: allow `extern "stdcall"` and `extern "fastcall"` with `#[link(kind = "raw-dylib")]`.
I've deliberately omitted support for vectorcall, as that doesn't currently work, and I wanted to get this out for review. (I haven't really investigated the vectorcall failure much yet, but at first (very cursory) glance it appears that the problem is elsewhere.)
Don't access pointer element type for nontemporal store
Simply shift the bitcast from the store to the load, so that
we can use the destination type. I'm not sure the bitcast is
really necessary, but keeping it for now.
Auto merge of #86888 - FabianWolff:issue-86600, r=davidtwco
Fix double warning about illegal floating-point literal pattern
This PR fixes #86600. The problem is that the `ConstToPat` struct contains a field `include_lint_checks`, which determines whether lints should be emitted or not, but this field is currently not obeyed at one point, leading to a warning being emitted more than once. I have fixed this behavior here.