Issue 1368 in google-caja: Support Function() constructor with string literal

[googl...@googlecode.com]

Status: New
Owner: metaw...@gmail.com
Labels: Type-Defect Priority-Medium

New issue 1368 by metaw...@gmail.com: Support Function() constructor with
string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

Shouldn't be any harder than event handlers. Unlike eval(), Function()
doesn't close over the surrounding scope, so it should be hoisted to the
top and given some triple-underbar name.

[googl...@googlecode.com]

Updates:
Labels: -Type-Defect Type-Enhancement

Comment #1 on issue 1368 by eri...@google.com: Support Function()

constructor with string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

(No comment was entered for this change.)

[googl...@googlecode.com]

Comment #2 on issue 1368 by metaw...@gmail.com: Support Function()

constructor with string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

MathJax uses this technique to avoid creating unnecessary closures.

[googl...@googlecode.com]

Comment #3 on issue 1368 by metaw...@gmail.com: Support Function()

constructor with string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

If we have a string literal as the argument to Function, we can replace the
whole thing with a reference to a global function that returns a closure:

var x = 1;
function f() {
var x = 2;
return new Function("return x;");
}

becomes

function makeClosure1() {
return function () {
return x;
}
}
var x = 1;
function f() {
var x = 2;
return makeClosure1();
}

Note that f()() === 1, not 2.

[googl...@googlecode.com]

Comment #4 on issue 1368 by eri...@google.com: Support Function()

constructor with string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

These are not equivalent in SES for many of the same reasons as raised in
our recent conversation about eval vs cajaVM.eval. Unlike the global "eval"
function, which can be shadowed per SES evaluation, the Function
constructor is necessarily global to a SES frame, since it can also be
reached by (function(){}).constructor. Therefore, it is analogous to
cajaVM.eval in that it cannot provide access to mutable state or authority.
In SES

var x = 1;
function f() {
var x = 2;
return new Function("return x;");
}

f();

Must fail, since the newly made function may not mention "x" and does not
have this x in its scope regardless.

[googl...@googlecode.com]

Comment #5 on issue 1368 by metaw...@gmail.com: Support Function()

constructor with string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

Do you see a way to support the Function constructor in SES, even with a
statically known argument? If not, should we even bother to try supporting
it in ES5/3?

[googl...@googlecode.com]

Comment #6 on issue 1368 by eri...@google.com: Support Function()

constructor with string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

We support the Function constructor, with the only caveat being that the
only free variables that may occur in the argument strings are the
whitelisted globals. So long as this restriction is enforced, your current
technique for supporting it in ES5/3 should work fine.

[googl...@googlecode.com]

Comment #7 on issue 1368 by metaw...@gmail.com: Support Function()

constructor with string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

In ES53, Function("...") would have to be a special form.

var F = Function;
new F("...");

would be different from

new Function("...");

We don't have any other special forms at the moment that I can find. (I
recall that "load" used to be one, but I think that was removed at the same
time as valija.) Does our aversion to special forms outweigh our desire to
support MathJax and other such libraries?

[googl...@googlecode.com]

Comment #8 on issue 1368 by eri...@google.com: Support Function()

constructor with string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

> Does our aversion to special forms outweigh our desire to support MathJax
> and other such libraries?

Which of these would be satisfied with a confining Function special form?
Of those, how much do we care?

By "confining" here I mean the restriction mentioned in comment 6: "the

[googl...@googlecode.com]

Updates:
Status: Fixed

Comment #9 on issue 1368 by ihab.a...@gmail.com: Support Function()

constructor with string literal
http://code.google.com/p/google-caja/issues/detail?id=1368

Supported in ES5. No longer a factor.

--
You received this message because this project is configured to send all
issue notifications to this address.
You may adjust your notification preferences at:
https://code.google.com/hosting/settings