Check child nodes before the layout style in Widget.is_container

evidence/container-height-membership-20261007/README.md

@@ -0,0 +1,38 @@
+# Native container membership during height measurement
+
+Native `is_container` is true for either committed child membership or an
+explicit layout. Child membership now answers first, avoiding RenderStyles
+layout resolution on every populated step of a relative-height traversal.
+Empty native child lists still consult the current layout declaration.
+
+This changes only the original Widget property. Subclass property overrides,
+relative percentage/fraction admission, hidden children, auto-size propagation,
+layout policies, measurement generations and reparenting remain unchanged.
+NodeList owns real child membership; RenderStyles owns the independently
+changing layout declaration. Neither answer is stored or inferred from bounds.
+
+The recursive relative-height query remains necessary: even independent
+Stream content measurements can have auto boxes stretched by relative-height
+descendants. Removing that query or treating content independence as box
+independence would change geometry. This patch removes a predicate cost; it
+does not claim to remove descendant traversal or solve burst frame latency.
+
+source.json records existing Package AST declarations and consumers across
+249 native and 288 Toad modules, with zero parse omissions. Runtime replacement
+of the entire is_container property continues to dispatch normally. Private
+replacement of native NodeList/RenderStyles with side-effectful nonconforming
+objects is not a native measurement contract.
+
+64 affected measurement/arrangement/structural-generation checks passed in
+4.33s. The existing loaded source Toad App completed with 568 widgets/10 tabs,
+equal relative-height answers for every widget, unchanged geometry and empty
+stderr. It compares old/current short-circuit order on the same acquired tree.
+Five alternating unprofiled trials of 20 query sweeps had medians 49.73ms and
+47.92ms; trial variation overlaps, so this is not a reliable latency improvement.
+The cProfile files retain instrumentation output; summed recursive cumulative
+values are not an aggregate wall-time result and are not used as speed evidence.
+
+No original session, submitted input, provider, package or public App is used.
+The full recursive descendant query remains the outstanding cost; this patch
+does not justify bypassing it, storing another readiness/dependency answer or
+claiming the original 2.869s writer gap has been solved.

evidence/container-height-membership-20261007/app.stdout.log

@@ -0,0 +1,40 @@
+{
+  "scope": "source Toad App recursive height queries, not burst/frame latency",
+  "widgets": 568,
+  "tabs": 10,
+  "sweeps_per_trial": 20,
+  "seconds": {
+    "original": [
+      0.05167253699619323,
+      0.04497357609216124,
+      0.043833934934809804,
+      0.0542515879496932,
+      0.04973278904799372
+    ],
+    "membership_first": [
+      0.0633440789533779,
+      0.04001189791597426,
+      0.04056658991612494,
+      0.048753642942756414,
+      0.04791633493732661
+    ]
+  },
+  "median_seconds": {
+    "original": 0.04973278904799372,
+    "membership_first": 0.04791633493732661
+  },
+  "profiles": {
+    "original": {
+      "is_container_seconds": 0.012723577,
+      "relative_height_seconds": 0.15487751200000002
+    },
+    "membership_first": {
+      "is_container_seconds": 0.013725005,
+      "relative_height_seconds": 0.006692936
+    }
+  },
+  "all_relative_answers_equal": true,
+  "geometry_unchanged": true,
+  "inputs": 0,
+  "providers": 0
+}

evidence/container-height-membership-20261007/controls.log

@@ -0,0 +1,2 @@
+................................................................         [100%]
+64 passed in 4.33s