Stop emitting the `dso_local` LLVM attribute for external symbols under the static relocation model on macOS.
This matches Clang's behavior:
973cb2c326/clang/lib/CodeGen/CodeGenModule.cpp (L1038-L1040)
Even if `dso_local` were properly supported in this way on macOS, it seems
incorrect to add this annotation as liberally as we did. The `dso_local`
annotation is for symbols that ultimately end up in the same linkage unit, but
we were adding this annotation even for `static` values inside `extern` blocks
marked with `#[link(type="framework")]`, which should be considered dynamically
linked. Note that Clang likewise avoids emitting `dso_local` for `dllimport`
symbols:
973cb2c326/clang/lib/CodeGen/CodeGenModule.cpp (L1005-L1007)
This issue caused breakage in the `ring` crate, which links to a symbol defined
in `Security.framework` that ultimately resolves to address `0x0`:
b94d61e044/src/rand.rs (L390)
For this symbol, the use of `dso_local` causes LLVM to emit a relocation of
type `X86_64_RELOC_SIGNED`, which is a 32-bit signed PC-relative offset. If the
binary is large enough, `0x0` might be out of range, and the link will fail.
Avoiding `dso_local` causes LLVM to use the GOT instead, emitting a relocation
of type `X86_64_RELOC_GOT_LOAD`, which will properly handle the large offset
and cause the link to succeed.
As a side note, the static relocation model is effectively deprecated for
security reasons on macOS, as it prohibits PIE. It's also completely
unsupported on Apple Silicon, so I don't think it's worth going to the effort
of properly supporting this model on that platform.
Rollup of 7 pull requests
Successful merges:
- #86747 (Improve wording of the `drop_bounds` lint)
- #87166 (Show discriminant before overflow in diagnostic for duplicate values.)
- #88077 (Generate an iOS LLVM target with a specific version)
- #88164 (PassWrapper: adapt for LLVM 14 changes)
- #88211 (cleanup: `Span::new` -> `Span::with_lo`)
- #88229 (Suggest importing the right kind of macro.)
- #88238 (Stop tracking namespace in used_imports.)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Stop tracking namespace in used_imports.
This changes `used_imports` from a `FxHashSet<(NodeId, Namespace)>` to a `FxHashSet<NodeId>`, as the Namespace information isn't used.
The only point that uses it did three lookups, `|=`'ing them together.
r? `@estebank`
PassWrapper: adapt for LLVM 14 changes
These API changes appear to have all taken place in
https://reviews.llvm.org/D105007, which moved HWAddressSanitizerPass and
AddressSanitizerPass to only accept their options type as a ctor
argument instead of the sequence of bools etc. This required a couple of
parameter additions, which I made match the default prior to the
mentioned upstream LLVM change.
This patch restores rustc to building (though not quite passing all
tests, I've mailed other patches for those issues) against LLVM HEAD.
Generate an iOS LLVM target with a specific version
This commit adds the `LC_VERSION_MIN_IPHONEOS` load command to the Mach-O header generated for `aarch64-apple-ios` binaries. The operating system will look for this load command to determine the minimum supported operating system version and will not allow the binary to run if it's absent. This logic already exists for the simulator toolchain.
I've been using `otool` from a [cctools](https://github.com/tpoechtrager/cctools-port) toolchain to parse the header and validate that this change adds the required load command.
This change appears to be enough to build Rust binaries that can run on a jailbroken iPhone.
Improve wording of the `drop_bounds` lint
This PR addresses #86653. The issue is sort of a false positive of the `drop_bounds` lint, but I would argue that the best solution for #86653 is simply a rewording of the warning message and lint description, because even if the lint is _technically_ wrong, it still forces the programmer to think about what they are doing, and they can always use `#[allow(drop_bounds)]` if they think that they really need the `Drop` bound.
There are two issues with the current warning message and lint description:
- First, it says that `Drop` bounds are "useless", which is technically incorrect because they actually do have the effect of allowing you e.g. to call methods that also have a `Drop` bound on their generic arguments for some reason. I have changed the wording to emphasize not that the bound is "useless", but that it is most likely not what was intended.
