Always lazy compile arrow functions

Always lazy compile arrow functions

This causes the scope of default parameters to be correct, where otherwise
we'd get it wrong, as expressions in default parameters would be evaluated
in the scope containing the arrow function.

This is something of a hack. The right way to fix it would likely be to
do something like pattern-rewriter, but for scopes.

BUG=v8:4395
LOG=n

[adamk, 2015-09-19 00:06:04]

adamk@chromium.org changed reviewers:
+ caitpotter88@gmail.com, rossberg@chromium.org, wingo@igalia.com

[adamk, 2015-09-19 00:06:05]

I'm not at all sure we want to land this, but I wanted to bump its visibility up
a bit.

[adamk, 2015-09-19 00:06:15]

Patchset #1 (id:1) has been deleted

[caitp (gmail), 2015-09-19 21:34:20]

On 2015/09/19 00:06:05, adamk wrote:
> I'm not at all sure we want to land this, but I wanted to bump its visibility
up
> a bit.

Given that this bug is observable with --harmony-destructuring alone, and that I
think we're wanting to ship destructuring soon (if I am remembering my email
correctly), it might be worth landing this as an interim fix and improving on it
later.

[caitp (gmail), 2015-09-19 21:34:36]

https://codereview.chromium.org/1317033005/diff/20001/src/preparser.h
File src/preparser.h (right):

https://codereview.chromium.org/1317033005/diff/20001/src/preparser.h#newcode...
src/preparser.h:3288: !IsArrowFunction(result->AsFunctionLiteral()->kind())) {
The only time you _need_ to force re-parsing of arrows is when the parameter
list contains expressions (eg an initializer), so we could probably add an extra
condition to escape lazy parsing here. But, it might not matter that much
anyway, so.

[rossberg, 2015-09-22 15:33:05]

https://codereview.chromium.org/1317033005/diff/20001/src/preparser.h
File src/preparser.h (right):

https://codereview.chromium.org/1317033005/diff/20001/src/preparser.h#newcode...
src/preparser.h:3288: !IsArrowFunction(result->AsFunctionLiteral()->kind())) {
On 2015/09/19 21:34:36, caitp wrote:
> The only time you _need_ to force re-parsing of arrows is when the parameter
> list contains expressions (eg an initializer), so we could probably add an
extra
> condition to escape lazy parsing here. But, it might not matter that much
> anyway, so.

That's a good point. It might make sense to avoid enforcing lazy parsing for
arrow functions with simple parameters. I expect arrow IIFE to become a not
too-uncommon use case.

[adamk, 2015-09-25 21:17:39]

On 2015/09/22 15:33:05, rossberg wrote:
> https://codereview.chromium.org/1317033005/diff/20001/src/preparser.h
> File src/preparser.h (right):
> 
>
https://codereview.chromium.org/1317033005/diff/20001/src/preparser.h#newcode...
> src/preparser.h:3288: !IsArrowFunction(result->AsFunctionLiteral()->kind())) {
> On 2015/09/19 21:34:36, caitp wrote:
> > The only time you _need_ to force re-parsing of arrows is when the parameter
> > list contains expressions (eg an initializer), so we could probably add an
> extra
> > condition to escape lazy parsing here. But, it might not matter that much
> > anyway, so.
> 
> That's a good point. It might make sense to avoid enforcing lazy parsing for
> arrow functions with simple parameters. I expect arrow IIFE to become a not
> too-uncommon use case.

What's the right way to get ahold of this information in the PreParser? I don't
have a real FunctionLiteral, so there's no Scope to check
has_simple_parameters() on.

[adamk, 2015-09-26 00:32:34]

This is not a sufficient fix, see attached bug for details.

[caitp (gmail), 2015-09-26 12:20:45]

On 2015/09/26 00:32:34, adamk wrote:
> This is not a sufficient fix, see attached bug for details.

So, a rewriter is probably all we can do