Message 285667 - Python tracker (original) (raw)

I checked the effect of individual patches:

tp_fastcall-2 + print + struct + decimal: 16.3 ms +- 0.6 ms tp_fastcall-2 + struct + decimal: 21.2 ms +- 0.3 ms tp_fastcall-2 + print: 16.7 ms +- 0.2 ms

Oh wow, I didn't expect that print would be the bottleneck of this benchmark!? There is a single print() in the hotcode of bm_telco:

print(t, file=outfil)

Maybe it's the expensive price of creating a temporary dictionary?