> Looking at the ES6 spec draft, I've drawn the following conclusions
> about Completion Records. Can someone confirm?
>
> If [[type]] is 'normal', 'return', or 'throw',
>       [[target]] must be 'empty' (can't be an identifier).
>
> Moreover, if [[type]] is 'return' or 'throw',
>       [[value]] must be a language value (can't be 'empty').

Looks reasonable but I haven't verified it myself.

>
> If [[type]] is 'continue' or 'break', no restrictions:
>       [[target]] can be 'empty' or an Identifier,
>       [[value]] can be 'empty' or a language value,
>       and all combinations can occur.
>
> ...
>
> The only ones whose existence is less than obvious are those where:
>     [[type]] is 'continue' or 'break', and [[value]] is a language value
> I can see in the pseudo-code the points where they could be created
> (in 12.1.1.2, 12.11.1.2, and 14.1.2), but I haven't tried to reverse-
> engineer a program that would cause it to happen. Could someone explain
> why it's useful to allow such Completion Records, and post a snippet of
> code that would cause one to be created?

12.1.1.2 is the important one here. For example take this test case 
which is expected to return "value":
---
L1:
do
{
   "value";      Completion {[[type]]: normal, [[value]]: "value", 
[[target]]: empty}
   break L1;     Completion {[[type]]: break,  [[value]]: empty, 
[[target]]: "L1"}
}               Completion {[[type]]: break,  [[value]]: "value", 
[[target]]: "L1"}
while(false);   Completion {[[type]]: normal, [[value]]: "value", 
[[target]]: empty}
---

'continue' works the same and just ignore for now that 12.6.1.1 step 2.c 
in the current draft is invalid (it needs to add an additional step to 
convert a break completion value to a normal completion value, cf. 
12.6.1 in the ES5.1 spec).

14.1.2 is only needed to inspect the completion record value:
eval('L1: do { "value"; break L1; } while(false);') === "value"

>
> Thanks,
> -Michael

- André

> Looking at the ES6 spec draft, I've drawn the following conclusions
> about Completion Records. Can someone confirm?
>
> If [[type]] is 'normal', 'return', or 'throw',
>       [[target]] must be 'empty' (can't be an identifier).
>
> Moreover, if [[type]] is 'return' or 'throw',
>       [[value]] must be a language value (can't be 'empty').

Looks reasonable but I haven't verified it myself.

> If [[type]] is 'continue' or 'break', no restrictions:
>       [[target]] can be 'empty' or an Identifier,
>       [[value]] can be 'empty' or a language value,
>       and all combinations can occur.
>
> ...
>
> The only ones whose existence is less than obvious are those where:
>     [[type]] is 'continue' or 'break', and [[value]] is a language value
> I can see in the pseudo-code the points where they could be created
> (in 12.1.1.2, 12.11.1.2, and 14.1.2), but I haven't tried to reverse-
> engineer a program that would cause it to happen. Could someone explain
> why it's useful to allow such Completion Records, and post a snippet of
> code that would cause one to be created?

12.1.1.2 is the important one here. For example take this test case 
which is expected to return "value":

```
L1:
do
{
   "value";      Completion {[[type]]: normal, [[value]]: "value", [[target]]: empty}
   break L1;     Completion {[[type]]: break,  [[value]]: empty, [[target]]: "L1"}
}               Completion {[[type]]: break,  [[value]]: "value", [[target]]: "L1"}
while(false);   Completion {[[type]]: normal, [[value]]: "value",  [[target]]: empty}
```
'continue' works the same and just ignore for now that 12.6.1.1 step 2.c 
in the current draft is invalid (it needs to add an additional step to 
convert a break completion value to a normal completion value, cf. 
12.6.1 in the ES5.1 spec).

14.1.2 is only needed to inspect the completion record value:
`eval('L1: do { "value"; break L1; } while(false);') === "value"`