A small amount of cleanup for the top-level of this repository, where
the `installer` directory just had one misc file for the Windows
installer which we can put elsewhere inside of `ci`, another misc folder
but one that already has a few files in it.
* General Cargo.toml cleanup.
- Remove travis-ci attributes.
- Remove "experimental" badges from actively-developed crates.
- Reflow some long lines.
- Use dependency features consistently.
- Add readme attributes
* Update WASI to the latest trunk.
This notably adds a .gitignore file for the WASI directory.
* Refactor Lightbeam's tests.
This refactors Lightbeam's tests.rs file into several pieces, separating
quickcheck tests into their own file, and moving tests which can be run as
wast tests into `tests/misc_testsuite`, and creating a tests directory
for the rest.
* Remove the old filetests tests.
These are all covered by misc_testsuite and spec_testsuite tests.
* rustfmt
* Remove the "bench" feature.
* Try and assign directly to return registers; backtrack to use struct-return param
Rather than trying to count number of return registers that would be used by a
given set of return values, optimistically assign the return values to
registers. If we later find that we can't fit them all in registers, then
backtrack and introduce the use of a struct-return pointer parameter.
* Rename `rets2` and wrap it in an option so we avoid the clone for non-multi-value
In some cases, compute_size() is used to choose between various different Encodings
before one has been assigned to an instruction. For x86, the REX.W bit is stored
in the Encoding. To share recipes between REX/non-REX, that bit must be inspected
by compute_size().
This commit fixes a subtle bug in `fd_readdir` implementation on
BSD-style nixes. In particular, the bug only resurfaced when testing in
release mode, and I can only assume it was due to a unlucky combination
of compiler optimizations and at first sight correct casting of `d_name`
pointer from `const* i8` to `const* u8`. This is now fixed by first
converting `d_name` to `std::str` by using `std::ffi::CStr::to_str`, and
then using the resultant `std::str` to copy the properly validated `u8`
contents into the Wasm buffer.
Furthermore, this commit fixes incorrect handling of the `readdir` loop
break condition. Although undocumented in BSD man pages, the signalling
is the same as on Linux where a null ptr returned by `readdir` signals
the end of the dir stream when the errno code has not changed since
before calling `readdir` inside a loop. Upon a fault such as an invalid
file descriptor, the errno *will* change after executing `readdir`.
* Update the top-level README.md and embedding documentation.
wasmtime-api is now the primary external API crate, so recommend that
instead of wasmtime-jit.
Also, enable wasmtime-api's C API by default, so that it shows up on
docs.rs, and to make it easier to use.
And, add basic embedding documentation and link to it from the
README.md. Credit to @yurydelendik for the content.
* Use the new wasm-c-api URL.
* Don't pass --features wasm-c-api, as it is now on by default.
This commit fixes the build errors in the unwind info implementation for
the x86 ABI by changing `byteorder` to build `no_std`.
This copies two simple functions from the `WriteBytesExt` trait so that
we can easily write to a `Vec<u8>` with a particular endianness.
Fixes#1203.
This commit simplifies the build script slightly for generating tests by
doing a few dull refactorings:
* Leaves formatting to `rustfmt`
* Extract bulk of code execution into a top-level shared `run_wast`
function so each test is a one-liner
* Use `anyhow` for errors both in the script and in tests
Some wasi-common tests assume that stdin is never ready to be read, but
on CI stdin is closed so it's always ready to be read. Work around this
by guaranteeing that wasi-common tests always have an unreadable stdin
pipe by creating our own pipe.
This commit switches the beta/nightly tests to happen on Ubuntu instead
of macOS. Turns out GitHub Actions has scheduling limitations on macOS
that limit repositories to 5 concurrent jobs per repository, so let's
reduce the load a bit by running more builds on Linux than mac.
