Benchmarks

How to run

python benchmarks/bench.py # print the comparison table
python benchmarks/bench.py --json benchmarks/results/results.json # write JSON

What is measured

• Construction: kwargs, mapping, nested (reference + C; plain dict where applicable).
• Key lookup / assignment / deletion (d[k], d[k]=v, del d[k]).
• Attribute lookup / assignment / deletion (d.name, d.name=v, del d.name).
• Iteration (list(d)).
• Shallow copy (d.copy).

Each operation is compared across plain dict, the pure-Python reference (attributedict._reference), and the C AttributeDict.

Methodology

• timeit with N=100_000 iterations, 5 repeats; report mean and median microseconds per operation.
• Warm cache: one untimed run precedes the timed repeats.
• Environment (Python version, platform, CPU) is captured in the JSON output for reproducibility.
• Import/module setup is never measured (imports happen before timing).

Results

Committed as data in benchmarks/results/results.json. Summarized in performance.md.

CI

Benchmarks are not run in full in CI; a smoke run validates the script executes. Full results are produced on demand by maintainers.