I've added two pieces here: first is a `defer` block right after the creation of the `socketDescriptor` variable. If it's non-nil and not `Socket.SOCKET_INVALID_DESCRIPTOR` then the socket is closed pronto. Secondly, I'm explicitly setting this variable to `nil` when it's copied into `self.socketfd`. From that point on, any failure or error thrown will not result in the socket being orphaned, since it's referenced by the `Socket` class instance, and will be cleaned up in its `deinit` implementation.
- Now uses an unsigned type to perform the bit-shift operation and then casts to signed type using `init(bitPattern:)`.
- Added a unit test case which ensures fd values from zero to (arbitrarily) 128 can all be inserted and removed from the fd_set.
- `fd_set` API is now much nicer and will compile about as close to the C implementation as we can get— especially on release builds, with compiler inlining.
- Socket structure's `readBuffer` now initializes all its bytes to zero, not just the first.
- More idiomatic/concise return expression in `Socket.wait(for:timeout:waitForever:)` (don't build an output array manually, we have `filter(_:)` for that).
- Various non-Linux code to set the `SO_NOSIGPIPE` option has been tucked into a private function now, still containing the compilation conditional. It's a no-op on Linux, and will be inlined by the compiler (since it's private) and the calling code looks much more straightforward.
- BUGFIX: calls to `gai_strerror(_:)` take the result of the `getaddrinfo()` call as a parameter, NOT `errno`.
asAddr funcs were passing unsafe pointers outside of the callbacks in which try should be used
Using callback to make all the access within these callbacks
Fixes bug where IPv6 address was 'cast' to sockaddr but actually was zero'd beyond the length of sockaddr_in and packets written to the wrong address
Jim Dovey
Bill Abt
ShihabMehboob
Bill Abt
Fred Cox
David Jones
Quan Vo