Add NetworkInterface::get_default_instance()

[kjbracey]

Provide an initial framework to make it easier to find a default network
interface.

This is the germ of a concept. Follows the same pattern as just established for EMAC::get_default_instance() and OnboardNetworkStack::get_default_instance().

Envisaged use:

NetworkInterface &net = NetworkInterface::get_default_instance();
net.connect();

// for advanced users:
WiFiInterface *wifi = net.wifiInterface();
if (wifi) {
    wifi->do_something_wifiy();
}

[kjbracey]

I've put this on top of feature-emac, but it could be done without the feature-emac stuff to master as-is. Only difference would be the condition for selecting EthernetInterface - it would be if FEATURE_LWIP is enabled and LWIP_ETHERNET is 1.

[sg-]

what does MESH, LOWPAN, ETHERNET, ... 0x2345 mean?
Why does Wi-Fi not get a default instance like MESH and THREAD?

[0xc0170]

@kjbracey-arm Please fix travis and jenkins CI

[kjbracey]

The 0x2345 is just magic to let you do a "string comparison" in the C preprocessor.

#define CONF  FOO
#define FOO 0x2345
#if CONF == FOO   // evaluates as 0x2345 == 0x2345: true

#define CONF BAR
#define FOO 0x2345
#if CONF == FOO  // evaluates as 0 == 0x2345: false

Aim is to make this namespace open-ended - new code/systems can invent just new configuration names. Easy connect uses the same mechanism but closed, with:

#define ETHERNET 1
#define MESH 2
#define WIFI 3
#if CONF == WIFI

Wi-fi doesn't get a default instance because we don't (yet) have WiFiInterface:;get_default_instance(). If we define that, we get it. It would make sense to do so.

(We also don't have NanostackRfPhy::get_default_instance() - needs to be done).

I'm liking this pattern, but my main concern is how to indicate at compile time rather than link time that each of those definitions exist, to help probing for test etc.

[0xc0170]

one endif missing here (see Travis failure)

[lauri-piikivi]

I like the basic approach. Is the advanced usage (wifyish in the original comment) still in plans?

Should we have a getInterfaceType() method, which would return if the interface is actually Wifi or Mesh etc? Thinking about the device that have e.g. Wifi and cellular, the app might want to know and act on the type of interface (cast the interface to correct class and use the methods?).

[kjbracey]

The "wifyish" isn't that advanced, but yes.

The "xxxInterface" methods function as the "get type" - it's equivalent to C++ dynamic_cast, returning NULL if it's not that sort of interface, so that example is checking if it's wifi.

If we had run-time type information enabled in the compilers that could have been this:

WiFiInterface *wifi = dynamic_cast<WiFiInterface *>(&net);
if (wifi) {
    wifi->do_something_wifiy();
} else {
    // it's not a wifi interface
}

Little bit clunky. But one day in C++17(?) you could do:

if (WiFiInterface *wifi = dynamic_cast<WiFiInterface *>(&net)) {
     wifi->do_something();
} else if (CellularNetwork *cell = dynamic_cast<CellularNetwork *>(&net)) {
     cell->cellular_device()->do_something_new_api();
} else if (CellularBase *cellular = dynamic_cast<CellularBase *>(&net)) {
     cell->do_something_old_api();
} ...

[cmonr]

@kjbracey-arm Is this still in progress?

[kjbracey]

Yes, still work in progress.

[0xc0170]

@kjbracey-arm Please reopen with an update