IDL: Fix exception handling when converting V8 value to union/dictionary

IDL: Fix exception handling when converting V8 value to union/dictionary

The toImpl() functions generated for union and dictionary types should
report exceptions via their ExceptionState argument. But internally, they
were doing value conversions using the TONATIVE_VOID_EXCEPTIONSTATE(),
TONATIVE_VOID_EXCEPTIONSTATE_ARGINTERNAL() and TOSTRING_VOID() macros,
that always throw exceptions all the way to V8 (in the TONATIVE_*() cases
by calling exceptionState.throwIfNeeded().)

To fix this, introduce _NOTHROW variants of the TONATIVE_*() macros, that
end with

  if (exceptionState.hadException())
    return;

instead of

  if (exceptionState.throwIfNeeded())
    return;

and use those instead, and use TOSTRING_VOID_EXCEPTIONSTATE() instead of
TOSTRING_VOID(). TOSTRING_VOID_EXCEPTIONSTATE(), unlike the
TONATIVE_*_EXCEPTIONSTATE() macros, does not throw exceptions all the way
to V8, but rather leaves them in the ExceptionState object.

BUG=240176, 321462

[Jens Widell, 2015-02-23 13:17:06]

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

[Jens Widell, 2015-02-23 13:17:07]

PTAL

Needless to say, these macros are getting out of hand...

This problem is contributing to the unit test failures in PS#1 of
https://codereview.chromium.org/947923002/. The unit tests are calling toImpl()
with a TrackExceptionState, on which calling throwException()/throwIfNeeded() is
effectively illegal, because none of the throw*Error() functions actually throw
an exception (which throwExecption() asserts must have been done.)

[haraken, 2015-02-23 13:48:42]

On 2015/02/23 13:17:07, Jens Widell wrote:
> PTAL
> 
> Needless to say, these macros are getting out of hand...
> 
> This problem is contributing to the unit test failures in PS#1 of
> https://codereview.chromium.org/947923002/. The unit tests are calling
toImpl()
> with a TrackExceptionState, on which calling throwException()/throwIfNeeded()
is
> effectively illegal, because none of the throw*Error() functions actually
throw
> an exception (which throwExecption() asserts must have been done.)

The complexity of these macros is getting unacceptable... I'm sorry to say this
without any alternative ideas, but we might want to spend time in fixing the
issue without messing up the macros :-/