IDL: Support deleter operation with any return type

IDL: Support deleter operation with any return type

Web IDL does not require that a deleter operation has boolean as its
return type, it merely defines behavior that is specific to the case that
it has.

Thus, remove our restriction, to avoid rejecting valid IDL declarations
that actual specifications might contain.

BUG=462920
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=191063

[Jens Widell, 2015-03-01 16:01:31]

jl@opera.com changed reviewers:
+ haraken@chromium.org, philipj@opera.com

[Jens Widell, 2015-03-01 16:01:32]

PTAL

Internally, a deleter returns an enum with the values "success", "failure" and
"not handled". Regardless of what the IDL deleter is declared to return, this
enum return value serves the purpose of reporting whether a property delete was
handled or not, back to V8.

We could change the code generation so that we ignore (or assert against) a
non-boolean deleter returning "failure", since that's only possible for deleters
that are defined to return boolean (and also only if they're defined with an
identifier.)

I don't really feel such code generation changes are necessary. WDYT?

[Jens Widell, 2015-03-01 16:33:05]

On 2015/03/01 16:01:32, Jens Widell wrote:
> We could change the code generation so that we ignore (or assert against) a
> non-boolean deleter returning "failure", since that's only possible for
deleters
> that are defined to return boolean (and also only if they're defined with an
> identifier.)
> 
> I don't really feel such code generation changes are necessary. WDYT?

Never mind (this question, not the patch), actually. I slightly misread the
specification.

There is a mismatch between the different C++ layers (in V8, in bindings, and in
the implementation) that we have and the algorithms in Web IDL, so the code
generation changes I talked about are not meaningful.

[haraken, 2015-03-01 16:34:01]

LGTM

> Internally, a deleter returns an enum with the values "success", "failure" and
> "not handled". Regardless of what the IDL deleter is declared to return, this
> enum return value serves the purpose of reporting whether a property delete
was
> handled or not, back to V8.
> 
> We could change the code generation so that we ignore (or assert against) a
> non-boolean deleter returning "failure", since that's only possible for
deleters
> that are defined to return boolean (and also only if they're defined with an
> identifier.)
> 
> I don't really feel such code generation changes are necessary. WDYT?

If the assert is a one-line change, it might be worth doing, but I don't think
the change is worth adding otherwise :)

[Jens Widell, 2015-03-01 16:43:48]

The CQ bit was checked by jl@opera.com

[commit-bot: I haz the power, 2015-03-01 16:45:02]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/968033002/1

[commit-bot: I haz the power, 2015-03-01 17:52:40]

Committed patchset #1 (id:1) as
https://src.chromium.org/viewvc/blink?view=rev&revision=191063

[philipj_slow, 2015-03-02 00:54:50]

Awesome, thanks! I'll update the DOMStringMap interface to make use of this when
I have some spare time.