posix: redo socket API to use conn

[miri64]

Depends on #3611 (merged), #3612 (merged), #3615 (merged) and #3660 (merged).

[BytesGalore]

oh that's not really what I want, sry. (see #3646 (comment))

[miri64]

From miri64@902d305#commitcomment-12756195:

how about:

case AF_INET:
case AF_INET6:
    s->domain = domain;
    s->type = type;
    if ((s->protocol = _choose_ipproto(type, protocol)) < 0) {
        res = -1;
    }
    break;

That's different from the current code in that this code segment will be entered even if the associated connection modules are not there. But I can put the EAFNOSUPPORT check for them to the top.

[BytesGalore]

ahh domain may be AF_INET or AF_INET6 even if CONN_IPV4 || CONN_IPV6 are not available, the forget my noise :)

[miri64]

Rebased to current master

[miri64]

Rebased to current master and dependencies.

[miri64]

oh that's not really what I want, sry.

@BytesGalore what did you mean with that?

[miri64]

Rebased to current dependencies

[BytesGalore]

@BytesGalore what did you mean with that?

nvm. @OlegHahm pointed me that I commented on commits from this PR and not on the PR, so that was something I didn't wanted :)

[miri64]

Rebased to current master and dependencies.

[miri64]

And now for real.

[emmanuelsearch]

Maybe the READ ME could be more complete -- especially for programmers expected above a POSIX socket in this context. E.g. in EXAMPLE OUTPUT, indicate that ifconfig gives you the IP address, and say that the UDP port number used here is not to be confused with the port used for the shell indicated in USAGE. In USAGE, indicate that on native, no need to indicate a port.

[miri64]

In USAGE, indicate that on native, no need to indicate a port.

Since when? If you want to use a different tap interface apart from tap0 you need to inticate a port.

[emmanuelsearch]

In USAGE, indicate that on native, no need to indicate a port.

oops my bad. Ignore ;)

[miri64]

Update README.

[emmanuelsearch]

Looks good. ACK.

[miri64]

Squashed

[miri64]

Fixed issues with coap app and added low-memory boards to BOARD_INSUFFICIENT_MEMORY.

[miri64]

Another (bug) one bites the dust. This time it was Arduino with its pesky 16-bit ints again

[miri64]

(hopefully) last one. tests/coap is due to gnrc too big for stm32f0discovery.

[miri64]

Let's give it one last round for today, shall we (blacklisted msp430 and avr boards for tests/coap again, since they have problems with libcoaps :x notation in structs and added include to kernel_types.h to sys/posix.h to have ssize_t on msp430)

[OlegHahm]

The last comment sounded like a question...

[miri64]

It was a rhetorical question intended to mask my frustration about the last minutes of this PR.