Carl Brown

Ok, so good news and bad news. The good news is, I've discovered via brute force that the regression happens between the 2017-05-30 and 2017-05-31 snapshots (e.g. the tests pass...

Quite possibly. I'm just really not sure how a change to `URLSession` (happening on one queue) could possibly cause `Glibc.close(socket)` not to interrupt `Glibc.recv(Socket)` on a different queue.

I've discovered can fix this problem by applying a change to `BlueSocket` where we call `Glibc.shutdown()` when `close()`ing a socket that `isConnected` in addition to `isListening`. (c.f. https://github.com/carlbrown/BlueSocket/commit/f5dc916bf707b12b11634a89655f92c3836f5051#diff-cec92f2c4685d05c3d246f2bb8a42733L3324). This means...

You have to remove the `link` line from the module.map for CHTTPParser. See the `0.5.0` tag: https://github.com/IBM-Swift/CHTTPParser/commit/2f5dc431b51544b4c1bd322854f3fbd76abe35b4#diff-f73852fef28faf8dabf66d2f28db6998L20

There's also [a tar file](https://github.com/swift-server/http/files/1160419/issue_22_complete_source.tar.gz) in a comment above (https://github.com/swift-server/http/issues/22#issuecomment-316462312) that has the whole directory structure with everything checked out, if that's easier.

Any status on this @pushkarnk @ianpartridge ?

@pushkarnk No, that means the socket was shutdown while the `accept()` system call was listening for new connections (which invalidates the `socket` passed as an argument). You'll see that message...

What snapshot are you seeing them passing with? What changed since 29 days ago when you said you saw the failures?

I put comments from @helje5 into two other issues so we can track them explicitly. They're both valid points, so I explained why I did what I did, and I'm...