fake-fs should probably use 'bind' to fix the 'this' of its 'node-like' methods.

[kdvolder]

The node-like filesystem api produced by

var fs = new Fs()

Is not quite equivalent to node fs in the sense that code like the following, which works with node fs doesn't work with a fake fs.

var fs = new Fs();
var readFile = fs.readFile;
readFile('/foo/bar.txt', ...);

This code breaks because the 'this' of readFile call is incorrectly set. It only works correctly if I call it like so: fs.readFile(...).

With 'raw' nodefs I can either call fs.readFile(...) or just readFile(...). Both are equivalent.

That means I have to be a tad more careful how I use/call the function from a fake fs instance. Ideally a fake fs instance should as much as possible try to emulate the behavior of raw node fs.

So, I wonder if it would be possible for the fake constructor to automatically 'seal' the returned api and bind the this object automatically for the node-like methods in the object returned by new Fs().

[eldargab]

Makes sense. I agree in general because I can't give any argument against, except, may be, performance. Haha, sounds crazy, but such binding things are extremely slow, so, who knows? :) Also once you know that fake fs is a classic object (which is not so unexpected BTW) it's not hard to avoid var readFile = fs.readFile. Also, personally, I only used fake-fs for patching global fs module (that's because of third party dependencies). That's why I'd like to leave things as is for now.

[kdvolder]

It is entire your call :-)

I'm using fake-fs to test a 'composable' filesystem I'm developing. The idea is you can create something that looks and behaves like nodefs by plugging together different things that provide a way to read/write files and directories.

So you can do something like:

  var nodefs = require('fs');
  var myFs = compose(
       withPrefix('/foo', subDir('/home/kdvolder/foo', nodefs)),
       withPrefix('/tmp', subDir('/tmp', nodefs))
  );

To create regression tests for this I'm using 'fake-fs' instances instead of nodefs.

In a way, actually the fact that 'fake-fs' doesn't 'auto-bind' the functions may be good for my tests. If my composite-fs implementation doesn't take care of proper binding or this arguments, then it will still work for nodefs but it may not work for other potential fs implementations that don't 'auto-bind'.

Now that my code is tested to behave with fakefs it will also work with nodefs, but the other way around may not be true. So in a way having fake-fs not do the 'auto-bind' while it deviates from standard nodefs behavior may actually make it a better test :-)

Anyway, I guess I just made a pretty good argument not to do auto-binding :-)

I won't be offended at all if you close this issue.

[eldargab]

Yeah, I'll remember the argument :)