Proposal: strict built-in functions

# Jack Works (4 months ago)

In the current spec, strictness of the built-in functions are implementation-dependent behaviors. This proposal is going to fix this problem. Jack-Works/proposal

# Isiah Meadows (4 months ago)

You might do better to file a pull request directly against the spec for this: tc39/ecma262. To me, it looks more like an oversight, not something that would likely have to go through all 4 stages. (If it does, at least you already have a repo for it.)

# Jack Works (4 months ago)

It is a runtime level change. I think it might break some code that depends on the strictness of the built-in function like code that using func.caller.

# Mark S. Miller (4 months ago)

Wow, Edge is definitely in violation of the intent of the spec, and the intent of the history of treatments of .caller in previous discussions and specs over the years. The Edge behavior is grossly unsafe.

Hi Jack, thanks for catching this and raising it!! It does need to be fixed. I do think it can be a PR because the historical intent was clear. Since the FF behavior is correct, the correct behavior is web compat, so we should be able to handle this in a PR.

Whether a PR or a full proposal, in either case I'm happy to take this to the committee. I invested years of my life killing non-sloppy function.caller. I can spend a little more ;)

# Claude Pache (4 months ago)

Le 8 déc. 2019 à 14:43, Jack Works <zjwpeter at gmail.com> a écrit :  In the current spec, strictness of the built-in functions are implementation-dependent behaviors. This proposal is going to fix this problem. Jack-Works/proposal-strict-built-in-functions

Hi,

See tc39/ecma262#562 for another testcase.

The real problem is that the semantics of f.caller is left to implementations. There are in fact some restrictions in the spec, see tc39.es/ecma262/#sec-forbidden-extensions, but they are insufficient.

Note that it doesn’t really make sense to mandate that builtin functions be “strict”: The notion of strictness is defined only for so-called ECMAScript functions, which are functions whose implementation is written in ECMAScript code. That excludes builtin functions (unless the implementation choose to implement them in ECMAScript), but also bound functions, proxies, and probably some other cases.

# Jack Works (4 months ago)

I thought the "caller" has been removed from the spec, so there isn't much to do with the "caller" since it is not standard. It's implementation's own extension. But maybe we can also extend The forbidden extensions section.

# Claude Pache (4 months ago)

Le 8 déc. 2019 à 20:49, Jack Works <zjwpeter at gmail.com> a écrit :

I thought the "caller" has been removed from the spec, so there isn't much to do with the "caller" since it is not standard. It's implementation's own extension. But maybe we can also extend The forbidden extensions section.

The caller property of functions (not to be confused with the caller property of the arguments object, which is indeed dead) has never been in the spec, except some restrictions in the [Forbidden Extensions] section introduced at the same time as strict-mode functions. And the problem is, precisely, that it is underspecified.

Now, it is indeed almost trivial to resolve spec-wise the particular issue you noted, by amending the [Forbidden Extensions] section: Just find the bullet that begins with “If an implementation extends any function object with an own property named "caller"...”, and replace therein all occurrences of “strict function” with “strict function or built-in function” (or maybe with “anything but a non-strict function”). That, however, is not the most implementation-friendly way, because that leaves up to them to guess what to do instead (return null, throw an error, etc.), that will not break the web for their users.

Moreover, the caller property has most probably some other gotchas; one of them I recall to have noted some time ago, is that, in at least one implementation, it is publicised as a non-writable, non-configurable own data property, but it may change its value.

I think that it is better, at this point, to specify Function#caller and Function#arguments, as proposed in [gh-issue 562].

[Forbidden Extensions]: tc39.es/ecma262/#sec-forbidden-extensions, tc39.es/ecma262/#sec-forbidden-extensions [gh-issue 562]: tc39/ecma262#562, tc39/ecma262#562

# Claude Pache (4 months ago)

Le 8 déc. 2019 à 15:39, Mark S. Miller <erights at gmail.com> a écrit :

Wow, Edge is definitely in violation of the intent of the spec, and the intent of the history of treatments of .caller in previous discussions and specs over the years. The Edge behavior is grossly unsafe.

Hi Jack, thanks for catching this and raising it!! It does need to be fixed. I do think it can be a PR because the historical intent was clear. Since the FF behavior is correct, the correct behavior is web compat, so we should be able to handle this in a PR.

Whether a PR or a full proposal, in either case I'm happy to take this to the committee. I invested years of my life killing non-sloppy function.caller. I can spend a little more ;)

Hi Mark,

I think it is time to revive tc39/ecma262#562, tc39/ecma262#562 and its associated repo claudepache/es-legacy-function-reflection, claudepache/es-legacy-function-reflection.

# Mark S. Miller (4 months ago)

Thanks Claude, I agree. I have added it to the agenda as well. tc39/agendas/blob/master/2020/02.md