Upstream initial test suite from wasm-tools
#85
Conversation
This change starts this proposal's test suite by adding encoding, decoding, and validation tests for the new instructions. This is a wholesale copy of the current state of the `wasm-tools` [tests]. Some things to note: - some of the tests were generated mechanically to attempt to "cover all possibilities;" not all proposals do this, though, so it's unclear whether we want to here (it's verbose, but hopefully more complete) - these tests include the component model built-ins, which are not executable in many engines; it's unclear whether to keep that file (the proposal speaks of this) or not - these tests are a start, not the end state: we're missing anything to do with "thread locals," `pause`, and possibly some other bits. [tests]: https://github.com/bytecodealliance/wasm-tools/tree/5a83828/tests/local/shared-everything-threads
|
@rossberg, looks like we don't even have CI turned on here... |
There was a problem hiding this comment.
I do think it makes sense to inlude mechanically generated tests to improve coverage, and we might as well include the component model tests in case they are useful to more folks. We can always remove things later once we have an interpreter implementation and CI running.
test/core/shared-everything-threads/shared-everything-threads/arrays.wast
Outdated
Show resolved
Hide resolved
test/core/shared-everything-threads/shared-everything-threads/arrays.wast
Outdated
Show resolved
Hide resolved
test/core/shared-everything-threads/shared-everything-threads/globals.wast
Outdated
Show resolved
Hide resolved
test/core/shared-everything-threads/shared-everything-threads/i31.wast
Outdated
Show resolved
Hide resolved
test/core/shared-everything-threads/shared-everything-threads/i31.wast
Outdated
Show resolved
Hide resolved
| (table shared (ref null (shared func)) (elem (ref.null (shared func)))) | ||
| ) | ||
|
|
||
| ;; Note that shared elements can live within an unshared table. |
There was a problem hiding this comment.
It would be good to test that the reverse is invalid. Also, we can test the interaction with shared and unshared globals used in the table initializers.
There was a problem hiding this comment.
It would be good to test that the reverse is invalid.
We do; they're just a few lines down. I'll move those closer to this assertion.
Also, we can test the interaction with shared and unshared globals used in the table initializers.
Were you thinking of variations on this kind of thing?
(module
(type $f (shared (func)))
(global $g (shared (ref null $f)) (ref.null $f))
(table $t shared 1 (ref null $f) (ref.null $f))
(elem (global.get $g))
)There was a problem hiding this comment.
Yes, although that test could be simplified by removing the table entirely and just keeping the element segment.
There was a problem hiding this comment.
Really? Both wasm-tools and wabt seem to expect an elem to have a preceding table definition or we get an error. E.g., (module (elem (i32.const 0) funcref)) is invalid in wat2wasm v1.0.33.
There was a problem hiding this comment.
...but see 740c560 (a lot more notation is necessary than I might have expected to convince wasm-tools it is valid).
There was a problem hiding this comment.
Really? Both
wasm-toolsandwabtseem to expect anelemto have a precedingtabledefinition or we get an error. E.g.,(module (elem (i32.const 0) funcref))is invalid inwat2wasmv1.0.33.
That module is invalid because it defines an active segment, which needs a table to be initialized into. Get rid of the (i32.const 0) offset and it should validate.
Co-authored-by: Thomas Lively <[email protected]>
Co-authored-by: Thomas Lively <[email protected]>
This change starts this proposal's test suite by adding encoding, decoding, and validation tests for the new instructions. This is a wholesale copy of the current state of the
wasm-toolstests. Some things to note:pause, and possibly some other bits.