Have you done benchmarks like the Unicode Changes PEP has?
I don't expect it to have any interesting impact. What benchmarks do you have any mind?
I would personally be satisfied with just timing the regression suite before and after, though others (Unladen Swallow?) may have more representative workloads.
My biggest concern is that any memory increase (let alone 6%) may matter if it forces the use of extra cache lines.
I've just timed running the test suite (in non-debug mode) and the result (user CPU time) is not significantly different: 3m5s without the patch, 3m7s with it. pybench also shows similar results. Without patch: Test minimum average operation overhead ------------------------------------------------------------------------------- BuiltinFunctionCalls: 59ms 60ms 0.12us 0.164ms ComplexPythonFunctionCalls: 60ms 62ms 0.31us 0.275ms PythonFunctionCalls: 55ms 56ms 0.17us 0.165ms PythonMethodCalls: 74ms 74ms 0.33us 0.099ms ------------------------------------------------------------------------------- Totals: 248ms 252ms With patch: Test minimum average operation overhead ------------------------------------------------------------------------------- BuiltinFunctionCalls: 59ms 61ms 0.12us 0.163ms ComplexPythonFunctionCalls: 59ms 60ms 0.30us 0.273ms PythonFunctionCalls: 55ms 55ms 0.17us 0.164ms PythonMethodCalls: 77ms 78ms 0.35us 0.099ms ------------------------------------------------------------------------------- Totals: 251ms 254ms Regards Antoine.