This commit updates how rustc compiler metadata is stored in rlibs.
Previously metadata was stored as a raw file that has the same format as
`--emit metadata`. After this commit, however, the metadata is encoded
into a small object file which has one section which is the contents of
the metadata.
The motivation for this commit is to fix a common case where #83730
arises. The problem is that when rustc crates a `dylib` crate type it
needs to include entire rlib files into the dylib, so it passes
`--whole-archive` (or the equivalent) to the linker. The problem with
this, though, is that the linker will attempt to read all files in the
archive. If the metadata file were left as-is (today) then the linker
would generate an error saying it can't read the file. The previous
solution was to alter the rlib just before linking, creating a new
archive in a temporary directory which has the metadata file removed.
This problem from before this commit is now removed if the metadata file
is stored in an object file that the linker can read. The only caveat we
have to take care of is to ensure that the linker never actually
includes the contents of the object file into the final output. We apply
similar tricks as the `.llvmbc` bytecode sections to do this.
This involved changing the metadata loading code a bit, namely updating
some of the LLVM C APIs used to use non-deprecated ones and fiddling
with the lifetimes a bit to get everything to work out. Otherwise though
this isn't intended to be a functional change really, only that metadata
is stored differently in archives now.
This should end up fixing #83730 because by default dylibs will no
longer have their rlib dependencies "altered" meaning that
split-debuginfo will continue to have valid paths pointing at the
original rlibs. (note that we still "alter" rlibs if LTO is enabled to
remove Rust object files and we also "alter" for the #[link(cfg)]
feature, but that's rarely used).
Closes#83730
Use pattern matching instead of checking lengths explicitly
This piece of code checks that there are exaclty two variants, one having
exactly one field, the other having exactly zero fields. If any of these
conditions is violated, it returns `None`. Otherwise it assigns that one
field's ty to `field_ty`.
Instead of fiddling with indices and length checks explicitly, use pattern
matching to simplify this.
`@rustbot` modify labels +C-cleanup +T-compiler
Turn off frame pointer elimination on all Apple platforms.
This ends up disabling frame pointer elimination on aarch64_apple_darwin
which matches what clang does by default along with the
aarch64_apple_ios and x86_64_apple_darwin targets.
Further, the Apple docs "Writing ARM64 Code for Apple Platforms" has a section
called "Respect the Purpose of Specific CPU Registers" which
specifically calls out the frame pointer register (x29):
The frame pointer register (x29) must always address a valid frame
record. Some functions — such as leaf functions or tail calls — may
opt not to create an entry in this list As a result, stack traces
are always meaningful, even without debug information.
Other platforms are updated to not override the default.
rustc: Allow safe #[target_feature] on wasm
This commit updates the compiler's handling of the `#[target_feature]`
attribute when applied to functions on WebAssembly-based targets. The
compiler in general requires that any functions with `#[target_feature]`
are marked as `unsafe` as well, but this commit relaxes the restriction
for WebAssembly targets where the attribute can be applied to safe
functions as well.
The reason this is done is that the motivation for this feature of the
compiler is not applicable for WebAssembly targets. In general the
`#[target_feature]` attribute is used to enhance target CPU features
enabled beyond the basic level for the rest of the compilation. If done
improperly this means that your program could execute an instruction
that the CPU you happen to be running on does not understand. This is
considered undefined behavior where it is unknown what will happen (e.g.
it's not a deterministic `SIGILL`).
For WebAssembly, however, the target is different. It is not possible
for a running WebAssembly program to execute an instruction that the
engine does not understand. If this were the case then the program would
not have validated in the first place and would not run at all. Even if
this were allowed in some hypothetical future where engines have some
form of runtime feature detection (which they do not right now) any
implementation of such a feature would generate a trap if a module
attempts to execute an instruction the module does not understand. This
deterministic trap behavior would still not fall into the category of
undefined behavior because the trap is deterministic.
