mirror of
https://github.com/abhigyanpatwari/GitNexus.git
synced 2026-09-24 00:51:53 +00:00
d9d6318b64
* feat(group): add Kotlin Spring HTTP consumer extraction Follow-up to #1849 (Kotlin providers). Extends `http-patterns/kotlin.ts` with three call-site patterns common in Kotlin Spring projects: - RestTemplate: `restTemplate.getForObject("/x", ...)` and the full verb family (getForObject/getForEntity → GET, postForObject/postForEntity → POST, put → PUT, delete → DELETE, patchForObject → PATCH). Mirrors the Java plugin's `REST_TEMPLATE_TO_HTTP` map so polyglot repos coalesce on a single contract id. - WebClient short form: `webClient.get().uri("/x")` and the `.post()` / `.put()` / `.delete()` / `.patch()` siblings. The chain parses as two nested `call_expression` nodes; the query anchors on the outer `.uri(...)` and walks one level inward to constrain the verb. - OkHttp: `Request.Builder().url("/x")`. Kotlin parses `Request.Builder()` as a `call_expression` whose callee is a `navigation_expression` (not Java's `object_creation_expression`), so the query shape differs from `java.ts` but the receiver/method constraints (`Request` / `Builder` / `url`) and emitted contract format match. Out of scope: `webClient.method(HttpMethod.X).uri("/y")` long form. The verb sits on a sibling `call_expression` two hops away, so it needs a walk-up helper rather than a flat tree-sitter query. A dedicated anti-overreach test pins the current behavior so a future short-form change can't accidentally start matching the long form. Receiver name constraints (`#eq? @obj "restTemplate"`, `#eq? @cls "Request"`) match the Java plugin's heuristic — a project that aliases the receiver under a different name won't be picked up. This trade-off keeps false-positive rates low and is documented in the file header. Tests: 5 new cases under `consumer extraction — fetch patterns`, gated by tree-sitter-kotlin grammar availability. positive (3) - RestTemplate verbs (5 calls × 5 verbs) - WebClient short-form verbs (5 calls × 5 verbs) - OkHttp Request.Builder().url("/x") anti-regression (2) - WebClient long form `.method(HttpMethod.X)` produces no consumer (deferred-feature pin) - non-restTemplate receiver does not match (receiver-name pin) Reverse-validated: removing the `(#eq? @obj "restTemplate")` constraint causes the receiver-name anti-regression test to fail. Local validation: - test/unit/group/http-route-extractor.test.ts: 59/59 ✅ - test/unit/group: 539/539 ✅ - npm run format:check: clean ✅ * test(group): pin Kotlin OkHttp POST-chain heuristic-default GET behavior Address Claude review on PR #1855 (Finding 1). The OkHttp query in `kotlin.ts:OK_HTTP_PATTERNS` matches the `.url("/x")` sub-expression of a builder chain, but the verb is encoded on a separate sibling call (`.post(body)` / `.delete()` / ...). The query intentionally does not walk the chain to recover the verb — it emits `method: 'GET'` for every match, mirroring the Java plugin's `OK_HTTP_PATTERNS` (java.ts). Concretely: `Request.Builder().url("/x").post(body).build()` becomes `http::GET::/x`, not `http::POST::/x`. This is an already-accepted Java parity heuristic, but it was untested on the Kotlin side. This commit: - Adds an anti-overreach test pinning the current behavior: * exactly one consumer is emitted with method=GET * no second http::POST::/x consumer appears - Documents the limitation in kotlin.ts as a "Known limitation" block tied to the test, so a future verb-walk implementation has to update the comment in lockstep with the assertion. Rationale for not implementing verb-walk in this PR: - Verb-walk requires walking sibling call_expression nodes (the `.post(body)` chain), which is the same shape as the deferred WebClient long-form work - Java has the same limitation in production today; fixing only Kotlin would create polyglot drift - A coordinated future PR can add verb-walk to both plugins at once and update both comments + the pin tests together Finding 2 (silent test-skip when tree-sitter-kotlin grammar is unavailable) is intentionally NOT addressed here — same gating pattern was accepted in #1849 for Provider tests, and a coordinated follow-up should add a CI sentinel covering both Provider and Consumer suites in one place. Local validation: - test/unit/group/http-route-extractor.test.ts: 60/60 ✅ - test/unit/group: 540/540 ✅ - npm run format:check: clean ✅ --------- Co-authored-by: henry <zhangwei2017@unipus.cn> |
||
|---|---|---|
| .. | ||
| fixtures | ||
| helpers | ||
| integration | ||
| unit | ||
| utils | ||