1926357 - Update wasm spec tests to target wasm-3.0 branch

Status: RESOLVED FIXED

(Core :: JavaScript: WebAssembly, task, P1)

Description

[Ben Visness [:bvisness]]

The core WebAssembly spec repo is actually lagging very far behind the state of the art, because all the new stuff (and I mean all of it) is on a branch called wasm-3.0. The result is that we have lots of core-wasm spec tests scattered across our various proposals, e.g. memory64 and gc. Many of these are duplicated. We should just update our core spec tests to target wasm-3.0 to reduce this duplication and track with the actual current state of wasm spec tests.

Comment 1

[Ben Visness [:bvisness]]

Attachment: Bug 1926357: Update wasm spec tests to wasm-3.0. r=rhunt

Many proposals have been merged into the wasm-3.0 branch of the
WebAssembly spec repo. It currently represents the state of the art of
wasm, which we keep up with - proposals that are officially standardized
but not yet super widely implemented.

Basing our spec tests on this reduces duplication across proposal repos
and should help us catch spec problems earlier.

This patch reworks the spec test generator slightly to handle some new
module definition syntax, and removes proposals which have already been
merged into wasm-3.0.

Comment 2

[Ben Visness [:bvisness]]

Attachment: Bug 1926357: Rework wasm tag types and result types. r=rhunt

The intersection of tag types and rec groups causes a subtle situation
where two structurally-identical tag types should not match because
the tag types are not the same. This patch changes tag types to store
TypeDef* instead of a vector of val types. This disallows some module
linking that was allowed before, but only in cases where the types were
equal anyway and therefore no incorrect behavior would occur.

This patch also fixes one other situation where ResultTypes were being
compared for equality, instead of subtyping being used. This fixes a
subtle validation issue with if statements lacking an else.

Depends on D227573

Comment 3

[Ben Visness [:bvisness]]

Attachment: Bug 1926357: Tweak message about globals in constant exprs. r=rhunt

This was intended to ship in a prior patch but I missed it.

Depends on D227574

Comment 4

[Ben Visness [:bvisness]]

Attachment: Bug 1926357: Fix up types around ref.func and elem segments. r=rhunt

Now that function references are standard, our handling of ref.func
should produce the most specific possible type instead of the broad
"funcref" type. This also applies to globals and elem segments.

Depends on D227575

Comment 5

[Ben Visness [:bvisness]]

Attachment: Bug 1926357: Don't ignore destination index on array.init_data. r=rhunt

The value is tired of being ignored. It wants to be known. It wants to
be seen and acknowledged and loved.

Depends on D227576

Comment 6

[Ben Visness [:bvisness]]

Attachment: Bug 1926357: Rework wasmFailValidateText and assertErrorMessage for clarity. r=rhunt

The message for wasmFailValidateText made it sound like the compile had
not failed, when in fact it just did not fail "as expected", i.e. in the
expected way. That message is now clearer.

The message printing for assertErrorMessage mixed up the order in a
confusing way that made it seem like the top-level user message was
actually the most proximal cause of the exception. This was weird and
has been reordered. The code has also been reworked for clarity, making
it obvious that no exception can slip through, and handling the common
mistake of forgetting to provide an error message. (It is at least a
common mistake for me.)

Depends on D227573

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/35ff4665b69f
Update wasm spec tests to wasm-3.0. r=rhunt,supply-chain-reviewers
https://hg.mozilla.org/integration/autoland/rev/7790b0f22ca3
Rework wasmFailValidateText and assertErrorMessage for clarity. r=rhunt
https://hg.mozilla.org/integration/autoland/rev/3fa6248e7d59
Rework wasm tag types and result types. r=rhunt
https://hg.mozilla.org/integration/autoland/rev/b26eb8c10962
Tweak message about globals in constant exprs. r=rhunt
https://hg.mozilla.org/integration/autoland/rev/cf339037b119
Fix up types around ref.func and elem segments. r=rhunt
https://hg.mozilla.org/integration/autoland/rev/0e0a725c49fd
Don't ignore destination index on array.init_data. r=rhunt

Comment 8

[agoloman]

https://hg.mozilla.org/mozilla-central/rev/35ff4665b69f
https://hg.mozilla.org/mozilla-central/rev/7790b0f22ca3
https://hg.mozilla.org/mozilla-central/rev/3fa6248e7d59
https://hg.mozilla.org/mozilla-central/rev/b26eb8c10962
https://hg.mozilla.org/mozilla-central/rev/cf339037b119
https://hg.mozilla.org/mozilla-central/rev/0e0a725c49fd