add version of Richards that uses super() #271
Conversation
|
The 3.12 failure looks like an issue building greenlet, not related to this PR. |
|
Yep, see #263 for 3.12+greenlet. |
|
Nice I will take a look |
|
Overall looks good, but why not just replace the Richards to use super()? cc @pablogsal |
I think the main reason is because this means Richards timing is no longer comparable to previous timings? (I'm not sure if this is an issue or not; I'm quite happy to just change Richards if that's ok with the maintainers.) |
pyperformance/data-files/benchmarks/bm_richards_super/run_benchmark.py
Outdated
Show resolved
Hide resolved
|
This looks good to me. I agree that adding (rather than replacing) is the way to go here. We can always remove the old benchmark once we learn more from this one. |
Currently, none of the benchmarks in pyperformance exercise
super()method calls significantly. But real-world modern object-oriented Python code often does (one source: Instagram server codebase uses lots ofsuper()method calls, some in hot paths.)To correct this gap in visibility, this PR adds a copy of the Richards benchmark that is modified to use
super()calls. Existing super-method calls that were done by naming the base class directly (Task.__init__(self, ...)) are modified to usesuper(), and the already-polymorphicTask.fnmethod is modified slightly such that overrides now call up to the base method usingsuper().If we accept that Richards is a roughly reasonable representation of some object-oriented Python code, I think this is a similarly reasonable representation of the real-world use of
super()in such code.Thanks to @pablogsal for this suggestion.