rethink distribution

[nthallen]

In our homogeneous QNX system, distribute could just loop through EXP_NODES and make distribution. Now we have the added complication that EXP_NODES may represent multiple architectures. Note that so far we support:

• Cygwin64
• Linux x86_64
• Linux armv7l Not all nodes will be directly accessible (as before), but a GSE system can reasonably be expected to be able to distribute locally and to the flight computer at least, and those are very likely to be two different architectures.

Ideally, we could have out-of-tree builds with appgen setting up everything in build dirs.

[nthallen]

As things currently stand, independent builds for different architectures will cause command clients connecting across architectures to fail, since the server and client must be compiled from the same source file, which is generated from the .cmd inputs. I see a couple options:

• Promote the *cmd.cc files into SRC rather than OBJ, and get them into the repository
  • That is pretty much guaranteed to fail because the file will be regenerated
• Change how the command configuration string is calculated (not date/time based, but based on the contents of the sources)

The latter solution has been implemented sometime prior to today, 2/8/20

[nthallen]

For distribution via scp, need to resolve how permissions will work. If we distribute to flight@$FlightNode and expect to use ssh keys, then /home/flight cannot be g+w. That complicates various operations, like moving flight.sh.log into a run or setting up a new experiment. What are the alternatives?

• Create another user or users for distribution that are members of flight and never login as flight.
• anything better?