fix: correctly re-establishes pipe destinations

[reggi]

NODE-2172

I tried pretty hard to recreate this locally, and I ran into issues, I was also considering stubbing out the #pipe here and testing the createChangeStreamCursor function directly but that would mean exporting it which would make it public. I can't get this code to fire and it does not exist in master anymore.

It does seem verifiably wrong.

const x = ['pipe1', 'pipe2', 'pipe3']
> for (let pipe in x) { console.log(pipe) }
0
1
2
> for (let pipe of x) { console.log(pipe) }
pipe1
pipe2
pipe3

[mbroadst]

@reggi I think this ticket was about reenabling this test. Looks like you found a bug here, but were you able to get that test passing? Are there other tests we should include here to validate your fix?

[reggi]

@reggi I think this ticket was about reenabling this test. Looks like you found a bug here, but were you able to get that test passing? Are there other tests we should include here to validate your fix?

@mbroadst good learning here to do a search for the jira ticket in the codebase (which I don't normally do). I fixed the test, but in 3.6 it wasn't broken because of this pipe issue, it was the file /tmp/ dir not existing. Fixed this by setting it to a already existing dir, then cleaning up the file using this.defer, master may show different issues.

In the description I wrote about attempting to test this in => of switch, and I had spent some time on different solutions and came up blank. Any ideas? Is it necessary? This code is not in master anymore. It might not be worth it.

[emadum]

This fix LGTM for 3.6. Just to clarify, we also need to get this functionality working again in master. That might be a little more complicated, since we've pulled the Readable stream off the Cursor object. There are currently two skipped tests in master - the one you fixed here, and a new somewhat simplified one I wrote. We should try to get at least one of them passing.