915419 - Support "object" in union return values and unions inside dictionaries/sequences

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect)

Description

[Boris Zbarsky [:bzbarsky]]

Patch coming up.

Comment 1

[Boris Zbarsky [:bzbarsky]]

Attachment: Add support for "object" types in owning unions (so union return values and unions in dictionaries and sequences. smaug

Adds RootedUnion and NullableRootedUnion structs that are used on the
stack for union return values when the union needs rooting.  It also
adds a TraceUnion method on union return values, which is called by
(Nullable)RootedUnion or by dictionary tracing.  TraceUnion traces
typed array and object members of unions (the only things unions can
contain so far that might need tracing).  Union return values are
changed to store raw JSObject* or typed array structs instead of
Rooted versions of both; the tracing is now handled via TraceUnion.

The wrapping code for dictionary arguments to constructors is fixed to
actually work.  This required adding GetAs* methods to union return
values that return references to the internal data.

The SetToObject method is adjusted to actually work for union return
value structs and not assume it's being generated for a
union-conversion struct, and we now generate SetToObject methods as
needed for union return values.

Comment 2

[David Zbarsky (:dzbarsky)]

Comment on attachment 803331 [details] [diff] [review]
Add support for "object" types in owning unions (so union return values and unions in dictionaries and sequences.   smaug

Review of attachment 803331 [details] [diff] [review]:
-----------------------------------------------------------------

::: dom/bindings/BindingUtils.h
@@ +1816,5 @@
> +  {
> +    this->TraceUnion(trc);
> +  }
> +};
> +

Shouldn't this disallow copy constructor/operator=?

::: dom/bindings/test/TestCodeGen.webidl
@@ +498,1 @@
>    (CanvasPattern? or CanvasGradient) receiveUnionContainingNull();

Do you have tests for nullable unions with objects?

Comment 3

[Boris Zbarsky [:bzbarsky]]

> Shouldn't this disallow copy constructor/operator=?

It gets that automatically from AutoGCRooter disallowing it, right?

> Do you have tests for nullable unions with objects?

Good catch.  Added some.

Also, note that for now we don't support unions containing gcthings inside a sequence because I didn't add a SequenceTracer specialization for them.  We can worry about it if we ever need them.

Comment 4

[Boris Zbarsky [:bzbarsky]]

Attachment: With more tests

Comment 5

[David Zbarsky (:dzbarsky)]

(In reply to Boris Zbarsky [:bz] from comment #3)
> > Shouldn't this disallow copy constructor/operator=?
> 
> It gets that automatically from AutoGCRooter disallowing it, right?
> 

Oh right, I guess the default copy ctor calls the superclass so its ok.

Comment 6

[David Zbarsky (:dzbarsky)]

Comment on attachment 803419 [details] [diff] [review]
With more tests

Review of attachment 803419 [details] [diff] [review]:
-----------------------------------------------------------------

::: dom/bindings/Codegen.py
@@ +4729,1 @@
>                  memberWraps.append(memberWrap)

Yeah, this makes more sense than what I had before ;)

@@ +6314,1 @@
>              unionValues.append(string.Template("UnionMember<${structType} > "

It would probably be clearer if you collapsed these and made the return type depend on isReturnValue

Comment 7

[Boris Zbarsky [:bzbarsky]]

Attachment: With that changed

Good idea!

Comment 8

[David Zbarsky (:dzbarsky)]

Comment on attachment 803462 [details] [diff] [review]
With that changed

Review of attachment 803462 [details] [diff] [review]:
-----------------------------------------------------------------

I don't know how JS_CallObjectTracer works but I'm assuming you got that part right.

::: dom/bindings/Codegen.py
@@ +3304,1 @@
>              # This is a bit annoying.  In a union we don't want to have a

Want to add a TypedArray in union test?

@@ +4492,1 @@
>              resultArgs = None

Why do you need this?

Comment 9

[Boris Zbarsky [:bzbarsky]]

> Want to add a TypedArray in union test?

Not convinced about building and linking more test-only union code...

> Why do you need this?

So I don't get "referenced before assignment" exceptions when I return resultArgs?

Comment 10

[David Zbarsky (:dzbarsky)]

(In reply to Boris Zbarsky [:bz] from comment #9)
> > Want to add a TypedArray in union test?
> 
> Not convinced about building and linking more test-only union code...
> 

Sure, I guess this won't be used in real idl.  I wonder if we can write a fuzzer that generates valid WebIDL to make sure we at least codegen things that compile.

> > Why do you need this?
> 
> So I don't get "referenced before assignment" exceptions when I return
> resultArgs?

Ah, true.

Comment 11

[Olli Pettay [:smaug][[e-mail]]]

Any chance to see some generated code. This is dealing with tracing in a bit complicated setup and
it just need to be right.

Comment 12

[Boris Zbarsky [:bzbarsky]]

Attachment: The relevant generated code for ObjectOrLongReturnValue

Olli, does this help?  Or should I attach the entirety of TestCodeGen and UnionTypes?

Comment 13

[Boris Zbarsky [:bzbarsky]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/54247b7b87e6

Comment 14

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/54247b7b87e6