script/coffeescript tags don't work in IE 8

[liammclennan]

I get an error immediately when I go to http://jashkenas.github.com/coffee-script/ in Internet Explorer 8.

The text of the error is:

User Agent: Mozilla/4.0 (compatible; MSIE 8.0; Windows NT 6.1; WOW64; Trident/4.0; SLCC2; .NET CLR 2.0.50727; .NET CLR 3.5.30729; .NET CLR 3.0.30729; Media Center PC 6.0; InfoPath.2; .NET4.0C; .NET4.0E)
Timestamp: Sun, 9 May 2010 06:29:18 UTC

Message: 'prototype' is null or not an object
Line: 1
Char: 251857
Code: 0
URI: http://jashkenas.github.com/coffee-script/extras/coffee-script.js

[ben-x9]

I just ran into this issue myself running the script locally.

I attempted to rollback to 0.5.6 just in case this bug was something recently introduced, but got a different error:

User Agent: Mozilla/4.0 (compatible; MSIE 8.0; Windows NT 5.1; Trident/4.0; .NET CLR 1.1.4322; .NET CLR 2.0.50727; FDM; .NET CLR 3.0.4506.2152; .NET CLR 3.5.30729)
Timestamp: Mon, 10 May 2010 13:48:20 UTC

Message: 'undefined' is null or not an object
Line: 1
Char: 21693
Code: 0
URI: lib/coffee-script-0.5.6.js

[jashkenas]

Sorry about the IE broken-ness. extras/coffee-script.js is now fixed on master, and should work live on the Internet Explorer of your choosing.

The two issues which had to be solved were these:

• We needed a custom implementation of Array#indexOf, naturally. I added one to Helpers.
• IE is buggy with respect to named function expressions and lexical scope. A named function defined within an inner function will effectively prevent the variable from being passed out to the outer function, where the variable was declared. This is enough of a showstopper to force us to remove named functions from CoffeeScript altogether, unfortunately. It's only a two-line change, but they were great for debugging, so it's sad to see them go.

Closing the ticket.

[d2kagw]

That's a massive shame.
I was using the named functions when looping through classes to detect the type of the class.
For example:
class Person
getName: ->
@name

class Neo extends Person
  name: "Neo

n: new Neo()
console.warn n.constructor.name

Would have output "Neo", but now it's just "" (empty string).
Any ideas of other ways of getting the class name?

[jashkenas]

The function isn't named anymore (because of the IE bug), so no -- there isn't a way to get its name, apart from tagging it manually with a property. A real shame, I completely agree.

[satyr]

Why not keep them named, but with some prefix and/or suffix to avoid name collisions. e.g.:
fn = ->
into
var fn;
fn = function __fn() {}

[devinrhode2]

@jashkenas are there ANY scenarios where we can put in function names? I've been working on TraceKit.js and was oblivious to this IE error, me and @occ will probably re-write TraceKit in coffeescript.

@satyr mentioned giving them a prefix, which is way better for a stack trace than a helpless (anonymous function)

We could even add a few random numbers to the end of the function name, a very simple and easy to implement solution, that would also be very robust. Soon thereafter, we could focus on prefixing by the class name and more deterministic identifiers.

I could totally work on this, very carefully of course.

[toomim]

This feature indeed is useful for debugging. What about enabling it as an option that people can enable in development, when testing on chrome/firefox/safari? They could then disable it in production.

I'd also be able to use this option for most of my coffeescript projects because they compile in the browser. So I can detect at run-time if the browser is not IE, and if so, enable the anonymous named functions.