try adding codspeed benchmarks

[davidhewitt]

Closes #3407

I'm curious to see if this works / how it looks like. If it's annoying or noisy we can turn it off.

[codspeed-hq]

CodSpeed Performance Report

Congrats! CodSpeed is installed 🎉

🆕 78 new benchmarks were detected.

You will start to see performance impacts in the reports once the benchmarks are run from your default branch.

Detected benchmarks

• list_get_item (22.5 ms)
• list_new (33.6 ms)
• call_0 (1.3 ms)
• sequence_from_list (347.8 ns)
• dict_new (132.4 ms)
• call_method_0 (3.7 ms)
• tuple_new_list (7.1 ms)
• iter_dict (57.4 ms)
• dict_get_item (88.5 ms)
• extract_btreemap (219.9 ms)
• iter_tuple (27.8 ms)
• tuple_new (33.4 ms)
• tuple_get_item_unchecked (15.5 ms)
• extract_hashmap (111.8 ms)
• mapping_from_dict (318.3 ns)
• tuple_into_py (4.7 µs)
• tuple_to_list (3.6 ms)
• list_via_downcast (126.1 ns)
• enum_from_pyobject (41.8 µs)
• not_a_list_via_downcast (126.1 ns)
• not_a_list_via_extract_enum (26.7 µs)
• f64_from_pyobject (272.2 ns)
• clean_acquire_gil (5 µs)
• clean_gilpool_new (677.2 ns)
• dirty_acquire_gil (6.3 µs)
• list_via_extract (273.9 ns)
• not_a_list_via_extract (2.5 µs)
• extract_bigint_small (4.3 µs)
• collect_generic_iterator (333.9 ms)
• extract_bigint_big_negative (8.7 µs)
• extract_bigint_big_positive (7.8 µs)
• extract_bigint_extract_fail (17.5 µs)
• identify_object_type (27.2 µs)
• sequence_from_tuple (320 ns)
• extract_bigint_huge_negative (49.7 µs)
• set_new (71.4 ms)
• extract_bigint_huge_positive (43.5 µs)
• extract_hashset (108.6 ms)
• extract_btreeset (97.4 ms)
• extract_str_extract_success (510.6 ns)
• extract_str_extract_fail (2.4 µs)
• extract_str_downcast_success (422.2 ns)
• extract_str_downcast_fail (266.1 ns)
• extract_float_downcast_success (488.9 ns)
• extract_float_extract_fail (17.7 µs)
• extract_int_extract_success (706.7 ns)
• extract_int_downcast_fail (266.1 ns)
• extract_int_downcast_success (861.1 ns)
• getattr_intern (4 µs)
• test_none_py (17.4 µs)
• test_empty_class_init (31.9 µs)
• ordered_dunder_methods (4.1 µs)
• test_simple_py (27.4 µs)
• extract_int_extract_fail (17.4 µs)
• test_simple_args_py (27.7 µs)
• test_none_rs (17.2 µs)
• test_simple_rs (27.4 µs)
• test_empty_class_init_py (24.8 µs)
• test_simple_kwargs_rs (36.5 µs)
• getattr_direct (9.2 µs)
• err_new_restore_and_fetch (10.8 µs)
• test_simple_args_rs (32.6 µs)
• err_new_without_gil (1.6 µs)
• test_simple_args_kwargs_py (34.2 µs)
• test_simple_kwargs_py (34.8 µs)
• first_time_init (74.6 µs)
• ordered_richcmp (4.1 µs)
• test_args_kwargs_rs (22.9 µs)
• drop_many_objects (55.3 µs)
• decimal_via_extract (22.5 µs)
• iter_list (37.4 ms)
• list_get_item_unchecked (19.5 ms)
• iter_set (58.7 ms)
• tuple_get_item (18.7 ms)
• test_args_kwargs_py (32.7 µs)
• extract_float_extract_success (332.8 ns)
• extract_float_downcast_fail (455 ns)
• test_simple_args_kwargs_rs (37.1 µs)

[davidhewitt]

Looks like the basics are working, I need to add codspeed criterion compatibility as per https://docs.codspeed.io/benchmarks/rust#from-criterionrs

[davidhewitt]

CodSpeedHQ/codspeed-rust#13

Also, it seems extremely slow to build the benchmarks 🤔

[davidhewitt]

Well, it built and ran! Took 63 minutes but I think that a lot of that is due to release builds, nothing being cached, and GHA runners being pretty slow. I'm going to merge this as an experiment and then we can revisit later whether this was a good idea, if it loads or blocks CI a lot we can remove.