repl breaks something about constructor/type info

[eventualbuddha]

As illustrated by this gist, Node.js v0.4.8 (and apparently v0.5.8, tested independently in #node.js) exhibits a strange behavior in the repl where loading a module function that passes a constructor (e.g. String or Number) in to be checked against an object's constructor doesn't behave as expected. Here is the function to be run:

// 00-type.js
exports.is = function(object, constructor) {
  return object instanceof constructor || (object != null ? object.constructor : void 0) === constructor;
};

$ node
> type = require('./00-type')
{ is: [Function] }
> type.is('foo', String)
false

However, if you then re-create this simple is function and use the new version, it works fine:

> eval('global.test = '+type.is.toString())
[Function]
> test('foo', String)
true

This is different from how this works when running node with a file argument and not in a repl:

// 01-test.js
type = require('./00-type')

console.log('type.is =>', type.is('foo', String))
eval('global.test = '+type.is.toString())
console.log('test    =>', test('foo', String))

$ node 01-test.js
type.is => true
test    => true

In that case, both return the expected true. Something is clearly wrong here, but I'm not sure what.

[bnoordhuis]

Not a bug. Modules and the REPL run in separate VM contexts. Each context has its own set of "root" objects (Object, String, Date, etc.). Look at e.g. isRegExp() in lib/util.js to see how we handle that:

function isRegExp(re) {
  return re instanceof RegExp ||
      (typeof re === 'object' && objectToString(re) === '[object RegExp]');
}

[eventualbuddha]

I had a feeling it might be something like that. Nevertheless, it's pretty strange behavior that can make testing code in the REPL somewhat less useful. Yes, real tests should be written in Jasmine or something but it's nice to be able to duck into the REPL sometimes to "see how it works", and sharp corners like this make that kinda dangerous. You don't see this kind of behavior, afaik, in Ruby, Python, or Scala.

[eventualbuddha]

Anyway, thanks for checking it out. Maybe this implies that there should be bigger warnings about this kind of behavior.

[bnoordhuis]

It's a common source of confusion. #1484 intends to address it but it's one of those TODOs no one ever seems to get around to.