Josh Rosen

I've commented on a Scalariform PR to push for making incremental progress here: if Scalaiform can get to a point where the parser doesn't crash, even if preservation of trailing...

In Spark, https://github.com/apache/spark/pull/24905 may partially mitigate this by significantly reducing the number of calls to `fastestInstance()`.

@m-e-conroy I'm running into a similar issue, where updates in the Ace editor don't seem to update the associated `ng-model`. Did you ever find a solution to this problem?

Cross-reference (for easier discoverability): this is related to #98.

Another workaround: if you don't actually need / want XML reports, pass the `-P:scapegoat:reports:none` compiler flag to disable all report generation. This seems to bypass the parts of the code...

I think that SBT runs tests in parallel by default; Spark's [SparkBuild.scala](https://github.com/apache/incubator-spark/blob/master/project/SparkBuild.scala#L102) contains a line to disable parallel tests: ``` scala // Only allow one test at a time, even...

> Looking at this again, I think I should extract the predicate for classes which can't be loaded to a separate change. It should be generally helpful because the cost...

I think that we should skip logging of these metrics in JsonProtocol when push-based shuffle is disabled, so I've filed https://issues.apache.org/jira/browse/SPARK-42203

Process-wise, I think the right thing to do here is for us to work with the Hive project to publish Spark's fork of Hive via the official Hive release process....

Implementing the confusion matrix could be difficult because it looks like cuda-convnet's GPU workers only return the per-batch loss rather than the actual misclassifications.