fix(build): prefer higher-info relation on same-pair edge collapse

[arimu1]

Summary

Fixes #2391.

build_from_json sorts edges by (source, target, relation) then inserts them into a simple nx.Graph / nx.DiGraph with last-write-wins add_edge. That made alphabetical order the survivor whenever two different relations landed on the same node pair — so references always overwrote calls ("calls" < "references"), silently deleting the higher-information call edge.

This adds a small _RELATION_PRIORITY table so the higher-information relation wins the collapse instead. Unlisted relations score 0 (first-wins), so obscure types do not pick up a new alphabetical bias. Equal priority keeps the first-seen edge, same spirit as the #1061 reverse-direction guard (left intact).

Out of scope (can follow up if wanted):

• Multigraph path / --multigraph wiring (related open PR [build] Add multigraph option to preserve parallel edges #709)
• Retaining discarded relations in an also_relations attribute

Test plan

• uv run pytest tests/test_build.py -q — 72 passed
• Regression: calls + references both input orders → survivor is calls with source_location=L271
• indirect_call beats contains
• Sole references still kept
• Same priority rule on directed=True
• Existing v0.8.22 regression of #760: fresh /graphify build round-trips correct AST output into reversed calls edges (build_from_json -> to_json undirected round-trip) #1061 bidirectional calls direction test still green

Notes

• Author: arimu1
• Platform: macOS (local pytest); CI should cover Linux matrix
• AI-assisted implementation (Grok 4.5 / Grok Build), human-reviewed

[graphify-labs]

Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).

Graphify reviewed this change.

Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).

---

Graphify review — findings

This PR introduces a relation-priority scheme in build_from_json to control which edge survives when a simple Graph/DiGraph collapses parallel edges on the same node pair. It adds a _RELATION_PRIORITY table and _relation_priority helper, and modifies the edge-adding logic so that when two edges share a pair but have different relations, the higher-priority relation is kept instead of relying on alphabetical last-write-wins (covering both undirected and directed cases). New tests in test_build.py exercise these collapse scenarios (e.g., calls vs references, indirect_call vs contains, sole-relation retention, and the directed variant).

Worth a look

• Undirected reverse-pair branch never applies relation priority — graphify/build.py:1075 · Escalate · medium
  • agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 804 functions depend on the 202 functions this change touches.

Health — this change adds coupling hotspots:

• worse: build_from_json() — 146 callers, 15 callees

Verification — 804 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

• verification_scope: 603 function(s) in the blast radius were not formally verified this run

· 1 more finding(s) on lines outside this diff (see the check run).