* Use thiserror proc macros for auto From impls
This commit refactors `wasi_common::error::Error` by using `#[from]`
proc macro to autoderive `From` for wrapped errors.
* Back port changes to snapshot0
* Auto impl Display for WasiError
* Fix stack overflow when auto generating Display for WasiError
* Initial support for securing tty I/O.
* Update the tests.
* Fix warnings
* Update crates/wasi-common/src/fdentry.rs
Co-Authored-By: Jakub Konka <jakub.konka@golem.network>
* Properly sandbox stderr.
* Document why the scratch buffer is 4 elements long.
* Update crates/wasi-common/src/sandboxed_tty_writer.rs
Co-Authored-By: Jakub Konka <jakub.konka@golem.network>
* Update crates/wasi-common/src/sandboxed_tty_writer.rs
Co-Authored-By: Jakub Konka <jakub.konka@golem.network>
* Add comments explaining how we report the number of bytes written.
* Always sanitize stderr.
* Port the changes to the snapshot_0 directory.
* Fix snapshot_0 compilation error.
* Replace the scratch buffer with a temporary buffer.
* Update crates/wasi-common/src/sandboxed_tty_writer.rs
Co-Authored-By: bjorn3 <bjorn3@users.noreply.github.com>
* Format with latest stable rustfmt.
Co-authored-by: Jakub Konka <kubkon@jakubkonka.com>
Co-authored-by: bjorn3 <bjorn3@users.noreply.github.com>
* Add unimplemented stubs for Cranelift interfaces
Cranelift changes to FuncEnvironment, TargetEnvironment, and GlobalInit (see https://github.com/bytecodealliance/cranelift/pull/1073) require these changes to compile wasmtime.
* Upgrade Cranelift to 0.52.0
This commit updates _all_ WASI test programs to use the latest
version of the `wasi` crate (`v0.9.0`). While at it, it also
unifies asserting error conditions across all test programs.
As an implementation-specific limit, we do not allow the full index space of
`0..=2^32 - 1` because we reserve index `2^32 - 1` for ourselves in
`cranelift-entity`.
Fixes#1306
This commit disables the Emscripten CI job until a regression
in rustc for `wasm32-unknown-emscripten` target is fixed.
Tracking issue: [rust-lang/rust 66308](https://github.com/rust-lang/rust/issues/66308).
Our corpora are too large to run in full on every pull request, they just take
too long. Instead, we sample some of them and make sure that running our fuzzers
over the sampled inputs still works OK.
This patch adds a third mode for templates: REX inference is requestable
at template instantiation time. This reduces the number of recipes
by removing rex()/nonrex() redundancy for many instructions.
This uses `--locked` on CI to ensure that if the lock file needs
changing it's reflected in the PR instead of letting CI accidentally and
silently update the lock file for us.
* Update the `*.wast` runner to use the `wasmtime` API
This commit migrates the `wasmtime-wast` crate, which executes `*.wast`
test suites, to use the `wasmtime` crate exclusively instead of the raw
support provided by the `wasmtime-*` family of crates.
The primary motivation for this change is to use `*.wast` test to test
the support for interface types, but interface types is only being added
in the `wasmtime` crate for now rather than all throughout the core
crates. This means that without this transition it's much more difficult
to write tests for wasm interface types!
A secondary motivation for this is that it's testing the support we
provide to users through the `wasmtime` crate, since that's the
expectation of what most users would use rather than the raw
`wasmtime-*` crates.
* Run rustfmt
* Fix the multi example
* Handle v128 values in the `wasmtime` crate
Ensure that we allocate 128-bit stack slots instead of 64-bit stack
slots.
* Update to master
* Add comment
error[E0425]: cannot find value `ones` in this scope
--> cranelift-codegen/meta/src/isa/x86/legalize.rs:564:33
|
564 | def!(c = vconst(ones)),
| ^^^^ not found in this scope
This commit uses the `*.witx` files describing the current wasi API to
reduce the boilerplate used to define implementations in the
`wasmtime-wasi` crate. Eventually I'd like to remove lots of boilerplate
in the `wasi-common` crate too, but this should at least be a good start!
The boilerplate removed here is:
* No need to list each function to add it to the
`wasmtime_runtime::Module` being created
* No need to list the signature of the function in a separate
`syscalls.rs` file.
Instead the `*.witx` file is processed in a single-use macro inside the
`wasmtime-wasi` crate. This macro uses the signatures known from
`*.witx` to automatically register with the right type in the wasm
module as well as define a wrapper that the wasm module will call into.
Functionally this is all the same as before, it's just defined in a
different way now!
The shim generated by this macro which wasmtime calls into only uses
`i32`/`i64`/etc wasm types, and it internally uses `as` casts to convert
to the right wasi types when delegating into the `wasi-common` crate.
One change was necessary to get this implemented, however. The functions
in `wasi-common` sometimes took `WasiCtx` and sometimes took a slice of
memory. After this PR they uniformly all require both `WasiCtx` and
memory so the wrappers can be auto-generated. The arguments are ignored
if they weren't previously required.
* Correctly handle possibly misaligned pointers in readdir
This reapplies #615, which was inadvertently reverted.
* Tidy up unneeded `self::` qualifiers.
* Make Dir's contents private.
Also remove the `unsafe` from `impl_iter`. With `Dir`'s field being
private, we can rely on the pointer being only what we've assigned to
it.
* Make `poll`'s timeout argument a `libc::c_int`.
This clarifies that there are no subsequent conversions before calling the
underlying libc API.
* Use clock_gettime instead of clock_getres to get the time.
* Mark FileType::from_raw as safe.
It handles unknown values, so it can be marked safe.
Previously `fsub` was used but this fails when negating -0.0 and +0.0 in the SIMD spec tests; using more instructions, this change uses shifts to create a constant for flipping the most significant bit of each lane with `bxor`.
Previously, the use of `enc_x86_64` emitted two 64-bit mode encodings for `scalar_to_vector.i64`, neither of which contained the REX.W bit telling `MOVD/MOVQ` to move 64 bits of data instead of 32 bits. Now, `scalar_to_vector.i64` will always use a sole 64-bit mode REX.W encoding and `scalar_to_vector` with other widths will have three encodings: a 32-bit mode move, a 64-bit mode move with no REX, and a 64-bit mode move with REX (but not REX.W).
It appears there are two trailing null bytes at the end of the string.
This does not seem right. But it might be a good idea generally to remove
any null bytes that get into error messages.
