Normative: CreateDynamicFunction early concatenates bodyText

[leobalter]

Ref tc39/test262#2102
Ref tc39/test262#2109

This normative change matches the current behavior of V8, SpiderMonkey, and JSC (e.g. Function("-->"). This is not only for the convenience of reflecting the observed behaviors, but this will allow some guard for potential false positives like Function("#!hashbang").

[anba]

but this will allow some guard for potential false positives like Function("#!hashbang").

I don't understand this part, hashbang is only allowed when parsing with the goal symbol Script or Module, so it never applies in CreateDynamicFunction.

[leobalter]

@anba right you are. Maybe the only advantage is the convenience for reflecting engines behavior in a very specific usage.

[leobalter]

@ljharb the current tests we have on Test262 matches this proposed behavior. If tc39/test262#2101 tc39/test262#2102 get merged those will invert.

If we have those PRs on hold until the TC39 meeting the tag for need tests could be discarded.

---

Just to inform, I just want a closing answer for this problem. One way or another, I'm fine with any decision TC39 chooses for this PR.

[leobalter]

tc39/test262#2101 is now merged, that means I need a new PR to now reflect this change.

[leobalter]

Doing more investigation over this issue I found some similar cases:

<script>-->
console.log('foo');</script>

Chrome, Firefox, Safari: will log `'foo'`.
Edge: SyntaxError

Anyway a file starting with --> will have the observed SyntaxError properly on v8, ChakraCore, SpiderMonkey, JSC, and Node.

The same issue is also observed with eval:

$ eshost -se 'eval("-->")'
#### d8, jsc, jsshell, node
undefined

#### ch
SyntaxError: Syntax error

---

The error is similar to what happens to Function("-->"), but I'm not convinced the fix should happen at the same place for whatever happens to Function (CreateDynamicFunction), eval (PerformEval), or the script tag (not ecma262).

[leobalter]

There is an interesting observation for the actual output of the Function ctor:

eshost -se 'Function(42).toString()'
#### ch
function anonymous(
) {42
}

#### d8, jsshell, node
function anonymous(
) {
42
}

#### jsc
function anonymous() {
42
}

That means that ChakraCore is also not compliant to CreateDynamicFunction when the SourceText is not correctly set, it misses the preceding line feed. I wonder if fixing this part would have the same issue observed for HTMLCloseComments.

A function like this will be a SyntaxError anywhere:

function fn() {-->
}

but this function formatting is not:

function fn() {
--> valid code
}

And this matches the revealed code from Function("-->"):

$ eshost -se 'Function("-->").toString()'
#### d8, jsshell, node
function anonymous(
) {
-->
}

#### ch
SyntaxError: Syntax error

#### jsc
function anonymous() {
-->
}

---

As seen in the examples, it makes sense to observe the function body in conjunction with the padding of line feeds added at CreateDynamicFunction.

I'm ok to observe the string: \n-->\n can be correctly parsed.

[leobalter]

My conclusion is to stick with the proposed change and discuss the problems with eval elsewhere. These are another problem I don't have time and availability to have immediately fixed.

Fixing the Function ctor in the specs will not preclude anything else to be fixed.

[ljharb]

Discussed on the editor call; this is ready to go.