Throw a TypeError when calling "next" method of a newly created generator with a value

Throw a TypeError when calling "next" method of a newly created generator with a value

According to the ECMA262 6th rev20 draft 25.3.3.2-7, when calling "next"
method of a newly created generator with a value `generator.next(value)`
TypeError should be thrown. And when calling "throw" method of newly
created generator, generator's status becomes "completed" and throw a
provided value.

BUG=v8:3099

[yusukesuzuki, 2014-01-15 17:53:37]

PTAL

[yusukesuzuki, 2014-01-27 14:47:59]

Hi Michael and Andy, could you take a look at this patch?

[wingo, 2014-01-27 17:48:49]

On 2014/01/27 14:47:59, yusukesuzuki wrote:
> Hi Michael and Andy, could you take a look at this patch?

Sorry for the delay; I'll take a look tomorrow morning.  Thanks for the ping.

[wingo, 2014-01-28 08:56:37]

On 2014/01/27 17:48:49, wingo wrote:
> On 2014/01/27 14:47:59, yusukesuzuki wrote:
> > Hi Michael and Andy, could you take a look at this patch?
> 
> Sorry for the delay; I'll take a look tomorrow morning.  Thanks for the ping.

Probably best to wait until this week's TC39 meeting is over:
https://github.com/tc39/agendas/pull/25

[yusukesuzuki, 2014-01-28 09:02:29]

On 2014/01/28 08:56:37, wingo wrote:
> Probably best to wait until this week's TC39 meeting is over:
> https://github.com/tc39/agendas/pull/25

Nice catch! Agreed.

[wingo, 2014-02-11 08:59:20]

https://codereview.chromium.org/139653003/diff/1/src/ia32/full-codegen-ia32.cc
File src/ia32/full-codegen-ia32.cc (right):

https://codereview.chromium.org/139653003/diff/1/src/ia32/full-codegen-ia32.c...
src/ia32/full-codegen-ia32.cc:1972: __ bind(&resume);
Instead of storing the initial suspend offset in the generator object, could we
not insert code at the resume point of the initial yield that throws unless the
"next" value is undefined?  That would save a word in each generator object.

WDYT?

[yusukesuzuki, 2014-02-11 11:46:59]

https://codereview.chromium.org/139653003/diff/1/src/ia32/full-codegen-ia32.cc
File src/ia32/full-codegen-ia32.cc (right):

https://codereview.chromium.org/139653003/diff/1/src/ia32/full-codegen-ia32.c...
src/ia32/full-codegen-ia32.cc:1972: __ bind(&resume);
On 2014/02/11 08:59:21, wingo wrote:
> Instead of storing the initial suspend offset in the generator object, could
we
> not insert code at the resume point of the initial yield that throws unless
the
> "next" value is undefined?  That would save a word in each generator object.
> 
> WDYT?

Yeah, saving a word in each generator object is important. However, an error is
needed to be raised from Generator.prototype.next(), not from generator's code.
So we need to raise an error in the Generator.prototype.next context.

So I've changed the code. Instead of storing initial suspend offset, when a
value is passed to an unborn generator, yield returns the hole value as a
marker. On the caller (Generator.prototype.next) side, the caller checks the
result and if the result value is the hole value, the caller throws an error.

What do you think about this implementation?

[yusukesuzuki, 2014-02-12 07:03:49]

On 2014/02/11 11:46:59, yusukesuzuki wrote:
> https://codereview.chromium.org/139653003/diff/1/src/ia32/full-codegen-ia32.cc
> File src/ia32/full-codegen-ia32.cc (right):
> 
>
https://codereview.chromium.org/139653003/diff/1/src/ia32/full-codegen-ia32.c...
> src/ia32/full-codegen-ia32.cc:1972: __ bind(&resume);
> On 2014/02/11 08:59:21, wingo wrote:
> > Instead of storing the initial suspend offset in the generator object, could
> we
> > not insert code at the resume point of the initial yield that throws unless
> the
> > "next" value is undefined?  That would save a word in each generator object.
> > 
> > WDYT?
> 
> Yeah, saving a word in each generator object is important. However, an error
is
> needed to be raised from Generator.prototype.next(), not from generator's
code.
> So we need to raise an error in the Generator.prototype.next context.
> 
> So I've changed the code. Instead of storing initial suspend offset, when a
> value is passed to an unborn generator, yield returns the hole value as a
> marker. On the caller (Generator.prototype.next) side, the caller checks the
> result and if the result value is the hole value, the caller throws an error.
> 
> What do you think about this implementation?

Or is it preferable to unwind frame manually and raise error in YieldExpression
INITIAL phase?

[yusukesuzuki, 2014-02-15 04:04:59]

ping

[wingo, 2014-06-10 12:32:18]

On 2014/02/15 04:04:59, yusukesuzuki wrote:
> ping

Thank you very much for bringing up this issue.  In the meantime, and in
response to your patch, the spec has changed to standardize the current
behavior:
http://people.mozilla.org/~jorendorff/es6-draft.html#sec-generatorresume

So thanks again for the patch, although it shouldn't go in now.  Marking not
LGTM to indicate that it is no longer applicable.

[yusukesuzuki, 2014-07-04 02:59:41]

On 2014/06/10 12:32:18, wingo wrote:
> On 2014/02/15 04:04:59, yusukesuzuki wrote:
> > ping
> 
> Thank you very much for bringing up this issue.  In the meantime, and in
> response to your patch, the spec has changed to standardize the current
> behavior:
> http://people.mozilla.org/~jorendorff/es6-draft.html#sec-generatorresume
> 
> So thanks again for the patch, although it shouldn't go in now.  Marking not
> LGTM to indicate that it is no longer applicable.

Ah! Thank you for your clarification!
So I'll close this CL.