<ronny>
anyone aware of the efforts in cython to support hyp - om wondering whether i should create certain utilities i want to make for pluggy on top of hyp or on top of cython (i want pypy speedups as well)
<antocuni>
ronny: fangerer is working on a cython backend for hpy
<antocuni>
I don't know much more than that
<mattip>
for pypy/graalpython, the best speedups would be to keep the pure python code.
<mattip>
You could try to add cython pure-python mode pxd files to get a cython speedup on cpython
<mattip>
the idea is to use type-annotations to help cython, while keeping the code in python
<ronny>
mattip: the code i want to create/speed up is function/method invokers, that given a bound method or a function, and a mapping thats a superset of the invoked calls with a subset (the argument handling of pluggy)
<ronny>
(this would be less of a pain if the stdlib itemsgetter api wasnt borked and slow/messy)
<mattip>
ronny did you finish the sentence? I am not following.
<mattip>
if you are looking to speed up function invokers, I am not sure cython will help, but benchmarking a prototype would be the best way to tell
<mattip>
funcion invokers? function invocation?
<mattip>
in general, we have many reports of PyPy CI runs being slown I hope adding cython into the mix not going to make the PyPy CI-experience even worse
<ronny>
mattip: sorry, the matrix bridge is acting up
<ronny>
The idea is to create objects and helper functions that speed up the hook calling of pluggy (major part of the pytest profiles) by factor of 2-5
<ronny>
Currently pluggy pulls out the arguments to one hook by fetching the items required from the argument dict, then unpacking them as positional arguments
<ronny>
The pulling out happens in a list compression
<mattip>
ok
<mattip>
I guess you are pretty sure cython will speed this up?
mattip has quit [Ping timeout: 268 seconds]
mattip has joined #hpy
<ronny>
mattip: both cython and mypyc did generate certain speedups (between 2 and 10 x, but the micro benchmarks where inconclusive