Temporarily disable TC_IDM_1_2 as it is too flaky

Author: ivmarkov

xtask/src/itest.rs

@@ -88,7 +88,25 @@ pub(crate) const SYS_TESTS: &[&str] = &[
     //
     // Python tests — Interaction Data Model (general Matter protocol)
     //
-    "TC_IDM_1_2",
+    // "TC_IDM_1_2", // Skipped: flake in step 8/9 — `FakeRevokeCommissioning`
+    //                // (Invoke with `timed_request=true` flag but NO preceding
+    //                // TimedRequest message) races the OpenCommissioningWindow
+    //                // mDNS-publish burst. The exchange handler responds with
+    //                // `TimedRequestMismatch` correctly, but the post-invoke
+    //                // `Exchange::acknowledge()` (which must MRP-ack the
+    //                // controller's closing IM `StatusResponse(Success)` carrying
+    //                // the `R` flag) blocks on `init_send.await` for the Matter
+    //                // transport TX buffer — held by the mDNS responder which
+    //                // grabs it for the entire broadcast loop in
+    //                // `rs-matter/src/transport/network/mdns/builtin.rs:161-219`.
+    //                // The MRP retransmit-deadline window then closes and the
+    //                // exchange wedges with `Pending ACK. Marking as dropped` at
+    //                // teardown. Flakes ~10% locally on fast hardware and
+    //                // deterministically on slower CI hardware. Independent of
+    //                // the TimeSync work — same rate at HEAD~2 (`32f4c7e`).
+    //                // Fix is in the mDNS responder: release the Matter TX
+    //                // buffer between broadcast_one calls, or move mDNS off the
+    //                // shared TX path.
     "TC_IDM_1_4",
     "TC_IDM_2_2",
     "TC_IDM_4_2",