- Second, it claims that `std::mem::needs_drop` detects whether a type has a destructor. But I think this is also technically wrong: The `Drop` bound says whether the type has a destructor or not, whereas `std::mem::needs_drop` also takes nested types with destructors into account, even if the top-level type does not itself have one (although I'm not 100% sure about the exact terminology here, i.e. whether the "drop glue" of the top-level type counts as a destructor or not).
cc `@jonhoo,` does this solve the issue for you?
r? `@GuillaumeGomez`
canonicalize consts before calling try_unify_abstract_consts query
Fixes#88022Fixes#86953Fixes#77708Fixes#82034Fixes#85031
these ICEs were all caused by calling the `try_unify_abstract_consts` query with inference vars in substs
r? `@lcnr`
marker_traits: require `EvaluatedToOk` during winnowing
closes#84955, while it doesn't really fix it in a way that makes me happy it should prevent the issue for now and this
test can't be reproduced anyways, so it doesn't make much sense to keep it open.
fixes#84917 as only one of the impls depends on regions, so we now drop the ambiguous one instead of the correct one.
cc https://rust-lang.zulipchat.com/#narrow/stream/144729-wg-traits/topic/winnowing.20soundly/near/247899832
r? `@nikomatsakis`
Instead, avoid registering the problematic well-formed obligation
to begin with. This removes global untracked mutable state,
and avoids potential issues with incremental compilation.
Trait upcasting coercion (part 3)
By using separate candidates for each possible choice, this fixes type-checking issues in previous commits.
r? `@nikomatsakis`
Instead of updating global state to mark attributes as used,
we now explicitly emit a warning when an attribute is used in
an unsupported position. As a side effect, we are to emit more
detailed warning messages (instead of just a generic "unused" message).
`Session.check_name` is removed, since its only purpose was to mark
the attribute as used. All of the callers are modified to use
`Attribute.has_name`
Additionally, `AttributeType::AssumedUsed` is removed - an 'assumed
used' attribute is implemented by simply not performing any checks
in `CheckAttrVisitor` for a particular attribute.
We no longer emit unused attribute warnings for the `#[rustc_dummy]`
attribute - it's an internal attribute used for tests, so it doesn't
mark sense to treat it as 'unused'.
With this commit, a large source of global untracked state is removed.
Force warn improvements
As part of stablization of the `--force-warn` option (#86516) I've made the following changes:
* Error when the `warnings` lint group is based to the `--force-warn` option
* Tests have been updated to make it easier to understand the semantics of `--force-warn`
r? `@nikomatsakis`
Upgrade to LLVM 13
Work in progress update to LLVM 13. Main changes:
* InlineAsm diagnostics reported using SrcMgr diagnostic kind are now handled. Previously these used a separate diag handler.
* Codegen tests are updated for additional attributes.
* Some data layouts have changed.
* Switch `#[used]` attribute from `llvm.used` to `llvm.compiler.used` to avoid SHF_GNU_RETAIN flag introduced in https://reviews.llvm.org/D97448, which appears to trigger a bug in older versions of gold.
* Set `LLVM_INCLUDE_TESTS=OFF` to avoid Python 3.6 requirement.
Upstream issues:
* ~~https://bugs.llvm.org/show_bug.cgi?id=51210 (InlineAsm diagnostic reporting for module asm)~~ Fixed by 1558bb80c0.
* ~~https://bugs.llvm.org/show_bug.cgi?id=51476 (Miscompile on AArch64 due to incorrect comparison elimination)~~ Fixed by 81b106584f.
* https://bugs.llvm.org/show_bug.cgi?id=51207 (Can't set custom section flags anymore). Problematic change reverted in our fork, https://reviews.llvm.org/D107216 posted for upstream revert.
