the biggest performance hotspot seems to be the use of twitter Eval, which has been integrated into CBT. It is bearable for everyday use, but 33% of the time of running the tests is lost to it according to my profiling investigations with VisualVM.
This was predicted and in fact we should probably use CBT's internal incremental compilation mechanism on generated source files instead of whatever Eval is doing and get rid of eval. This should probably solve this.
This mainly affects the ~0.5 s overhead for users when using the DynamicOverride feature.
the biggest performance hotspot seems to be the use of twitter Eval, which has been integrated into CBT. It is bearable for everyday use, but 33% of the time of running the tests is lost to it according to my profiling investigations with VisualVM.
This was predicted and in fact we should probably use CBT's internal incremental compilation mechanism on generated source files instead of whatever Eval is doing and get rid of eval. This should probably solve this.
This mainly affects the ~0.5 s overhead for users when using the DynamicOverride feature.