1249524 - Baldr: add the other (param) declaration form to the text format

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Luke Wagner [:luke]]

Attachment: multi-param

It's in ml-proto and Binaryen uses it.

Comment 1

[Michael Bebenita [:mbx]]

Comment on attachment 8721146 [details] [diff] [review]
multi-param

Review of attachment 8721146 [details] [diff] [review]:
-----------------------------------------------------------------

Small fixes needed, looks good otherwise.

::: js/src/asmjs/WasmText.cpp
@@ +431,4 @@
>        : name_(name),
>          sig_(sig),
>          varTypes_(Move(varTypes)),
> +        localNames_(Move(localNames)),

We should assert that localNames.length() == varTypes.length() + sig.args().length();

@@ +2553,5 @@
> +static bool
> +ParseLocal(WasmParseContext& c, WasmNameVector* localNames, WasmAstValTypeVector* localTypes)
> +{
> +    if (c.ts.peek().kind() == WasmToken::Name) {
> +        if (!localNames->append(c.ts.getIfName()))

We need to append a name even if it doesn't have one, so we can map names to indices correctly.

@@ +2578,5 @@
>  static WasmAstFunc*
>  ParseFunc(WasmParseContext& c, WasmAstModule* module)
>  {
> +    WasmAstValTypeVector varTypes(c.lifo);
> +    WasmAstValTypeVector argTypes(c.lifo);

If we rename this then we should also rename WasmAStSig::args_ and other WasmAstValTypeVector members that store types.

@@ +2604,5 @@
>      while (c.ts.getIf(WasmToken::OpenParen)) {
>          WasmToken token = c.ts.get();
>          switch (token.kind()) {
>            case WasmToken::Local:
> +            if (!ParseLocal(c, &localNames, &varTypes))

(param ..) and (local ..) s-expressions may actually be interleaved. This only works if we assume that they can't.

@@ +3141,2 @@
>      for (size_t i = 0; i < numVars; i++) {
> +        if (!r.registerVarName(func.localNames()[i], i))

I don't think this works. Suppose you have:

(local $a i32) (local i32) (local $b i32) (get_local $a) (get_local 1) (get_local $b)

The code now maps: $a -> 0, $b -> 1, but it should be $a -> 0, $b -> 2, otherwise (get_local 1) (get_local $b) alias the same local.

Comment 2

[Luke Wagner [:luke]]

(In reply to Michael Bebenita [:mbx] from comment #1)
> We should assert that localNames.length() == varTypes.length() +
> sig.args().length();

That's a great idea, but sig is unfortunately a WasmRef, not yet a Sig.

> We need to append a name even if it doesn't have one, so we can map names to
> indices correctly.

Good catch!  Added a test case for this.

> If we rename this then we should also rename WasmAStSig::args_ and other
> WasmAstValTypeVector members that store types.

I'll put 'em back; seems clear enough without the suffix.

> (param ..) and (local ..) s-expressions may actually be interleaved. This
> only works if we assume that they can't.

Yeah, that's a good point and a pre-existing issue.  I'll add "if (!locals.empty()) fail" in the Param case so this doesn't parse since this is grammatically weird anyhow (text order doesn't match index order).

Comment 3

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/78bf144d2e46

Comment 4

[Phil Ringnalda (:philor)]

https://hg.mozilla.org/mozilla-central/rev/78bf144d2e46