https://pantsbuild.org/ logo
Powered by
#general
  • a

    adorable-hairdresser-18353

    08/04/2015, 7:34 PM
    this is with regards to 
    round_engine
    and 
    group_task
    whose tests are throwing errors due to new calls to get_options on task objects inside the two classes.
  • e

    enough-analyst-54434

    08/04/2015, 7:39 PM
    The key to seeting up options is the call to self.context()
  • e

    enough-analyst-54434

    08/04/2015, 7:39 PM
    That takes a parameter - 1st, to list the task types that should be options-initialized
  • e

    enough-analyst-54434

    08/04/2015, 7:40 PM
    If you're not passing that list of types - that's probably the issue
  • e

    enough-analyst-54434

    08/04/2015, 7:45 PM
    Sorry for the slow link, its for_task_types right here: https://github.com/pantsbuild/pants/blob/master/tests/python/pants_test/base_test.py#L149 The TaskTestBase for example uses this to setup options for the type of task under-test for subclasses of it. You don't have the advantage of that help directly subclassing BaseTest
  • r

    rough-minister-58256

    08/04/2015, 7:55 PM
    yeah, self.context(for_task_types=[YourDummyTask])
  • a

    adorable-hairdresser-18353

    08/04/2015, 8:02 PM
    shouldn’t 
    self.set_options_for_scope('', explain=False)
    give it the global scope of options?
  • e

    enough-analyst-54434

    08/04/2015, 8:03 PM
    No - just that single global option. The machinery in BaseTest.context is what sets up all the rest of the global (and optionally task and subsystem) default options.
  • e

    enough-analyst-54434

    08/04/2015, 8:04 PM
    In fact. 
    self.set_options_for_scope('', explain=False)
    is probably no longer needed
  • e

    enough-analyst-54434

    08/04/2015, 8:05 PM
    That should be defaulted by the call to self.context()
  • a

    adorable-hairdresser-18353

    08/04/2015, 8:06 PM
    ah good to know
  • e

    enough-analyst-54434

    08/04/2015, 8:07 PM
    Yeah - confirmed - that line nuked on master still has all engine package tests passing
  • a

    adorable-hairdresser-18353

    08/04/2015, 8:13 PM
    So in 
    test_round_engine
    , there is a call to super method 
    install_task
    that eventually calls goal’s 
    install
    method, using the goal’s option scope instead of context. so it’s not the task I need to look for but the goal. However, in this case the goal is just a string.
  • a

    adorable-hairdresser-18353

    08/04/2015, 8:13 PM
    The context created in setup isn’t used?
  • e

    enough-analyst-54434

    08/04/2015, 8:15 PM
    Its definitely used by the test in test methods - looking at the install_task bit...
  • e

    enough-analyst-54434

    08/04/2015, 8:17 PM
    OK - basically there is a scope-mismatch with how things are currently done in that test and how you need to do them now that you need task option defaults plumbed
  • e

    enough-analyst-54434

    08/04/2015, 8:17 PM
    For example: def test_cli_goals_deduped(self): self.install_task('task1', goal='goal1') self.install_task('task2', goal='goal2') self.install_task('task3', goal='goal3') self.engine.attempt(self._context, self.as_goals('goal1', 'goal2', 'goal1', 'goal3', 'goal2'))
  • e

    enough-analyst-54434

    08/04/2015, 8:18 PM
    the problem there - is you cannot build a complete context with defaulted options for all 3 tasks until after all 3 tasks are installed
  • e

    enough-analyst-54434

    08/04/2015, 8:18 PM
    However self._context was built up in setUp before any tasks were installed
  • e

    enough-analyst-54434

    08/04/2015, 8:18 PM
    So that all needs a fundamental change
  • a

    adorable-hairdresser-18353

    08/04/2015, 8:18 PM
    okay that makes so much more sense!
  • a

    adorable-hairdresser-18353

    08/04/2015, 8:19 PM
    thanks for the help! I’ll have to look into other tests to see how they do it because I was running into a similar thing with 
    test_group_task
  • e

    enough-analyst-54434

    08/04/2015, 8:20 PM
    k - yeah - you might end up having a util pop out of this that creates N fake tasks and returns a context
    👍 1
  • r

    rough-minister-58256

    08/04/2015, 8:33 PM
    thanks for the guidance on the pantsd RB @enough-analyst-54434 .. will partition that up as suggested. btw, regarding ‘while True’ vs ‘while 1’ - this has always been my basis for the preferring the latter in 2.x: http://stackoverflow.com/questions/3815359/while-1-vs-for-whiletrue-why-is-there-a-difference + benchmark: http://pastebin.com/60xPmLiA
  • e

    enough-analyst-54434

    08/04/2015, 8:35 PM
    Aha - interesting
  • r

    rough-minister-58256

    08/04/2015, 8:35 PM
    the joys of python
  • e

    enough-analyst-54434

    08/04/2015, 9:06 PM
    Travis is in OSX incident mode today, so feel free to cancel OSX CI jobs to unblock the linux queues for PRs as-needed: https://travis-ci.org/pantsbuild/pants-for-travis-osx-ci
  • h

    happy-kitchen-89482

    08/04/2015, 9:52 PM
    The first two comments on that perfectly-formed Stack Overflow question are everything that is wrong with the internet.
  • e

    enough-analyst-54434

    08/04/2015, 9:53 PM
    Heh
  • c

    cool-thailand-74745

    08/04/2015, 10:29 PM
    @enough-analyst-54434 i'm going to get https://github.com/pantsbuild/pex/pull/142 merged and then cut 1.0.2
1...135136137...2420Latest