Mangle symbols with a mangled name close to PDB limits with v0 instead of legacy mangling to avoid linker errors
This is rust-lang/compiler-team#934
As PDB debuginfo has a 64KiB limit for symbol names, we use v0 mangling instead of legacy mangling for symbol names >= 65000 bytes if PDB is used. The cutoff number was chosen to leave some room for potential errors in the empirical measurement of the limit of 65521 bytes, as well as potential symbol prefixes and suffixes that are applied later, plus some generous extra space.
Tracking issue: rust-lang/rust#148429
This commit fixes an accidental regression from 144678 where wasm
targets would now accidentally use the wrong import module map for a
symbol causing a symbol to skip mangling. This can result in compilation
failures when symbols are used in cross-crate situations.
Closes 148347
Add `__isPlatformVersionAtLeast` and `__isOSVersionAtLeast` symbols
## Motivation
When Objective-C code uses ```@available(...)`,`` Clang inserts a call to [`__isPlatformVersionAtLeast`](https://github.com/llvm/llvm-project/blob/llvmorg-20.1.0/compiler-rt/lib/builtins/os_version_check.c#L276) (`__isOSVersionAtLeast` in older Clang versions). These symbols not being available sometimes ends up causing linker errors. See the new test `tests/run-make/apple-c-available-links` for a minimal reproducer.
The workaround is to link `libclang_rt.osx.a`, see e.g. https://github.com/alexcrichton/curl-rust/issues/279. But that's very difficult for users to figure out (and the backreferences to that issue indicates that people are still running into this in their own projects every so often).
For another recent example, this is preventing `rustc` from using LLVM assertions on macOS, see https://github.com/rust-lang/rust/pull/62592#issuecomment-510670657 and https://github.com/rust-lang/rust/pull/134275#issuecomment-2543067830.
It is also a blocker for [setting the correct minimum OS version in `cc-rs`](https://github.com/rust-lang/rust/issues/136113), since fixing this in `cc-rs` might end up introducing linker errors in places where we weren't before (by default, if using e.g. ```@available(macos`` 10.15, *)`, the symbol usually happens to be left out, since `clang` defaults to compiling for the host macOS version, and thus things _seem_ to work - but the availability check actually compiles down to nothing, which is a huge correctness footgun for running on older OSes).
(My super secret evil agenda is also to expose some variant of ```@available``` in Rust's `std` after https://github.com/rust-lang/rfcs/pull/3750 progresses further, will probably file an ACP for this later. But I believe this PR has value regardless of those future plans, since we'd be making C/Objective-C/Swift interop easier).
## Solution
Implement `__isPlatformVersionAtLeast` and `__isOSVersionAtLeast` as part of the "public ABI" that `std` exposes.
**This is insta-stable**, in the same sense that additions to `compiler-builtins` are insta-stable, though the availability of these symbols can probably be considered a "quality of implementation" detail rather than a stable promise.
I originally proposed to implement this in `compiler-builtins`, see https://github.com/rust-lang/compiler-builtins/pull/794, but we discussed moving it to `std` instead ([Zulip thread](https://rust-lang.zulipchat.com/#narrow/channel/219381-t-libs/topic/Provide.20.60__isPlatformVersionAtLeast.60.20in.20.60std.60.3F/with/507880717)), which makes the implementation substantially simpler, and we avoid gnarly issues with requiring the user to link `libSystem.dylib` (since `std` unconditionally does that).
Note that this does not solve the linker errors for (pure) `#![no_std]` users, but that's _probably_ fine, if you are using ```@available``` to test the OS version on Apple platforms, you're likely also using `std` (and it is still possible to work around by linking `libclang_rt.*.a`).
A thing to note about the implementation, I've choosen to stray a bit from LLVM's upstream implementation, and not use `_availability_version_check` since [it has problems when compiling with an older SDK](https://github.com/llvm/llvm-project/issues/64227). Instead, we use `sysctl kern.osproductversion` when available to still avoid the costly PList lookup in most cases, but still with a fall back to the PList lookup when that is not available (with the PList fallback being is similar to LLVM's implementation).
## Testing
Apple has a lot of different "modes" that they can run binaries in, which can be a bit difficult to find your bearings in, but I've tried to be as thorough as I could in testing them all.
