test(node-suite): stop counting fixtures/ modules as tests

.github/workflows/test.yml

@@ -871,6 +871,7 @@ jobs:
         run: |
           python3 scripts/check_test_registration.py --self-test
           python3 scripts/check_test_registration.py
+          python3 scripts/node_suite_run.py --self-test
 
       # #7672: the GC test guards CLEAR ~20 process-global side tables from
       # whatever libtest thread constructs a guard, and nothing requires a

run_parity_tests.sh

@@ -1308,7 +1308,7 @@ case "$TEST_SUITE" in
         if [[ -d "$node_suite_search_root" ]]; then
             while IFS= read -r test_file; do
                 TEST_FILES+=("$test_file")
-            done < <(find "$node_suite_search_root" -type f -name '*.ts' | sort)
+            done < <(find "$node_suite_search_root" -type f -name '*.ts' -not -path '*/fixtures/*' | sort)
         fi
         ;;
     *)

scripts/check_test_registration.py

@@ -368,7 +368,8 @@ def _stem(rel: str) -> str:
 # absence from MECHANISMS is exactly the silence this gate exists to remove.
 GLOB_DRIVEN = (
     ("run_parity_tests.sh", "find test-files -maxdepth 1 -name '*.ts'; "
-     "find test-parity/node-suite -name '*.ts'"),
+     "find test-parity/node-suite -name '*.ts' -not -path '*/fixtures/*' "
+     "(scripts/node_suite_run.py applies the same rule; its --self-test pins it)"),
     ("scripts/gc_root_dominance_corpus.sh", "PATTERNS globs over test-files/; "
      "a pattern that matches nothing is already loud, and MIN_COMPILED floors "
      "the corpus size"),

scripts/node_suite_run.py

@@ -1,7 +1,8 @@
 #!/usr/bin/env python3
 """Differential runner for the print-and-diff node-suite (test-parity/node-suite).
 
-For every `test-parity/node-suite/<module>/**/*.ts`, run `node <t>` and
+For every `test-parity/node-suite/<module>/**/*.ts` outside a `fixtures/`
+directory, run `node <t>` and
 `perry <t> -o out && out`, then compare stdout (trailing whitespace ignored).
 Prints a per-module pass/total table plus an overall figure.
 
@@ -20,7 +21,11 @@
    run STRICTLY SEQUENTIALLY so their numbers are trustworthy; everything else
    stays parallel.
 
+A `fixtures/` directory holds modules that tests import, never tests: its `.ts`
+files print nothing, so counting them would only add trivial passes.
+
 Usage: node_suite_run.py <perry-bin> <repo-root> [comma-separated-modules]
+       node_suite_run.py --self-test
 """
 import os, re, subprocess, sys, tempfile
 from concurrent.futures import ThreadPoolExecutor
@@ -71,6 +76,43 @@ def normalize(text: str) -> str:
     return "\n".join(out)
 
 
+FIXTURE_DIR = "fixtures"
+
+
+def discover(base, mods=None):
+    """(module, path) for every test entry under `base`, skipping `fixtures/`."""
+    found = []
+    for mod in (mods or sorted(os.listdir(base))):
+        md = os.path.join(base, mod)
+        if not os.path.isdir(md):
+            continue
+        for dp, dirs, files in os.walk(md):
+            dirs[:] = [d for d in dirs if d != FIXTURE_DIR]
+            for f in files:
+                if f.endswith(".ts") and not f.endswith(".d.ts"):
+                    found.append((mod, os.path.join(dp, f)))
+    return found
+
+
+def self_test():
+    with tempfile.TemporaryDirectory() as td:
+        for rel in ("m/a.ts", "m/sub/b.ts", "m/sub/fixtures/dep.ts",
+                    "m/fixtures/nested/dep.ts", "m/types.d.ts", "n/c.ts"):
+            path = os.path.join(td, rel)
+            os.makedirs(os.path.dirname(path), exist_ok=True)
+            open(path, "w").close()
+        got = sorted(os.path.relpath(p, td) for _, p in discover(td))
+        want = ["m/a.ts", "m/sub/b.ts", "n/c.ts"]
+        if got != want:
+            print(f"FAIL: discovered {got}, want {want}")
+            return 1
+    print("OK: node-suite discovery skips fixtures/ and .d.ts")
+    return 0
+
+
+if len(sys.argv) > 1 and sys.argv[1] == "--self-test":
+    sys.exit(self_test())
+
 PERRY = sys.argv[1]
 ROOT = sys.argv[2]
 MODS = sys.argv[3].split(",") if len(sys.argv) > 3 and sys.argv[3] else None
@@ -87,16 +129,7 @@ def normalize(text: str) -> str:
     "trace_events",
 }
 
