Fixes for multiple issues with computing analytic and approx derivatives for components with a mix of serial and distributed variables.

[naylor-b]

Related Issues

• Resolves Distributed component hangs during check_partials. #2137
• Resolves Distributed component hangs during check_totals. #2138
• Resolves Totals computed via compute_totals of a serial output wrt a distributed input are not gathered even though get_remote is set to True. #2139
• Resolves Component-fd hangs for component with mixed serial & distributed inputs and outputs. #2142

Backwards incompatibilities

• Users are no longer allowed to compute derivatives for a component having distributed input(s) with serial output(s) unless they use the matrix free API (so they must define either compute_jacvec_product or apply_linear).

New Dependencies

None

[Kenneth-T-Moore]

Are we keeping this permanently in API? I guess it's ok, and it is easier to just set the env var now.

[naylor-b]

Yeah I just put this in the api file so you and I could activate wing debugging with a single line, and I don't know about you but I often want to switch back and forth between wing debugging and just dumping stuff to stdout with prints, so it's easier to just precede my run command with WING_DBG=1 or not to control whether wing takes over or not. It's not a big deal though and I can take it out of the api file if people don't think it should be in there.

[coveralls]

Coverage decreased (-0.4%) to 88.889% when pulling 5fc7914 on naylor-b:dist_bugs into 827ef22 on OpenMDAO:master.