neo4j: relationship types are not namespaced, unlike every other analyzer's

[rahlk]

Plan

Found while checking whether JavaScript modules are distinguishable in the Neo4j projection. Planning context: roadmap.

Problem

Node labels are namespaced; relationship types are not. A shared database holding output from more than one analyzer mingles their edges.

What this analyzer emits (verified from a real --emit neo4j run on OWASP NodeGoat):

labels : Application:TSApplication Module:TSModule CallSite:TSCallSite
 Package:TSPackage Variable:TSVariable ← namespaced
rels : CALLS DECLARES DECLARES_VAR HAS_CALLSITE
 HAS_MODULE MEMBER_OF RESOLVES_TO ← NOT namespaced

What codeanalyzer-python emits for the same concepts (codeanalyzer/neo4j/schema.py):

PY_CALLS PY_DECLARES PY_DECLARES_VAR PY_HAS_CALLSITE PY_HAS_MODULE PY_RESOLVES_TO
PY_CDG PY_CFG_NEXT PY_DDG PY_DECORATED_BY PY_EXTENDS PY_HAS_ATTRIBUTE
PY_HAS_CFG_NODE PY_IMPORTS PY_PARAM_IN PY_PARAM_OUT PY_SUMMARY

Every one of this analyzer's seven relationship types has a PY_-prefixed counterpart on the Python side. Project a Python app and a TypeScript app into one database and CALLS is TypeScript-only while PY_CALLS is Python-only — a query written against one silently returns nothing for the other, and any query written against both has to know the asymmetry.

The Python analyzer's own changelog states the intent explicitly: edges are PY_-namespaced "like every other row family so a multi-language database never mingles analyzers' dependence edges."

Prior art — read before starting

PR #7 (patch/ts-prefix-neo4j-schema, "prefix graph labels and relationship types with TS") has been open against main since 2026-06-20 and reports mergeable=UNKNOWN. This issue is not a proposal to redo that work blind — decide first whether to revive, rebase, or supersede it. Filing this because the question is unresolved and undocumented, not because #7 is wrong.

Contract gate

This changes Neo4j output shape, which is a public contract with cross-analyzer parity obligations, so it enters at designing-cldk-changes, not as a direct fix. Two questions it has to settle, and neither is TypeScript-local:

1. Is the namespace prefix per language (TS_/JS_) or per analyzer family (TS_ covering both)? This is the same artifact-language-vs-family question the roadmap parks as candidate 2 — coining it here would decide it by accident.
2. Do the twin/generic labels survive? Today a node carries both :Module and :TSModule. If relationships gain a prefix, a bare cross-language query loses its only remaining language-agnostic handle.

Scope boundary

Settles relationship-type namespacing for the Neo4j projection. Does not decide the JSON analysis.json identity question (roadmap candidate 2), and does not add a language discriminator to nodes — a .js module and a .ts module are both :Module:TSModule today with no property distinguishing them, which is a separate gap.

Goals

• Decide revive / rebase / supersede for PR patch/ts-prefix-neo4j-schema: prefix graph labels and relationship types with TS #7, and record the decision
• Design session settles per-language vs per-family prefixing, in family-neutral terms so clang and jvm can inherit it
• Decide whether generic twin labels stay
• Whatever lands, codeanalyzer-python and codeanalyzer-typescript agree — one convention, not two

Caveats and known risks

• Breaking for anyone with an existing graph. Renaming relationship types invalidates every stored query and every projected database; it needs a migration path or a schema-version bump, not a silent rename.
• Deciding this coins vocabulary permanently. The parity clause makes a prefix scheme unrenameable once shipped — which is exactly why it belongs in a design session rather than a patch.
• Do not fix this on the 0.x line. 0.x is the supported line for consumers today but its Neo4j projection is v1; coining the scheme there and again on v2 is the double-coining failure the roadmap already flags.
• The v2 line's TS_PREFIX constant is dead (tracked separately) — do not treat its existence as evidence that prefixing is already half-done.

Definition of done

• A committed design decision naming the prefix scheme and its scope, in family-neutral terms.
• codeanalyzer-python and codeanalyzer-typescript emit relationship types under the same convention, demonstrated by projecting one app from each into a single database and running one query that returns rows from both.
• PR patch/ts-prefix-neo4j-schema: prefix graph labels and relationship types with TS #7 is merged, closed, or explicitly superseded — not left open a second time.

[rahlk]

Landed on release/0.x via #86, merge commit ea6eac3, targeting 0.6.0.

Closing manually: the PR carried Closes #88, but GitHub only honours closing keywords for PRs merged into the default branch, and this one merged into release/0.x.