fixed bug in merge points where deleted points would be replaced in merge
batch_filter doubles all TimingSummary counts in its batch.
This occurs because the batch.crop simply copies profiling_stats.
Merging batch and node_batch ends up merging the profiling_stats with itself.
My quick hacky solution:
Simply add a flag for if we want to merge profiling stats.
A potentially better solution:
Require process to create and return a batch. This way the node_batch would
not contain any upstream profiling_stats, and would only add its own timings
to the batch.
This would require modifying every subclass of BatchFilter but would be more
fitting with the design of node_dependencies.
batch_filter doubles all TimingSummary counts in its batch. This occurs because the batch.crop simply copies profiling_stats. Merging batch and node_batch ends up merging the profiling_stats with itself.
My quick hacky solution: Simply add a flag for if we want to merge profiling stats.
A potentially better solution: Require process to create and return a batch. This way the
node_batchwould not contain any upstreamprofiling_stats, and would only add its own timings to the batch. This would require modifying every subclass ofBatchFilterbut would be more fitting with the design of node_dependencies.