Use fast-write-atomic instead of write-file-atomic

[mcollina]

fast-write-atomic is 10-20% faster depending on the operating system,
mainly because it does not use fs.realpath() and so many promises.

This needs to be tested, especially on Windows.

I'm happy to move fast-write-atomic to the ipfs org if you want, I've put it on my profile for convenience of adding Travis etc.

See ipfs/js-ipfs#1785 for more details.

[achingbrain]

The commit message needs to be changed to be in line with the PL guidelines

Also pre 1.0.0 deps need to be specified with the ~ character instead of ^, guidelines, again.

Once that's done CI should be happier.

[dignifiedquire]

LGTM - mainly looked at what fast-write-atomic does under the hood, and that seems to match the minimal thing that we need to do to, to make things work.

[mcollina]

@achingbrain done.

[alanshaw]

One thing that write-file-atomic does that fast-write-atomic doesn't is it will queue up multiple writes to the same file and execute them in order. With fast-write-atomic there's no ordering like this.

[mcollina]

One thing that write-file-atomic does that fast-write-atomic doesn't is it will queue up multiple writes to the same file and execute them in order. With fast-write-atomic there's no ordering like this.

That was not part of the algorithm of the Go implementation.

[achingbrain]

@achingbrain done.

CI says:

⧗   input: 

src: use fast-write-atomic instead of write-file-atomic

fast-write-atomic is 10-20% faster depending on the operating system,
mainly because it does not use fs.realpath() and so many promises.

✖   type must be one of [build, chore, ci, docs, feat, fix, perf, refactor, revert, style, test] [type-enum]

Change src: to perf:...

[mcollina]

should be done now :D

[achingbrain]

LGTM.

There's recursion without process.nextTick in a couple of places in fast-write-atomic Do you think there's a chance of stack overflows on busy systems?

[mcollina]

There's recursion without process.nextTick in a couple of places in fast-write-atomic Do you think there's a chance of stack overflows on busy systems?

Can you send links? I think everything is taken care of.

[achingbrain]

Just these two places, I think:

• https://github.com/mcollina/fast-write-atomic/blob/ab82a19dabf169717b8ebf241ca539b24e023dbf/index.js#L30
• https://github.com/mcollina/fast-write-atomic/blob/ab82a19dabf169717b8ebf241ca539b24e023dbf/index.js#L41

[mcollina]

Those cannot overflow because open and write calls the callback asynchronously by design.

[alanshaw]

@pgte can we get your review, merge and release please 🙏

[alanshaw]

ping @pgte

[pgte]

Landed on v0.8.0.