A Proof of concept patch that 'fixes' #67

[ciaranj]

• I made various attempts to get flock working properly on windows for sinopia.
  They all seem to result in the process ending un-expctedly :/
• So in the short term this patch does seem to address a bunch of realworld issues people are seeing, however
  it also causes 3 new test breaks, so I doubt you want to actually merge it.

If I have time to cycle back to this problem I might look at re-jigging it
again

[rlidwka]

fs.exists is almost always a bad idea because of a potential race condition.

Can you try a rename-unlink-rename instead?

fs.rename(file1, file2, function(err) {
    if (err) {
        // windows
        fs.unlink(file2, function(err) {
            fs.rename(file1, file2, function(err) {
                cb()
            })
        })
    } else {
        cb()
    }
})

[ciaranj]

Interesting, I've tried that, and the failures come back :/ I should note that if any race exists, it is within the code, I trigger it with a single request for a package that sinopia has never seen before. (no other active clients.)

Whilst applying your suggested fix, I noted: https://github.com/rlidwka/sinopia/blob/master/lib/local-fs.js#L59 should that have an 'else' keyword before the 'stream.emit('success')' statement [incidentally, that doesn't fix the problems :/]

[rlidwka]

Whilst applying your suggested fix, I noted

I think you just found why our travis build fails randomly. This is great. :)

I'll try to get a windows box to debug those issues.

[ciaranj]

sweet. drive-by-bug-fix ;)

[rlidwka]

Fixed in a854d07 (sinopia@0.13.2), see #67 (comment) for details . Sorry for the delay.