[trungdong]

Three speed changes to the model from the 3.3.0 performance spec (spec 4, section 3.2), with every output, error and warning unchanged except one case spec 4 rules on (S15). Against main, unified is 22.7% faster, construct 12.5%, prov_to_graph 11.9%, deserialize[json] 7.2%, deserialize[jsonld] 6.9% and deserialize[xml] 5.5%, and every other benchmark case is within −2.7% and +0.6%.

• Record construction. _value_key tests Identifier first, since identifiers are the commonest attribute value and the type subclasses none of the others. add_attributes scans a call's attributes for prov:collection only when a single-valued attribute repeats, at most once per call, where it used to scan every call up front. A resolved qualified name is returned without the datetime and None checks it can never fail. The repeated-value comparison moves into a small helper, _check_repeated_value.
• unified(). The merge map is keyed by each record's id(), so finding a record's merge no longer hashes and compares the whole record. Every copy of an equal record still merges, including two equal copies sharing an identifier with a third record.
• add_namespace. An equal registered namespace is found by one lookup under its prefix instead of a scan of every namespace, so declaring many namespaces takes linear time (8,000 namespaces: 1.75 s on main, 0.01 s here). For an equal namespace it now returns the registered namespace, as its docstring says, rather than the argument; the two are equal (S16).

S15: on main, add_attributes compared every attribute name with prov:collection before storing anything, so a name object whose __eq__ raises made the call raise up front. It now raises only when a single-valued attribute repeats. Names from the readers and the API are strings or qualified names, so nothing they accept changes.

Why: these are the model-side changes the performance study measured as behaviour-free, after the writer-side changes of #698.

How it was checked: the output-identity script reads every fixture under src/prov/tests and writes the benchmark document, the example documents and every fixture document in each format. It then reads each one back and compares unified(), flattened(), document equality and prov_to_dot against main under PYTHONHASHSEED 1 and 2, and finds no difference. A probe also confirmed that a document repeating one formal attribute thousands of times reads as fast as on main (8,000 repeated prov:activity elements in PROV-XML: 0.024 s on both). New cases pin the prov:collection rule with the collection after the repeat and with a one-shot iterator, a single scan per call, two equal copies plus a third record merging into one, and add_namespace returning the registered namespace for an equal one without registering anything. Spec 4 records S15, S16 and the wording this PR needed. No HISTORY entry, since output is unchanged and add_namespace now returns what its docstring promised.

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

🟢 Metrics 5 complexity

Metric	Results
Complexity	5

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}

[coveralls]

coverage: 98.381% (-0.02%) from 98.401% — feat/3.3.0-model-speed into main