-tests = []
-base = os.path.join(ROOT, "test-parity", "node-suite")
-for mod in (MODS or sorted(os.listdir(base))):
-    md = os.path.join(base, mod)
-    if not os.path.isdir(md):
-        continue
-    for dp, _, files in os.walk(md):
-        for f in files:
-            if f.endswith(".ts") and not f.endswith(".d.ts"):
-                tests.append((mod, os.path.join(dp, f)))
+tests = discover(os.path.join(ROOT, "test-parity", "node-suite"), MODS)
 
 
 def run_one(args):

test-parity/node-suite/module/README.md

@@ -21,10 +21,11 @@ to 69**. The 69 entries are grouped by directory as follows:
 | `require`    |    13 |    15 | overloads, resolution, cache, cycles, metadata, packages      |
 | `source-map` |     7 |     7 | payloads, getters, mappings, origins, validation              |
 
-`loader/fixtures/**/*.ts` contains ten pre-existing executable fixture modules.
-The generic runner intentionally counts every `.ts` file recursively, so they
-remain represented in the total and baseline rather than being silently
-subtracted in this document.
+`loader/fixtures/**/*.ts` contains ten fixture modules that the loader tests
+import. The runners skip every `fixtures/` directory, so these modules are not
+counted as tests: the runner-visible lane is **59** entries (69 minus those
+ten). The historical figures below were measured when the fixtures were still
+counted.
 
 ## Primary evidence
 

test-parity/node_suite_baseline.json

@@ -2,7 +2,7 @@
   "_schema": {
     "description": "Floor baseline for scripts/node_suite_regression_check.py. Each module's run must produce pass >= floor.pass and total >= floor.total; dropping below either is a regression (exit 1). Improvements are always accepted and reported as ratchet candidates. Captured in the node-26 environment with scripts/node_suite_run.py (pre-warm + fast/slow lanes).",
     "oracle": "node v26.5.0 (pinned by .node-version)",
-    "note": "Full pass sets the floor for deterministic modules. Timing/racy modules (net, stream, diagnostics_channel, fs-promises) carry a small margin below observed pass so ordinary flake does not false-alarm; the guard still catches real regressions, which are large (e.g. dns 6->0, http 19->9). node_suite_run.normalize() scrubs environment-variant tokens (console.time hrtime durations, stack-trace frame lines) symmetrically before the stdout compare, so console uses a full floor (119) for its deterministic content. Three clean focused Perry runs against Node 26.5.0 set the expanded HTTP matched floor at 22; #5891 adds one current unmatched regression and raises the total floor to 53. The expanded net floor is stable at 16/47. The granular http2 matched floor is 32; #5891 adds one current unmatched regression and raises its total floor to 60. The https suite is stable at 6/47 with no margin. The deterministic readline suite is stable at 20/95 with no margin. A clean node-26 run at 2810/2863 (98.1%) refreshed the floors, then an independent Node 26.5.0 run set the deterministic child_process floor at 43/53."
+    "note": "Full pass sets the floor for deterministic modules. Timing/racy modules (net, stream, diagnostics_channel, fs-promises) carry a small margin below observed pass so ordinary flake does not false-alarm; the guard still catches real regressions, which are large (e.g. dns 6->0, http 19->9). node_suite_run.normalize() scrubs environment-variant tokens (console.time hrtime durations, stack-trace frame lines) symmetrically before the stdout compare, so console uses a full floor (119) for its deterministic content. Three clean focused Perry runs against Node 26.5.0 set the expanded HTTP matched floor at 22; #5891 adds one current unmatched regression and raises the total floor to 53. The expanded net floor is stable at 16/47. The granular http2 matched floor is 32; #5891 adds one current unmatched regression and raises its total floor to 60. The https suite is stable at 6/47 with no margin. The deterministic readline suite is stable at 20/95 with no margin. A clean node-26 run at 2810/2863 (98.1%) refreshed the floors, then an independent Node 26.5.0 run set the deterministic child_process floor at 43/53. Files under a fixtures/ directory are imported modules, not tests, and are no longer discovered; module's total floor drops from 69 to 59 for the ten loader/fixtures modules (its pass floor of 29 stays below the 57/59 measured without them)."
   },
   "modules": {
     "assert": {
@@ -117,7 +117,7 @@
     },
     "module": {
       "pass": 29,
-      "total": 69
+      "total": 59
     },
     "net": {
       "pass": 16,