ponder-lab / Hybridize-Functions-Refactoring / #2180

79%

push

github

Emit per-parameter input signatures as `input_signatures.csv` (#679)

* Emit per-parameter input signatures as input_signatures.csv

Add input_signatures.csv, one row per function parameter and source, so downstream
analysis of the input signatures is a group-by rather than a parse of the joined
input_signature string in functions.csv. Each non-self parameter of the inferred
signature (source inferred) and of the developer-supplied signature (source supplied)
gets a row with its raw dtype and shape; a blocked parameter gets a source-absent row
carrying its AbsenceReason. InputSignature gains a parameterSpecs() accessor exposing
the per-parameter dtype and shape, sharing the shape rendering with toTensorSpecList.

Closes #665.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FuhM8SsTRB6BraLt3Ph7ZC

* Extract the CSV column names and source values into constants

Name the printer-block column strings, so the columns shared across emitters stay
single-sourced: "param index" and "absence reason" between blocked_parameters.csv and
input_signatures.csv, and refactoring/severity/code/message between the failed-precondition
and status CSVs. Also names the input_signatures.csv source values. Addresses review
feedback on #665.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FuhM8SsTRB6BraLt3Ph7ZC

* Revert "Extract the CSV column names and source values into constants"

This reverts commit a29dab01d. The evaluator's
convention is plain string literals for CSV column names (see
buildFunctionAttributeColumnNames), so the constants were out of step; restoring the
literals.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FuhM8SsTRB6BraLt3Ph7ZC

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

10 of 10 new or added lines in 1 file covered. (100.0%)

1627 of 2087 relevant lines covered (77.96%)

0.78 hits per line