fix(parsers/go): classify entry points by signature, not return type or bare next(

[gadievron]

fix(parsers/go): classify entry points by signature, not return type or bare next(

Base: master · Type: bug fix · Findings: F10 + F7 (MED)

What

parsers/go/go_parser/extractor.go:

• F10 — isHTTPHandler no longer matches httpHandlerPatterns against the
  return type (dropped || pattern.MatchString(returnsStr)); it matches
  parameters only.
• F7 — isMiddleware no longer treats a bare next( substring as middleware;
  the next( heuristic is now gated on the function having an HTTP request
  signature (hasHTTPSignature && strings.Contains(code, "next(")). The
  http.Handler/http.HandlerFunc return and next.ServeHTTP signals are kept.

Why

Both heuristics over-matched, seeding false remote-web entry points:

• A factory func Logger() gin.HandlerFunc has no request param but matched the
  bare HandlerFunc pattern against its return type → tagged http_handler. A
  handler is defined by what it receives, not what it returns.
• Any function whose body contains next( — Go 1.23 range-over-func iterators
  (iter.Pull → next()), lexers, cursors — was tagged middleware.

Tests

parsers/go/go_parser/extractor_classify_test.go (new): factory returning
http.HandlerFunc is not http_handler; iterator using next() is not
middleware; a genuine (w, r) handler and genuine next.ServeHTTP middleware
are still detected.

• RED (pre-fix): both negative tests failed.
• GREEN (post-fix): 4 passed; go test ./... passes.

Reachability impact (verified empirically, base vs patched binary)

Function-id sets identical (no extraction loss). Reclassified units sampled and
confirmed all fake:

• samber/lo: middleware 1 → 0 (CutPrefix, a pull-iterator).
• gin-gonic/gin: http_handler 108 → 89; all 19 removed are factories
  returning HandlerFunc (Logger, Recovery, BasicAuth, …) with no request
  param. All 89 retained handlers carry a genuine request/context param.

Zero genuine handlers or middleware lost.

Scope / siblings (deferred, documented)

Same defect class but out of scope here (tightening risks false negatives,
needs its own corpus): isHTTPHandler body heuristics w.Write( / Handler =
(:289-290) over-match any local w with .Write(; isCLIHandler os.Args
body match (:318). Tracked for a follow-up.

Upstream coordination

Only PR #126 also edits extractor.go (different functions: typeToString
generics + duplicateIDWarning). No hunk overlap with isHTTPHandler/
isMiddleware; rebase trivially if #126 lands first. Binary go_parser not
included — reviewers/CI rebuild from source.

Author notes

• Fix lines: isHTTPHandler param-only loop; isMiddleware hasHTTPSignature
  gate on next(.
• Input it now handles: middleware factories and next()-using iterators that
  previously became false entry points.
• Likely pushback: "could a real handler be return-only?" — net/http and
  gin/echo/fiber handlers always take the request/context as a param; a return-only
  match is a factory. Body heuristics still catch param-less real handlers.

[gadievron]

Findings F7 + F10 (MED). Coordinates with #126 (pr/parser-generic-anon-class-keys), which edits the same file extractor.go but different functions (typeToString/duplicateIDWarning) — no hunk overlap; rebase trivially if #126 lands first. Reachability-verified: removed seeds are all fake (gin HandlerFunc factories, lo CutPrefix iterator); zero genuine handlers/middleware lost.

[gadievron]

Rebased onto origin/master (post-release #133 / Wave 1). Conflicts resolved by UNION — both this PR's fix and the Wave-1 parser fixes are preserved; no Wave-1 change reverted (independently judge-verified). Force-pushed with --force-with-lease.

Clean rebase (master did not touch extractor.go); go test green.

[gadievron]

Closing as already resolved on current master. This fix's behavior shipped via the parser-fix-stack release (#133/#134); confirmed by independent reproduction — the original bug (reconstructed from its bug-pipeline investigation) no longer manifests on master, not merely by line-presence. Part of a full validated sweep of the open-PR corpus.