Collecting metrics on built-in dictionary usage

This repository contains a modified version of the CPython 3.11.4 interpreter (see README_cpython.rst).

All the dictionary routines in dictobject.c are modified to keep track of the number of calls and the total running time for each method inside the dict_metrics global table (see dictmetrics.c).

The modified interpreter was used to run the pyperformance benchmark suite. The csv reports generated were summarized with the analysis.ipynb script and are attached here in dict-metrics_combined.csv. The separate reports, generated per interpreter invocation, are not included here.

Note that the results of the benchmarks executed are not relevant here, and thus not included.

Extras

Several methods also include extra information:

method	extra 1	extra 2	extra 3
dictkeys_get_index	successful count	failing count
_Py_dict_lookup	successful count	failing count
unicodekeys_lookup_unicode	successful count	failing count
PyDict_GetItemWithError	successful count	failing count
_PyDictKeys_StringLookup	successful count	failing count
dictkeys_generic_lookup	successful count	failing count
insertdict	successful count 1	update count 2	triggered resizes count
_PyObject_GetInstanceAttribute	successful count
_PyDict_GetItemHint	successful count	failing count
PyDict_Contains	successful count	failing count
_PyDict_GetItem_KnownHash	successful count	failing count
_PyDict_DelItem_KnownHash	successful count	failing count
dict_get_impl	successful count
_PyDict_LoadGlobal	successful count	failing count
PyDict_GetItem	successful count	failing count
_PyDict_KeysSize	average size
PyObject_GenericGetDict	successful count
_PyDict_GetItemStringWithError	successful count	failing count
_PyDict_MaybeUntrack	untracked count
PyDict_Size	average size
unicodekeys_lookup_generic	successful count	failing count
shared_keys_usable_size	average size
_PyDict_HasOnlyStringKeys	return true	return false
estimate_log2_keysize	average size
calculate_log2_keysize	average size
dict_new_presized	average pre-size
_PyDict_SizeOf	average size
dict_sizeof	average size

Footnotes

1. in this context success means non-occurrance of errors during the initial lookup (need to know whether to update an already existing entry), or during resize. From the data collected, no errors occurred. ↩

2. insertdict may either insert a new entry into the hash table, or update the value of an existing one. ↩