exit on null for target output #50
Conversation
R/race-wrapper.R
Outdated
| @@ -518,6 +518,7 @@ execute.experiments <- function(experiments, scenario) | |||
| scenario$targetRunnerParallel(experiments, exec.target.runner, | |||
| scenario = scenario, | |||
| target.runner = target.runner) | |||
| if (is.null(target.output)) {stop()} | |||
There was a problem hiding this comment.
This should give an irace.error() that mentions targetRunnerParallel.
Also, I'm considering that maybe we can get rid of the exec.target.runner parameter and use lapply() to call check_output_target_runner() after targetRunnerParallel returns. This means that the targetRunnerParallel does not benefit from automatic retries but you are anyway not using that feature and the other users of targetRunnerParalle also don't do that.
There was a problem hiding this comment.
Also, I'm considering that maybe we can get rid of the exec.target.runner parameter and use lapply() to call check_output_target_runner() after targetRunnerParallel returns. This means that the targetRunnerParallel does not benefit from automatic retries but you are anyway not using that feature and the other users of targetRunnerParalle also don't do that.
Yeah but would that require a major version bump because it would not be backward compatible (even if no one is using exec.target.runner)?
DE0CH
force-pushed
the
exit-target-output
branch
from
bd81fdd to
066cf59
Compare
This is mostly needed for auto-optimization/iracepy#22 because error from the user defined function would not propagate from a python function
targetRunnerParallelto the embedded R instance. If I raise an error,rpy2will just swallow the error and return NULL instead. I am not sure if this is a feature or a bug. But in any case, we need to stop irace at that point otherwise it will just go on and fail somewhere else, alerting the user of an "internal bug" and taking down the python interpreter with it due to auto-optimization/iracepy#27.