* https://bugs.llvm.org/show_bug.cgi?id=51211 (Regression in codegen for #83623). This is an optimization regression that we may likely have to eat for this release. The fix for #83623 was based on an incorrect premise, and this needs to be properly addressed in the MergeICmps pass.
The [compile-time impact](https://perf.rust-lang.org/compare.html?start=ef9549b6c0efb7525c9b012148689c8d070f9bc0&end=0983094463497eec22d550dad25576a894687002) is mixed, but quite positive as LLVM upgrades go.
The LLVM 13 final release is scheduled for Sep 21st. The current nightly is scheduled for stable release on Oct 21st.
r? `@ghost`
Refactor fallback code to prepare for never type
This PR contains cherry-picks of some of `@nikomatsakis's` work from #79366, and shouldn't (AFAICT) represent any change in behavior. However, the refactoring is good regardless of the never type work being landed, and will reduce the size of those eventual PR(s) (and rebase pain).
I am not personally an expert on this code, and the commits are essentially 100% `@nikomatsakis's,` but they do seem reasonable to me by my understanding. Happy to edit with review, of course. Commits are best reviewed in sequence rather than all together.
r? `@jackh726` perhaps?
This matches Clang's behavior:
973cb2c326/clang/lib/CodeGen/CodeGenModule.cpp (L1038-L1040)
Even if `dso_local` were properly supported in this way on macOS, it seems
incorrect to add this annotation as liberally as we did. The `dso_local`
annotation is for symbols that ultimately end up in the same linkage unit, but
we were adding this annotation even for `static` values inside `extern` blocks
marked with `#[link(type="framework")]`, which should be considered dynamically
linked. Note that Clang likewise avoids emitting `dso_local` for `dllimport`
symbols:
973cb2c326/clang/lib/CodeGen/CodeGenModule.cpp (L1005-L1007)
This issue caused breakage in the `ring` crate, which links to a symbol defined
in `Security.framework` that ultimately resolves to address `0x0`:
b94d61e044/src/rand.rs (L390)
For this symbol, the use of `dso_local` causes LLVM to emit a relocation of
type `X86_64_RELOC_SIGNED`, which is a 32-bit signed PC-relative offset. If the
binary is large enough, `0x0` might be out of range, and the link will fail.
Avoiding `dso_local` causes LLVM to use the GOT instead, emitting a relocation
of type `X86_64_RELOC_GOT_LOAD`, which will properly handle the large offset
and cause the link to succeed.
As a side note, the static relocation model is effectively deprecated for
security reasons on macOS, as it prohibits PIE. It's also completely
unsupported on Apple Silicon, so I don't think it's worth going to the effort
of properly supporting this model on that platform.
Motivation: in upcoming commits, we are going to create a graph of the
coercion relationships between variables. We want to
distinguish *coercion* specifically from other sorts of subtyping, as
it indicates values flowing from one place to another via assignment.
I didn't like the sub-unify code executing when a predicate was
ENQUEUED, that felt fragile. I would have preferred to move the
sub-unify code so that it only occurred during generalization, but
that impacted diagnostics, so having it also occur when we process
subtype predicates felt pretty reasonable. (I guess we only need one
or the other, but I kind of prefer both, since the generalizer
ultimately feels like the *right* place to guarantee the properties we
want.)
RFC2229 Only compute place if upvars can be resolved
Closes https://github.com/rust-lang/rust/issues/87987
This PR fixes an ICE when trying to unwrap an Err. This error appears when trying to convert a PlaceBuilder into Place when upvars can't yet be resolved. We should only try to convert a PlaceBuilder into Place if upvars can be resolved.
r? `@nikomatsakis`
RFC2229 Add missing edge case
Closes https://github.com/rust-lang/rust/issues/87988
This PR fixes an ICE where a match discriminant is not being read when expected. This ICE was the result of a missing edge case which assumed that if a pattern is of type `PatKind::TupleStruct(..) | PatKind::Path(..) | PatKind::Struct(..) | PatKind::Tuple(..)` then a place could only be a multi variant if the place is of type kind Adt.
