Wrong test expectation in const_conditional_test/08

Wrong test expectation in const_conditional_test/08

Committed: https://code.google.com/p/dart/source/detail?r=28169

[hausner, 2013-10-02 16:30:13]

Committed patchset #1 manually as r28169 (presubmit successful).

[hausner, 2013-10-02 16:30:45]

TBR=johnniwinther

[kasperl, 2013-10-02 16:58:23]

I think this CL is wrong. The test looks fine to me. It looks like the VM allows


   0 ? 1 : 2

as a compile-time constant, but that is not correct according to the
specification (if I remember correctly) because 0 isn't a bool.

[hausner, 2013-10-02 17:04:59]

Where does the spec say that in *production* mode an error has to be reported
for expressions like 0 ? 1 : 2 or statements like if (0) {...}

I will check with Gilad but we have many bugs if boolean conversion reports type
errors in production mode.

[kasperl, 2013-10-02 17:07:50]

On 2013/10/02 17:04:59, hausner wrote:
> Where does the spec say that in *production* mode an error has to be reported
> for expressions like 0 ? 1 : 2 or statements like if (0) {...}
> 
> I will check with Gilad but we have many bugs if boolean conversion reports
type
> errors in production mode.

This isn't about boolean conversion. It's about what constitutes a compile-time
constant.

[hausner, 2013-10-02 17:17:20]

Ok, you mean this part of the spec, discussing CT constants:

An expression of the form e1?e2:e3 where e1, e2 and e3 are constant expressions,
and e1 evaluates to a boolean value.

So, after the evaluation of e1 a type check is done even in production mode to
check whether e1 is bool?

[kasperl, 2013-10-02 17:21:29]

On 2013/10/02 17:17:20, hausner wrote:
> Ok, you mean this part of the spec, discussing CT constants:
> 
> An expression of the form e1?e2:e3 where e1, e2 and e3 are constant
expressions,
> and e1 evaluates to a boolean value.
> 
> So, after the evaluation of e1 a type check is done even in production mode to
> check whether e1 is bool?

That's how I read it. You need to check if e1 is true or false -- if not the
expression is not a compile-time constant regardless of what e2 and e3 evaluate
to.

[hausner, 2013-10-02 17:28:41]

Ok, get it. I have a change that reverts this and fixes the VM.

Stay tuned.