Module function_coupling 

function-coupling analysis.

For a single target file, identifies pairs of functions that co-change (appear together in the same revision’s hunk-attributed change set) and ranks them by Fisher exact significance. Only pairs with co_changes ≥ 2 are emitted.

Algorithm: for each revision that touched the target file, the set of HEAD-alive functions whose line spans overlapped any hunk is computed via the same hunk-overlap logic used by function-xray. Co-change counts are then accumulated over all revisions. For each pair (a, b) the Fisher 2×2 contingency table is:

          b touched   b not touched
a touched     co          a_only
a not         b_only      neither

where n = total distinct revisions touching the file (from the hunks table, regardless of which functions they touched) and neither = n − co − a_only − b_only.

Rename limitation: hunk attribution uses WHERE h.path = ? (current HEAD-relative path). Pre-rename history is not attributed — see function_xray module doc.

Output: sorted by p_value ASC (None first — degenerate marginal implies p → 0, i.e. perfectly coupled) then confidence DESC for determinism. confidence = co_changes / min(a_changes, b_changes).

Research basis: Adams et al., ICSM 2006 “The Co-Change Rule”; Fisher significance adapts the coupling analysis in Tornhill, “Your Code as a Crime Scene” (2015) to function granularity.

Structs

FunctionCouplingRow

One row per function pair with co_changes ≥ 2, sorted by p_value ASC.

Functions

run_function_coupling

Run the function-coupling analysis.