# Changelog ## 0.1.3 — 2026-10-31 - Adapt the native library, examples and scripts to Moon/moonrun 0.1.20261028 (`e1e1f15`) and moonc 0.10.18+6c83a7f62. - Upgrade the official Protobuf runtime to 0.2.0 and regenerate the unchanged Pulsar protocol with protoc-gen-mbt 0.3.0. This removes the upstream reserved `Option` declaration that blocked 0.1.2 documentation builds with newer compilers. - Migrate integer/default-parameter syntax and deprecated JSON construction; make script imports explicit frontmatter. - Bound Linux test compilation to a 32 MiB subprocess stack, then run the test executables separately at the normal stack limit. This is a measured workaround for input-sensitive compilation, not a proven compiler version regression. - Align CONNECT/AUTH_RESPONSE identity, example dependencies and CI archive consumers with 0.1.3. No existing client or protocol declarations were removed or renamed. Integer aliases are rendered as `Int32`/`UInt32`, and the generator explicitly exports trait methods. Applications directly using upstream `google.protobuf.Option` must migrate to `Option_`; custom Protobuf users should use the matching generator/runtime. The selected SDK and dependency upgrade may require consumer changes, so this is not a zero-migration claim for every application. Other SDK combinations remain unverified. Check registry availability before installation; actual publication is recorded in the matching GitHub Release. The functional behavior and verification limits of 0.1.2 remain. The compatibility source passed native debug/release CI and isolated Pulsar 3.3.9/4.2.4 checks; final version-specific CI and registry documentation results are recorded with the 0.1.3 Release. ## 0.1.2 — 2026-10-29 - Close the TCP transport owned by an ordinary TLS connection when TLS resources are released. - Isolate ECIES adapter errors from existing caller OpenSSL state on verified runtimes; document known error-mark compatibility limits. - Drain messages admitted during producer replay without resending a frame within the same registration epoch. - Honor producer close while replacement registration or replay is pending, preserving `ProducerClosed` for retired sends. - Reduce the CRC32-C dependency chain with portable slicing-by-eight and independent polynomial regression tests. - Preserve receive-window capacity across reconnects with buffered messages. - Wake blocked receivers before waiting for the Broker's close response. - Reconnect and replay unconfirmed producer frames after a Broker `CLOSE_PRODUCER` notification, including Topic unloads. - Add a bounded recovery example that records per-message send and delivery events. Unknown sends remain unresolved unless delivery is observed; no end-to-end exactly-once guarantee is implied. - Release closed-handle callbacks and failed initial-connection resources; prevent closed handles from reconnecting and release terminal chunk buffers. - End a producer's own memory-admission waits on close and bound blocked sends/replays by their original deadline. Cancellation after a write starts retires that transport; other handles sharing it follow their existing failure or reconnect policy. - Retire silent connections after bounded keepalive probes, including blocked PING writes; complete incoming commands clear the pending probe. - Use the module's 0.1.2 identity consistently in CONNECT and AUTH_RESPONSE. Keep internal maintenance plans and raw evidence out of the published archive. The public API and pinned dependencies are unchanged. These notes were prepared for publication on the date above; confirm registry availability before installation. Actual publication is recorded in the matching GitHub Release. The October 29 source-bound checks on Pulsar 4.2.4 passed ten-message send/receive/confirmed-ACK and a separate real Writer replay with concurrent admission of a second send. Both original receipt futures completed; the initial A was also delivered, so this is not a deduplication or exactly-once guarantee. Retaining an unconfirmed send's original future across a Broker restart is deferred beyond 0.1.2 and remains unverified. See [development and verification](docs/testing.md) for reproducible checks and [verification records](docs/verification-0.1.2.md) for maintenance changes and source-specific evidence boundaries. SEND timeouts can leave broker acceptance unknown; the existing user-supplied sequence-ID limitation remains documented in the client guide. ## 0.1.1 — 2026-09-29 - Fix residual OpenSSL errors from failed TLS setup affecting other connections on the same event-loop thread. This caused intermittent TLS failures and test timeouts. - Add a deterministic regression test for missing private-key error isolation. - Clarify build requirements, feature-specific runtime dependencies, and test/example prerequisites in both READMEs. The public API is unchanged. These notes describe the release candidate; confirm availability in the Mooncakes registry before installation. ## 0.1.0 — 2026-09-29 Initial version notes, prepared for the scheduled release date above. Confirm actual availability in the Mooncakes registry; these notes alone do not establish that publication succeeded. - Native MoonBit Pulsar client with connection pooling, lookup, reconnect, producers, consumers and partition discovery. - Batching, compression, chunking, flow control, ACK/redelivery, retry and dead-letter handling. - Readers, TableView, transactions, schema codecs and selected Admin REST operations. - Token, OAuth2, Athenz and TLS authentication; RSA/ECIES message encryption. - Local mock tests, isolated Broker interoperability checks and recorded performance comparisons. Targets the native backend. A C compiler and SDK/libc are required; enhanced TLS and ECIES load OpenSSL 3 at runtime. Windows remains unverified. Full Admin REST coverage and end-to-end exactly-once behavior are not promised. See the [capability matrix](docs/capabilities.md) for verified scope, the [client guide](docs/client-guide.md) for interoperability limits, and [performance documentation](docs/performance.md) for measurement boundaries. Historical reports do not certify every later commit.