fix: resolve string aggfun to callable in table_pivot step

[apoorvdarshan]

• fixes Bug: Pipeline.from_descriptor does not work for all steps #1764

---

Summary

A table-pivot step defined via a JSON/YAML descriptor with "aggfun": "sum" fails, while the equivalent Python API steps.table_pivot(..., aggfun=sum) works, even though the docs (and the step's own docstring) say descriptors mirror the Python API and that aggfun "can be sum, max, min, len etc.".

Reproducer from the issue:

json_descriptor = {
    "steps": [
        {"type": "table-normalize"},
        {"type": "table-pivot", "f1": "region", "f2": "gender", "f3": "units", "aggfun": "sum"},
    ]
}
Resource(data=data).transform(Pipeline.from_descriptor(json_descriptor))

fails with:

FrictionlessException: [step-error] Step is not valid: "table_pivot" raises
"[source-error] ... 'str' object is not callable"

Root cause: aggfun is typed Any with an empty metadata profile, so a JSON string like "sum" passes validation and is handed straight to petl's table.pivot(f1, f2, f3, aggfun), which requires a callable, not a string.

Fix

In frictionless/steps/table/table_pivot.py, transform_resource now resolves a string aggfun to the corresponding callable before calling petl.pivot, via a small map of the standard petl aggregation names:

sum, max, min, len, count (alias for len), mean (statistics.mean), first, last.

An actual callable aggfun is passed through unchanged (the existing Python-API behavior). An unrecognized name raises a clear StepError (aggregation function "<name>" is not supported) instead of the opaque petl TypeError. The docstring was updated to list the supported names.

Tests

Added test_step_table_pivot_string_aggfun_issue_1764 in frictionless/steps/table/__spec__/test_table_pivot.py, mirroring the existing test_step_table_pivot_issue_1220 but building the pipeline from a descriptor with "aggfun": "sum". It fails on main ('str' object is not callable) and passes with this change. Full frictionless/steps, frictionless/pipeline, and frictionless/transformer suites pass; ruff check, ruff format --check, and pyright are clean on the changed files.

Note on scope

This fixes the descriptor -> callable (input) direction, which is the reported failure and matches the documented usage. The issue also mentions that pipeline.to_json() fails when a Python callable was passed directly (e.g. the builtin sum), because a function is not JSON serializable. Reliably serializing an arbitrary callable back to a descriptor name is a separate design decision (lambdas / user functions have no canonical name), so it is intentionally left out here; using a string aggfun in the descriptor now round-trips cleanly through to_json/from_descriptor.

Disclosure: prepared with AI assistance; reviewed and verified locally.