* deps: bump wasmparser to 0.39.2
This has a bug fix for multi-value Wasm validation that is required for getting
the spec tests passing.
https://github.com/yurydelendik/wasmparser.rs/pull/135
* Update cranelift to 0.46.1 to get multi-value Wasm support
The `cranelift_wasm` APIs had to change a little bit to maintain state necessary
when translating multi-value Wasm blocks. The `translate_module` function now
returns a `ModuleTranslationState` that is borrowed during each function's
translation.
* Enable multi-value proposal's spec tests
This enables all the Wasm multi-value proposal's spec tests other than the ones
that rely on functions having more return values than registers available on the
target. That is not supported by cranelift yet.
* wasmtime-interface-types: always use multi-value Wasm
And remove the return pointer hacks that work around the lack of multi-value.
This situation could be triggered that can_add_var would return true
while a variable was already added for the given register.
For instance, when we have a reassignment (because of a fixed register
input requirement) and a fixed input conflict on the same fixed
register, this register will not be available in the regs_in set after
inputs_done (because of the fixed input conflict diversion) but will
have its own variable.
* Fixes `path_symlink_trailing_slashes` test case
This commit:
* adds a couple `log::debug!` macro calls in and around `path_get`
for easier future debugging
* changes impl of `path_symlink` hostcall to actually *require*
the final component (matching the impl of WASI in C)
* ignores the error `__WASI_ENOTDIR` in `path_get`'s `readlinkat` call
which is not meant to be an error at this stage (i.e., this
potentially erroneous condition *will be* handled later, in
one of the layers above)
* Fixes `path_symlink_trailing` slashes on BSD-nixes
This commit:
* makes `path_symlink` host-specific (Linux and BSD-like nixes
now have their own differing implementations)
* on BSD-like nixes, when `ENOTDIR` is returned from `symlinkat`
it checks whether the target path contains a trailing slash,
strips it, and then checks if the target path without the trailing
slash exists; if yes, then converts the error code to `EEXIST` to
match Linux/POSIX spec
This commit moves a couple of things around:
* separates the logic of `path_unlink_file` into separate impls
for linux and BSD-style nixes
* moves implementation consts into appropriate impl modules: linux
or bsd
* cleans up `utime_now` and `utime_omit` for BSD-style nixes
This commit fixes an issue with incorrect handling of /dev/(u)random
on Linux. It turns out that `nix::unistd::isatty` call handled only
the POSIX spec case where `ENOTTY` is returned in case the passed
in file descriptor is OK but not a TTY, whereas on Linux this is not
always the case. On Linux, it can be the case that `EINVAL` is returned
instead and this case AFAIK is not handled by the `nix` crate. This
commit fixes this by using `libc::isatty` syscall directly and checking
the return values.
Only the shifts with applicable SSE2 instructions are implemented here: PSRL* (for ushr) only has 16-64 bit instructions and PSRA* (for sshr) only has 16-32 bit instructions.
Previously, ConstantData was a type alias for `Vec<u8>` which prevented it from having an implementation; this meant that `V128Imm` and `&[u8; 16]` were used in places that otherwise could have accepted types of different byte lengths.
We weren't previously keeping track of quite the right information for whether
an `if .. else .. end`'s following block was reachable or not. It should be
reachable if the head is reachable and either the consequent or alternative end
reachable (and therefore fall through to the following block) or do an early
`br_if` to it.
This commit rejiggers `ControlStackFrame::If` to keep track of reachability at
the end of the consequent (we don't need to keep track of it at the end of the
alternative, since that is simply `state.reachable`) and adds Wasm tests for
every reachability situation we can encounter with `if .. else .. end`.
Fixes#1132
This allows replacing a function that has N results with N instructions
with the same result type. It also narrows down typing, so that
instructions creating F32/F64 values are replaced with a constant of the
correct type.
It has started breaking in ways unrelated to Cranelift, making it hard
to spot real CI failures in Cranelift. We should re-enable it at some
point, but disable it in the meanwhile.
This avoids doing multiple unpacking of the InstructionData for a single
legalization, improving readability and reducing size of the generated
code. For instance, icmp had to unpack the format once per IntCC
condition code.
* cranelift-wasm: replace `WasmTypesMap` with `ModuleTranslationState`
The `ModuleTranslationState` contains information decoded from the Wasm module
that must be referenced during each Wasm function's translation.
This is only for data that is maintained by `cranelift-wasm` itself, as opposed
to being maintained by the embedder. Data that is maintained by the embedder is
represented with `ModuleEnvironment`.
A `ModuleTranslationState` is returned by `translate_module`, and can then be
used when translating functions from that module.
* cranelift-wasm: rename `TranslationState` to `FuncTranslationState`
To disambiguate a bit with the new `ModuleTranslationState`.
* cranelift-wasm: Reorganize the internal `state` module into submodules
One module for the `ModuleTranslationState` and another for the
`FuncTranslationState`.
* cranelift-wasm: replace `FuncTranslator` with methods on `ModuleTranslationState`
`FuncTranslator` was two methods that always took ownership of `self`, so it
didn't really make sense as an object as opposed to two different functions, or
in this case methods on the object that actually persists for a longer time.
I think this improves ergonomics nicely.
Before:
```rust
let module_translation = translate_module(...)?;
for body in func_bodies {
let mut translator = FuncTranslator::new();
translator.translate(body, ...)?;
}
```
After:
```rust
let module_translation = translate_module(...)?;
for body in func_bodies {
module_translation.translate_func(body, ...)?;
}
```
Note that this commit does not remove `FuncTranslator`. It still exists, but is
just a wrapper over the `ModuleTranslationState` methods, and it is marked
deprecated, so that downstream users get a heads up. This should make the
transition easier.
* Revert "cranelift-wasm: replace `FuncTranslator` with methods on `ModuleTranslationState`"
This reverts commit 075f9ae933bcaae39348b61287c8f78a4009340d.
This adds a `DummyConstant` structure that is converted to something like `let const0 = pos.func.dfg.constants.insert(...)` in `gen_legalizer.rs`. This allows us to create constants during legalization with something like `let ones = constant(vec![0xff; 16])` and then use `ones` within a `def!` block, e.g.: `def!(a = vconst(ones))`. One unfortunate side-effect of this change is that, because the names of the constants in `ConstPool` are dynamic, the `VarPool` and `SymbolTable` structures that previously operated on `&'static str` types now must operate on `String` types; however, since this is a change to the meta code-generation, it should result in no runtime performance impact.
