[bfops/wasm-test]: diff-explain.txt

Author: bfops

diff-explain.txt

@@ -0,0 +1,48 @@
+ Mechanical only:
+
+  - sdks/rust/tests/test.rs:1 is mostly runner/config wiring so the same declared tests can execute either the native client or the wasm client.
+  - The bulk of the async churn in sdks/rust/tests/test-client/src/main.rs:41 is helper-level: shared dispatch, shared connect*, shared wait_for_all, wasm/native shims.
+  - sdks/rust/tests/connect_disconnect_client/src/main.rs:1 was similarly moved onto the shared async harness path, but its assertions did not change.
+
+  The test-body changes that are actually substantive are these:
+
+  - sdks/rust/tests/test-client/src/main.rs:1045 and sdks/rust/tests/test-client/src/main.rs:1347
+    Reason: Timestamp::now() is stubbed/unimplemented on wasm32-unknown-unknown.
+    Why the change is acceptable: those tests are not trying to validate wall-clock acquisition; they are validating timestamp serialization/round-trip through reducers/subscriptions. A fixed timestamp preserves that assertion and
+    removes a wasm-only runtime failure.
+  - sdks/rust/tests/test-client/src/main.rs:2179
+    Reason 1: it now subscribes only to one_u_8 and one_u_16 instead of subscribe_all.
+    Reason 2: it now marks the subscription as “ready” only after the relevant callbacks are installed.
+    Why the change is needed: this test only asserts on those two tables, so subscribe_all was unnecessary setup cost. More importantly, the old ordering could report “subscription applied” before the handlers under test were
+    actually registered, which is a real race on wasm.
+
+  If you are looking at my current local changes on top of that branch, there are a few additional meaningful test-body fixes:
+
+  - sdks/rust/tests/test-client/src/main.rs:2729
+    Reason: it registered TestCounter expectations but never waited for them.
+    Why the change is needed: the old version could fall off the end early on native and became an outright hang on wasm. Adding the wait does not broaden the test; it makes the existing assertion actually execute.
+  - sdks/rust/tests/test-client/src/main.rs:2821
+    Reason: it used a blocking Barrier inside callback-driven code.
+    Why the change is needed: that deadlocks wasm, because the callbacks and the wait share the same JS event loop. The atomic gate preserves the intended semantics: insert exactly once, only after all three subscriptions are
+    ready.
+  - sdks/rust/tests/test-client/src/main.rs:2919
+    Reason: it used a second TestCounter as a cross-client blocking barrier.
+    Why the change is needed: same wasm deadlock issue. The test’s actual assertion is per-client filtering after that client’s own subscription applies; it does not require both clients to reach a global barrier first.
+  - sdks/rust/tests/test-client/src/main.rs:2977
+    Reason: same pattern as test_parameterized_subscription.
+    Why the change is needed: the barrier was not part of the behavior under test, only a synchronization convenience that is invalid on wasm. Removing it keeps the test focused on RLS visibility rather than cross-client
+    simultaneity.
+
+  So the short version is:
+
+  - Most of the scary-looking diff is harness plumbing, not changed assertions.
+  - The real test-content changes were either:
+      - replacing native-only time acquisition with a deterministic value, or
+      - removing synchronization patterns that were accidentally relying on native’s threaded runtime and were invalid on wasm.
+
+  If you want, I can turn this into a reviewer-oriented checklist grouped as:
+
+  1. pure harness changes,
+  2. assertion-preserving test fixes,
+  3. true behavioral test changes.
+