Merge branch 'worktree-iso-phase-4' into macros

sx/sx/boundary.sx

@@ -59,3 +59,8 @@
   :params ()
   :returns "dict"
   :service "sx")
+
+(define-page-helper "run-spec-tests"
+  :params ()
+  :returns "dict"
+  :service "sx")

sx/sx/testing.sx

@@ -1,6 +1,6 @@
 ;; Testing spec page — SX tests SX.
 
-(defcomp ~spec-testing-content (&key spec-source)
+(defcomp ~spec-testing-content (&key spec-source server-results)
   (~doc-page :title "Testing"
     (div :class "space-y-8"
 
@@ -22,15 +22,27 @@
           (code :class "text-violet-700 text-sm" "if")
           " works. No code generation, no intermediate files — the evaluator runs the spec."))
 
-      ;; Live test runner
+      ;; Server-side results (ran when this page was rendered)
       (div :class "space-y-3"
-        (h2 :class "text-2xl font-semibold text-stone-800" "Run in browser")
+        (h2 :class "text-2xl font-semibold text-stone-800" "Server: Python evaluator")
+        (p :class "text-stone-600"
+          "The Python evaluator ran "
+          (code :class "text-violet-700 text-sm" "test.sx")
+          " when this page loaded — "
+          (strong (str (get server-results "passed") "/" (get server-results "total") " passed"))
+          " in " (str (get server-results "elapsed-ms")) "ms.")
+        (pre :class "text-sm font-mono bg-stone-900 text-green-400 rounded-lg p-4 overflow-x-auto max-h-96 overflow-y-auto"
+          (get server-results "output")))
+
+      ;; Client-side test runner
+      (div :class "space-y-3"
+        (h2 :class "text-2xl font-semibold text-stone-800" "Browser: JavaScript evaluator")
         (p :class "text-stone-600"
           "This page loaded "
           (code :class "text-violet-700 text-sm" "sx-browser.js")
           " to render itself. The same evaluator can run "
           (code :class "text-violet-700 text-sm" "test.sx")
-          " right here — SX testing SX, in your browser:")
+          " right here:")
         (div :class "flex items-center gap-4"
           (button :id "test-btn"
             :class "px-4 py-2 rounded-md bg-violet-600 text-white font-medium text-sm hover:bg-violet-700 cursor-pointer"

sx/sxc/pages/docs.sx

@@ -343,7 +343,8 @@
               :source (read-spec-file (get item "filename"))))
         extension-spec-items))
     "testing" (~spec-testing-content
-      :spec-source (read-spec-file "test.sx"))
+      :spec-source (read-spec-file "test.sx")
+      :server-results (run-spec-tests))
     :else (let ((spec (find-spec slug)))
       (if spec
         (~spec-detail-content

sx/sxc/pages/helpers.py

@@ -24,6 +24,7 @@ def _register_sx_helpers() -> None:
         "bundle-analyzer-data": _bundle_analyzer_data,
         "routing-analyzer-data": _routing_analyzer_data,
         "data-test-data": _data_test_data,
+        "run-spec-tests": _run_spec_tests,
     })
 
 
@@ -491,6 +492,76 @@ def _event_detail_data(slug: str) -> dict:
     }
 
 
+def _run_spec_tests() -> dict:
+    """Run test.sx against the Python SX evaluator and return results."""
+    import os
+    import time
+    from shared.sx.parser import parse_all
+    from shared.sx.evaluator import _eval, _trampoline
+
+    ref_dir = os.path.join(os.path.dirname(__file__), "..", "..", "shared", "sx", "ref")
+    if not os.path.isdir(ref_dir):
+        ref_dir = "/app/shared/sx/ref"
+    test_path = os.path.join(ref_dir, "test.sx")
+    with open(test_path, encoding="utf-8") as f:
+        src = f.read()
+
+    suite_stack: list[str] = []
+    passed = 0
+    failed = 0
+    test_num = 0
+    lines: list[str] = []
+
+    def try_call(thunk):
+        try:
+            _trampoline(_eval([thunk], {}))
+            return {"ok": True}
+        except Exception as e:
+            return {"ok": False, "error": str(e)}
+
+    def report_pass(name):
+        nonlocal passed, test_num
+        test_num += 1
+        passed += 1
+        lines.append("ok " + str(test_num) + " - " + " > ".join(suite_stack + [name]))
+
+    def report_fail(name, error):
+        nonlocal failed, test_num
+        test_num += 1
+        failed += 1
+        full = " > ".join(suite_stack + [name])
+        lines.append("not ok " + str(test_num) + " - " + full)
+        lines.append("  # " + str(error))
+
+    def push_suite(name):
+        suite_stack.append(name)
+
+    def pop_suite():
+        suite_stack.pop()
+
+    env = {
+        "try-call": try_call,
+        "report-pass": report_pass,
+        "report-fail": report_fail,
+        "push-suite": push_suite,
+        "pop-suite": pop_suite,
+    }
+
+    t0 = time.monotonic()
+    exprs = parse_all(src)
+    for expr in exprs:
+        _trampoline(_eval(expr, env))
+    elapsed = round((time.monotonic() - t0) * 1000)
+
+    return {
+        "passed": passed,
+        "failed": failed,
+        "total": passed + failed,
+        "elapsed-ms": elapsed,
+        "output": "\n".join(lines),
+    }
+
+
 def _data_test_data() -> dict:
     """Return test data for the client-side data rendering test page.