This results in a "ReferenceError: FooClass is not defined" in the 2016-05-28 Nightly and 46.0.1, but not in Chrome: var g = (function () { function f() { new FooClass(); } class FooClass { } return f; })(); g(); I expect it to succeed based on this explanation: http://stackoverflow.com/a/35537963/1026 As far as I understand, the "FooClass" binding should be in place (though not having a value yet) when the closure is initialized. After the outer function returns, the FooClass binding will point to the class. Everything works fine - if I move the `function f()` below the class declaration. - or if I use `let` with class expression (`let FooClass = class {}`) This reproduces in a web page (simply wrapped in <script>) and in the JS shell. Curiously, running the code from the Web Console works

Seems correct to me. Needinfo efaust for confirmation.

Yeah, this is a bug. Even if it weren't, the bytecode used for the name access is clearly wrong and would represent a bug: js> dis(g) flags: CONSTRUCTOR loc op ----- -- main: 00000: getgname "FooClass" 00005: undefined 00006: dupat 1 00010: new 0 00013: pop 00014: retrval Source notes: ofs line pc delta desc args ---- ---- ----- ------ -------- ------ 0: 2 0 [ 0] newline 1: 3 0 [ 0] colspan 8 3: 3 14 [ 14] xdelta 4: 3 14 [ 0] newline 5: 4 14 [ 0] colspan 4 Truly shocking that we'd get def-use chain information wrong anywhere, y'know?

I put this on the bakc burner, because that hoisting will be much easier to ensure is right with the new bindings story that shu is working on. It's clearly a bug, due to the way the Parser::classDefinition is written.

Fixed by bug 1263355.

Thanks! Indeed this works as intended in the 2016-08-28 nightly.