[LNT] Allow --use-perf=profile and --run-under to work together

[peter.smith]

Summary

The perf.py module invokes the run_under.mutateCommandLine member function before mutating its own command line. In a recent refactoring r314239 "litsupport: Rework test module support" the run_test.mutateCommandLine was renamed with a leading _ to make it private. This means that when --use-perf=profile (or all) and --run-under are used simultaneously, as suggested for benchmarking in the lnt manual, we get the error:

test-suite/litsupport/modules/perf.py", line 32, in mutatePlan
    run_under.mutateCommandLine)
AttributeError: 'module' object has no attribute 'mutateCommandLine'

For reference the mutatePlan function from perf.py is:

def mutatePlan(context, plan):
    script = context.parsed_runscript
    if context.config.run_under:
        script = testplan.mutateScript(context, script,
                                       run_under.mutateCommandLine)
    script = testplan.mutateScript(context, script, _mutateCommandLine)
    plan.profilescript += script
    plan.metric_collectors.append(
        lambda context: {'profile': context.tmpBase + '.perf_data'})

The patch just renames the _mutateCommandLine in run_under.py as I don't think that it should be marked private if run_under.py is calling it.

I've not been able to find an easy way to write a regression test for this. The existing run_under tests in the lnt unit tests run using a fake lit so I couldn't write a failing test. I'm an almost complete novice with lnt development so I'd welcome any suggestions for how to write a unit test.

[kristof.beyls]

It's clear from your description that the run_under _mutateCommandLine isn't a "private" function in the module as is, and therefore removing the under score makes sense to me.
Maybe it'd be better for the mutatePlan in perf.py to be implemented in a different way so that _mutateCommandLine really can be private in the run_under module.
I'm afraid I don't have enough experience here to know whether that would be a reasonable goal, implementable with reasonable effort.
I hope @MatzeB could share his opinion on this.

On writing a unit test for this - I also fear that @MatzeB may be the only one who's up to speed on the unit tests for the test-suite framework itself.

Despite all of that - I think it's reasonable to commit this patch as is to unbreak the breakage that was introduced by that past refactoring.

[peter.smith]

Committed under r335463 my apologies I forgot the Differential Revision commit that would automatically close this revision.

litsupport/modules/run_under.py

 """Test module to run benchmarks in a wrapper application. This is typically
 used to prefix the benchmark command with simulator commands."""
 from litsupport import shellcommand
 from litsupport import testplan
 
 
-def _mutateCommandLine(context, commandline):
+def mutateCommandLine(context, commandline):
     cmd = shellcommand.parse(commandline)
     run_under_cmd = shellcommand.parse(context.config.run_under)
 
     if run_under_cmd.stdin is not None or \
        run_under_cmd.stdout is not None or \
        run_under_cmd.stderr is not None or \
        run_under_cmd.workdir is not None or \
        run_under_cmd.envvars:
         raise Exception("invalid run_under argument!")
 
     cmd.wrap(run_under_cmd.executable, run_under_cmd.arguments)
 
     return cmd.toCommandline()
 
 
 def mutatePlan(context, plan):
     run_under = context.config.run_under
     if run_under:
         plan.runscript = testplan.mutateScript(context, plan.runscript,
-                                               _mutateCommandLine)
+                                               mutateCommandLine)