Alright, here is a more radical rework of this PR:

• I removed the epsim_standalone sub-module all together -> when aiming for a minimal setup, no one would ever want a dedicated thread for this simple task anyhow...
• made the RD address configurable during runtime (as function param). This makes (i) the implemenation simpler and removes the need for ipv6_addr_from_str() inside of epsim, and (ii) this allows at least in theory, to run multiple simplified registrations in parallel...
• removed the default values for CORD_SERVER_ADDR and CORD_SERVER_PORT -> for the former there is just no good value, and for the latter it is simpler to just use COAP_PORT directly
• added a guard around the address warning in the application Makefile in case of make clean

Overall, this leads to a slimmer and more efficient implementation and hopefully makes epsim also easier to use.

I like the simpler rework! I removed the testing label because I'm retesting now. I plan to post a review in the next day or so.

good to hear, thanks!

[edited for clearer explanation of error]

I like the simple main loop rather than a separate thread. Overall this version uses around 15K less flash and 600 bytes less RAM on a samr21-xpro. I agree it was a little odd to include ipv6_addr_from_str() internally. It's also nice not to have the separate CORD_SERVER_ definitions.

I tested this quite a bit, and I'm worried that the cord_epsim module is a little too simple. For example, suppose the app is built with default setttings and the RD server is unreachable. In this case, cord_epsim returns an error code some of the time. The error occurs when gcoap tries to send a registration confirmably before the previous confirmable send has timed out. By default, gcoap supports a single open confirmable message (GCOAP_RESEND_BUFS_MAX). On average a CoAP confirmable message will timeout after 77s, but cord_epsim resends after 45s. If the user sets a longer, more realistic timeout, they will not see any error feedback in a terminal.

I propose adding a response handler to cord_epsim to maintain a state variable. The response handler maintains the variable simply based on reception of an ACK from the server. Then the main module can query it as desired. Something like this for cord_epsim:

static unsigned _state = CORD_EPSIM_OK;

static void _reg_handler(unsigned req_state, coap_pkt_t* pdu,
                          sock_udp_ep_t *remote)
{
    (void)remote;
    (void)pdu;

    if (req_state == GCOAP_MEMO_RESP) {
        _state = CORD_EPSIM_OK;
    }
    else {
        _state = CORD_EPSIM_ERROR;
    }
}

unsigned cord_epsim_state(void)
{
    return _state;
}

Thats a very good point and I fully agree. Will include those proposed changes.

addressed comments and added the internal state tracking. This should indeed be far more failure tolerant now.

Fixed setting _state to BUSY after sending the registration request and included some minor doc and style fixes. I hope all comments are addressed.

ACK! Tests fine. Please squash.

done

@haukepetersen, please review the Murdock failure. It looks spurious to me. Should we just trigger a rebuild? How to do that?

I rebased. lets see what Murdock says now.

I took the warning part on the localhost address out of the Makefile. Thinking about this, I think it should be sufficient to have the warning in the readme file. Also the RD address is printed when the application starts, so that users at least have a good chance in seeing the ::1 when configured as RD address.

@kb2ma would you mind to have one final look if my latest changes are ok with you? Thanks!

Looks good to me. Tests still pass. The user will see the address in the CLI. Thanks to @miri64 for finding the problem.

Please squash. Let's try this again.

squashed. lets hope for the best :-)