Re: [Discuss-gnuradio] make check failures in inband on Cygwin

[Don Ward]

Thanks, George.

> These pass on unix machines, I've never tried running any of the inband 
> stuff in cygwin... but I don't see how it would change.  If the mblock QA 
> code passes, I would expect the inband QA code to pass in cygwin. It's 
> simply utilizing m-blocks.
>
> Can you go to the 'mblock' directory and run a 'make check' ?

Yes, make check works in the mblock directory.

> In short, nothing will be affected by inband failing to pass a check at 
> the current point in time if you're not using any of the inband code... 
> which I don't think anyone else in the world other than developers are.

Ok.  The only thing I lose is 'make check' on everything after inband.

I have found what is happening:  It appears (to the linker, at least) that 
test_inband.exe depends on libusrp_inband-qa.dll but not on 
libusrp_inband.dll.  But the code to initialize s_ALL_PROTOCOL_CLASSES (in 
mb_protocol_class.cc) is contained in usrp_{server,interface}_mbh.cc which 
is in libusrp_inband.dll.  Since libusrp_inband.dll is not used (as far as 
the linker can determine) by test_inband.exe, s_ALL_PROTOCOL_CLASSES is 
never initialized.  The "uncaught exception of type mbe_mblock_failed" is a 
result (I don't understand how) of the 'throw' on line 40 of mb_port.cc 
(after mb_protocol_class_lookup failed).

Somehow we need to be sure that any code that relies on initializations 
performed in libusrp_inband.dll actually references libusrp_inband.dll in a 
way that the linker can recognize.

Note that adding $(BUILT_SOURCES) to libusrp_inband_la_SOURCES helps but is 
not enough---apparently there is something else in libusrp_inband.dll that 
is needed.

-- Don W.