Tested using roughly the equivalent of `./x test library/std -- platform_version` on the following configurations:
- macOS 14.7.3 on a Macbook Pro M2
- `aarch64-apple-darwin`
- `x86_64-apple-darwin` (under Rosetta)
- `aarch64-apple-ios-macabi`
- `x86_64-apple-ios-macabi` (under Rosetta)
- `aarch64-apple-ios` (using Xcode's "Designed for iPad" setting)
- `aarch64-apple-ios-sim` (in iOS Simulator, as iPhone with iOS 17.5)
- `aarch64-apple-ios-sim` (in iOS Simulator, as iPad with iOS 18.2)
- `aarch64-apple-tvos-sim` (in tvOS Simulator)
- `aarch64-apple-watchos-sim` (in watchOS Simulator)
- `aarch64-apple-ios-sim` (in visionOS simulator, using Xcode's "Designed for iPad" setting)
- `aarch64-apple-visionos-sim` (in visionOS Simulator)
- macOS 15.3.1 VM
- `aarch64-apple-darwin`
- `aarch64-apple-ios-macabi`
- macOS 10.12.6 on an Intel Macbook from 2013
- `x86_64-apple-darwin`
- `i686-apple-darwin`
- `x86_64-apple-ios` (in iOS Simulator)
- iOS 9.3.6 on a 1st generation iPad Mini
- `armv7-apple-ios` with an older compiler
Along with manually inspecting the output of `version_from_sysctl()` and `version_from_plist()`, and verifying that they actually match what's expected.
I believe the only real omissions here would be:
- `aarch64-apple-ios` on a newer iPhone that has `sysctl` available (iOS 11.4 or above).
- `aarch64-apple-ios` on a Vision Pro using Xcode's "Designed for iPad" setting.
But I don't have the hardware available to test those.
``@rustbot`` label O-apple A-linkage -T-compiler -A-meta -A-run-make
try-job: aarch64-apple
This was done in #145740 and #145947. It is causing problems for people
using r-a on anything that uses the rustc-dev rustup package, e.g. Miri,
clippy.
This repository has lots of submodules and subtrees and various
different projects are carved out of pieces of it. It seems like
`[workspace.dependencies]` will just be more trouble than it's worth.
Couple of codegen_fn_attrs improvements
As noted in https://github.com/rust-lang/rust/pull/144678#discussion_r2245060329 here is no need to keep link_name and export_name separate, which the third commit fixes by merging them. The second commit removes some dead code and the first commit merges two ifs with equivalent conditions. The last commit is an unrelated change which removes an unused `feature(autodiff)`.
Currently they are skipped, which is a bit weird, and it sometimes
causes malformed output like `Foo<>` and `dyn Bar<, A = u32>`.
Most regions are erased by the time `type_name` does its work. So all
regions are now printed as `'_` in non-optional places. Not perfect, but
better than the status quo.
`c_name` is updated to trim lifetimes from MIR pass names, so that the
`PASS_NAMES` sanity check still works. It is also renamed as
`simplify_pass_type_name` and made non-const, because it doesn't need
to be const and the non-const implementation is much shorter.
The commit also renames `should_print_region` as
`should_print_optional_region`, which makes it clearer that it only
applies to some regions.
Fixes#145168.
Broken by #144776; this is reachable after all.
Fixes#144994.
The commit also adds a lot more cases to the `type-name-basic.rs`,
because it's currently very anaemic. This includes some cases where
region omission does very badly; these are marked with FIXME.
Three of them are named `AbsolutePathPrinter`, which is confusing, so
give those names that better indicate how they are used. And then there
is `SymbolPrinter` and `SymbolMangler`, which are renamed as
`LegacySymbolMangler` and `V0SymbolMangler`, better indicating their
similarity.
`Bound`/`Placeholder`/`Infer`/`Error` shouldn't occur, so we can handle
them in the second exhaustive `match`, and ignore them in the first
non-exhaustive `match`.
Currently they are mostly named `cx`, which is a terrible name for a
type that impls `Printer`/`PrettyPrinter`, and is easy to confuse with
other types like `TyCtxt`. This commit changes them to `p`. A couple of
existing `p` variables had to be renamed to make way.
This helps me understand the structure of the code a lot.
If any of these are actually reachable, we can put the old code back,
add a new test case, and we will have improved our test coverage.
