Create `utf.r2py` about utf

[aaaaalbert]

Alternatively, how about a seash module?

[aaaaalbert]

I'll summarize discussions with @lukpueh about this issue.

1. This is supposed to be a tool for checking the correctness of code running inside a remote sandbox. Thus, we consider mostly the repy_v2 and seattlelib_v2 tests for the candidate set.
2. The tool should not attempt to test the nodemanager controlling the test sandbox, nor should it test the interactions performed by seash or other components. These are orthogonal goals (see also SeattleTestbed/nodemanager#120).
3. Obviously, the tool won't run pure-Python unit tests, nor will it run setup/subprocess/shutdown scripts as they (mostly) depend on Python functionality.
4. #pragmas and command-line options should be respected as usual. (Reuse functionality from utf.py where possible!)

What's mentioned above (having a utf.r2py that runs test on its own vessel, or perhaps a seash module for this functionality) has practical ramifications: The usual protocol is that if a test case prints anything, the test is considered to have failed. utf.py can use plain output redirection to check that; if you run on a sandbox, you can either redirect log using a security layer above the test case, or inspect the vessel log after the test ran. The former technique is probably what a utf.r2py would do, the latter is apt for a seash module.

I think that overall, the seash module approach has the benefit of being simpler and less error-prone, as there is less code running on the vessel and the number of things that could go wrong is reduced. Potential non-transparency issues in dylink or encasementlib are avoided. On the negative side, checking for output might get more complicated due to --execinfo and such.

[lukpueh]

+1 for the seash module, thinking DRY this sounds more sensible than porting utf.py to repy.

Also I think we could and should take it a step further and decouple the seash module from seash, to run said unit tests programmatically. This could be achieved by updating and using the experimental experimentlib.r2py, where @aaaaalbert has already implemented a lot of the required functionality.
The workflow could be something like this:

1. Locally run: python utf.py -f <test.r2py> --keys <keys> --vessel <vessel>
2. utf.py executes python repy.py experimentlib.r2py, e.g. in a subprocess
3. experimentlib.r2py runs the seash module
4. the module uploads the the test file to the vessel, executes it, downloads the log and feeds it to utf.py's checking routine - verify_results

And here is what we have to do:

• Implement upload--run--download-log--reset-log seash module
• Modify experimentlib.r2py to use seash modules
• Modify utf.py to accept additional arguments for remote execution and to execute experimentlib.r2py
• Strip away output generated by verbosity options, e.g. --execinfo
• Somehow figure out which part of the output was stdout and which was stderr (this might be difficult)
• Make the output available to utf.py's verify_results function

[aaaaalbert]

Thumbs up for thinking this through from the user's perspective too!

I should add that "seash modules" are really implementations of additional seash commands, not modules in a Python sense. Thus, you can't really use a seash module in other code; I did however write a few seash modules to pull in functionality from other code! (And indeed, seash's overall engineering is a DRY problem in itself, see SeattleTestbed/seash#103).

The good news is that we don't require any seash module anyway --- experimentlib.r2py has most of the required functions already, so we could use it for your proposed modifications to utf.py almost any time.