Added feature list for K64F

[bogdanm]

Without this fix, some required FEATURE_ directories are not included,
which breaks compilation for existing code.

[bogdanm]

@sg- @geky @screamerbg

[jupe]

@kjbracey-arm @SeppoTakalo

[sg-]

@mbed-bot: TEST

HOST_OSES=windows
BUILD_TOOLCHAINS=GCC_ARM,ARM,IAR
TARGETS=K64F

[mbed-bot]

[Build 485]
SUCCESS: Building succeeded and tests were run! Be sure to check the test results

[sg-]

@mbed-bot: TEST

HOST_OSES=windows
BUILD_TOOLCHAINS=GCC_ARM,ARM,IAR
TARGETS=K64F

[sg-]

TLDR: IPV6 and CLIENT need to be reverted. Can re-run tests with only IPV4

A few things about this default inclusion and the current state:

1. The atmel RF driver is moving out of tree. IPV6 will fail to compile in this case until a weak implementation of the phy is implemented. Seppo is working on this
2. CLIENT cannot be enabled currently without IPV6 :(
3. CLIENT is enabled by an application that uses it and IPV6 is enabled by a transceiver that supports it

Another side effect of config:
The way FEATURE_IPV6 uses "SOMETHING" and (4) and {x,x,x,x,x,x,x,x} breaks exports for uvision and iar. This is a problem with using -D in the command line and also assemblers seem to have a limit on command line length.

@0xc0170 is working on moving these into a file which gets included rather than -D on the command line

[bogdanm]

CLIENT is enabled by an application that uses it

Then this is a dependency, not a feature.

[sg-]

For the time being until it removes its dependency on IPV6 but that is just a matter of doing some work, then the feature CLIENT can go away and it should be just a standard include such as CFSTORE

[bogdanm]

I'm not sure what you mean by a "standard include". It just feels weird to say that an application enables client, since what the application really does is use the client library, thus describing a dependency (would you say that an application "enables" mbed-os instead of saying that it depends on mbed-os?). In the past, we have concluded that FEATURE_ is not a mechanism for deplacing dependencies, did that change? It might be that I'm not aware of the latest developments in this area.

[kjbracey]

Don't see an obvious reason for CLIENT->IPV6 dependency. Unless mbed-trace is wrongly still in IPV6? Or maybe ns-event-loop and HAL are in there? They shouldn't be.

Tools will need to handle appropriate escaping for config strings - presumably different shells will need different escaping mechanisms. Moving into a file would help.

[SeppoTakalo]

I think CLIENT is depending only of IPV6 because of event loop. (maybe event tracing library).
But if we move them out, whose job it is to start the event-loop?

[kjbracey]

ns_hal_init starts it. Responsibility was moved out of mbed-mesh-api (which should be in IPv6), to allow use of HAL/event-loop without nanostack.

[kjbracey]

To clarify, any component who wants it, calls ns_hal_init. Currently mbed-client's connector and mbed-mesh-api both call this. Apps could also call it if they wanted to get in first with a bigger nsdynmemlib heap size or different error callback.

[SeppoTakalo]

Anyway, I would propose that we leave the "dependency" as it is, and solve it in Oulu.
So for K64F, just enable IPV4.
Leave it for us to cut the CLIENT<-->IPV6 dependency once we get the build working.

And @bogdanm , for me, it looks like FEATURE_X is the only way of dependency chain right now. If you guys have agreed something else, then this might be a workaround made by Sam&Mihail to get partner workshop running. And it is probably up to rest of the teams to continue implementing the dependencies you are talking about.

[bogdanm]

Cc @jaustin

[mbed-bot]

[Build 490]
SUCCESS: Building succeeded and tests were run! Be sure to check the test results

[SeppoTakalo]

Modified the commit so that only IPV4 gets enabled. That is the only one currently having driver in the build tree.

[SeppoTakalo]

@mbed-bot: TEST

HOST_OSES=windows
BUILD_TOOLCHAINS=GCC_ARM,ARM,IAR
TARGETS=K64F

[mbed-bot]

[Build 492]
SUCCESS: Building succeeded and tests were run! Be sure to check the test results

[sg-]

👍

[bogdanm]

The initial intent of this PR was to at least make the client example compile. It did with my original changes. Is that still the case?

[SeppoTakalo]

Nope. it does not.
We are working for fix for toolchain to have proper handling of "features" and then we will add features to client example.