Create an optimistic mkdirp

[jclem]

This implements a simpler, optimistic form of mkdirp. The code is significantly easier to read and will be faster in cases where most of the given path already exists.

[damccorm]

There's 2 orthogonal issues here:

1. Changing from iterative to recursive implementation. That seems fine to me, I don't see much advantage though.

2. Changing to the optimistic approach. I don't actually think its going to make things faster though. Looking only at more time consuming calls:

In the case the folder already exists:

The current implementaion stats it, then checks if its a directory. The proposed implementation tries to make the folder, fails, stats the folder, then checks if its a directory. So the proposed implementation is marginally slower.

In the case the folder doesn't exist, but the path to the folder exists:

The current implementation tries to stat it, fails, stats its parent, checks if its parent is a directory, and then calls mkdir on the new folder. The proposed implementation just creates the new folder. So the proposed implementation is marginally faster.

In the case multiple folders need to be created

The current implementation tries to stat each missing folder/check if they're directories and eventually calls mkdir on each missing folder. The proposed implementation tries to call mkdir on each missing folder, fails, and then eventually calls it successfully on each missing folder. At best, the proposed implementation seems to be marginally slower here, I would guess this is actually a more significant slow down though.

So to summarize:
Folder already exists: current implementation is faster
Only one folder needs to be created: proposed implementation is faster
Multiple folders need to be created: current implementation is faster

So I don't see a ton of advantage here.

EDIT: Took out "It seems to me like the multiple folders need to be created case is the most useful here as well, otherwise you can easily just use fs." Because as I thought about it that's not really true, the advantage of mkdirp is that you don't need to know how many folders you have to create (could be 0, 1, or many).

[jclem]

@damccorm It’s more about what I saw as some nice opportunities for maintainability improvements here, more than performance.

[damccorm]

It’s more about what I saw as some nice opportunities for maintainability improvements here, more than performance.

Cool, that sounds good. I agree its more readable and perf won't be hugely impacted, just wanted to make sure we weren't making these changes with the expectation that it would be.

[damccorm]

LGTM