Open stevehalliwell opened 3 years ago
This also or similar mechanism could account for profiling.
adding logic to readpacket is scary, it is the perf critical path.
We could do this now with the tracing elements that exist, compare tracing data against full walk of the compiled program.
with VM_STATS on, we actually dump out the full coverage of every opcode. We could in tests turn that on and dump those stats after each test for later cross compare.
VM instrumentation of some form would be required but once in, if cheap enough, could track which ops in which chunks are run. Useful for test coverage and dead code identification.