Summary

Optimize dict.fromkeys(d, val) when d is a combined, compact dict by cloning the source PyDictKeysObject in one memcpy instead of inserting each key individually.

Fixes gh-153229

Motivation

Previously dict_dict_fromkeys() always called insertdict() for every key in the source, performing a full hash probe on the (empty) target dict N times. This is wasteful: the target is always empty, so every probe is a guaranteed miss.

dict.copy() and {}.update(other) already use clone_combined_dict_keys() to fast-clone the PyDictKeysObject in one shot. dict.fromkeys() lacked an equivalent path.

Changes

Objects/dictobject.c

• Added clone_combined_dict_keys_with_value(orig, fill_value): a variant of clone_combined_dict_keys() that replaces every active entry's value with a new reference to fill_value instead of copying from the source.

• Added fast path in dict_dict_fromkeys(): when the source dict satisfies:

  • ma_values == NULL (combined, not split)
  • ma_used == dk_nentries (compact, no deleted entries)
  • Is a plain dict (not a subclass with custom tp_iter)
  • Source is non-empty

  …call clone_combined_dict_keys_with_value() and install the resulting keys object directly, bypassing the insertdict loop entirely.

Benchmark

1000-key string dict, 10,000 iterations of dict.fromkeys(d, None), Python 3.16.0a0, GCC 15.2.0:

Tests

./python -m test test_dict test_dictviews
== Tests result: SUCCESS ==   (157 tests passed)