For these reasons the `#[target_feature]` attribute is now allowed on
safe functions, but only for WebAssembly targets. This notably enables
the wasm-SIMD intrinsics proposed for stabilization in #74372 to be
marked as safe generally instead of today where they're all `unsafe` due
to the historical implementation of `#[target_feature]` in the compiler.
This ends up disabling frame pointer elimination on aarch64_apple_darwin
which matches what clang does by default along with the
aarch64_apple_ios and x86_64_apple_darwin targets.
Further, the Apple docs "Writing ARM64 Code for Apple Platforms" has a section
called "Respect the Purpose of Specific CPU Registers" which
specifically calls out the frame pointer register (x29):
The frame pointer register (x29) must always address a valid frame
record. Some functions — such as leaf functions or tail calls — may
opt not to create an entry in this list As a result, stack traces
are always meaningful, even without debug information.
Other platforms are updated to not override the default.
Avoid creating anonymous nodes with zero or one dependency.
Anonymous nodes are only useful to encode dependencies, and cannot be replayed from one compilation session to another.
As such, anonymous nodes without dependency are always green.
Anonymous nodes with only one dependency are equivalent to this dependency.
cc #45408
cc `@michaelwoerister`
This piece of code checks that there are exaclty two variants, one having
exactly one field, the other having exactly zero fields. If any of these
conditions is violated, it returns `None`. Otherwise it assigns that one
field's ty to `field_ty`.
Instead of fiddling with indices and length checks explicitly, use pattern
matching to simplify this.
Reduce the amount of untracked state in TyCtxt
Access to untracked global state may generate instances of #84970.
The GlobalCtxt contains the lowered HIR, the resolver outputs and interners.
By wrapping the resolver inside a query, we make sure those accesses are properly tracked.
As a no_hash query, all dependent queries essentially become `eval_always`,
what they should have been from the beginning.
Don't sort a `Vec` before computing its `DepTrackingHash`
Previously, we sorted the vec prior to hashing, making the hash
independent of the original (command-line argument) order. However, the
original vec was still always kept in the original order, so we were
relying on the rest of the compiler always working with it in an
'order-independent' way.
This assumption was not being upheld by the `native_libraries` query -
the order of the entires in its result depends on the order of entries
in `Options.libs`. This lead to an 'unstable fingerprint' ICE when the
`-l` arguments were re-ordered.
This PR removes the sorting logic entirely. Re-ordering command-line
arguments (without adding/removing/changing any arguments) seems like a
really niche use case, and correctly optimizing for it would require
additional work. By always hashing arguments in their original order, we
can entirely avoid a cause of 'unstable fingerprint' errors.
Emit a hard error when a panic occurs during const-eval
Previous, a panic during const evaluation would go through the
`const_err` lint. This PR ensures that such a panic always causes
compilation to fail.
Fix span of redundant generic arguments
Fixes#71563
Above issue is about lifetime arguments, but generic arguments also have same problem.
This PR fixes both help messages.
Use correct edition when parsing `:pat` matchers
As described in issue #85708, we currently do not properly decode
`SyntaxContext::root()` and `ExpnId::root()` from foreign crates. As a
result, when we decode a span from a foreign crate with
`SyntaxContext::root()`, we end up up considering it to have the edition
of the *current* crate, instead of the foreign crate where it was
originally created.
A full fix for this issue will be a fairly significant undertaking.
Fortunately, it's possible to implement a partial fix, which gives us
the correct edition-dependent behavior for `:pat` matchers when the
macro is loaded from another crate. Since we have the edition of the
macro's defining crate available, we can 'recover' from seeing a
`SyntaxContext::root()` and use the edition of the macro's defining
crate.
Any solution to issue #85708 must reproduce the behavior of this
targeted fix - properly preserving a foreign `SyntaxContext::root()`
means (among other things) preserving its edition, which by definition
is the edition of the foreign crate itself. Therefore, this fix moves us
closer to the correct overall solution, and does not expose any new
incorrect behavior to macros.
