s2-streamstore/s2-sdk-typescript

Patch Changes

• c543af5: fix: abort during an in-flight append no longer corrupts capacity accounting (#203)

• df2daab: - Treat browser TypeError("network error") as retryable SDK network error.

  • Make fetch read-session cleanup tolerate reader.cancel() failures instead of leaking raw errors.
  • Defensively retry if a transport read stream throws directly.
  • Increase fetch SSE inactivity watchdog from 20s to 25s.
  • Remove misleading stale-timeout debug log & add some additional context to other logs
  • Add regression tests and patch changeset.

• 1adac12: fix: metrics basin()/stream() no longer send path params in the query string (#235)

  Both methods spread ...args into the request query, so basin (and stream) were
  serialized into the URL query string in addition to the path. They now pass path and
  query explicitly — only { basin } / { basin, stream } in the path and only
  set/start/end/interval in the query — matching the OpenAPI spec and the Go and Python SDKs.

• 0acd383: fix: S2Stream.read() no longer ignores the count/bytes budget when filtering command records (#231)

  When called with ignoreCommandRecords: true, the unary read() previously looped and
  re-issued a full read request (with the original count/bytes) for every command-only
  batch, amplifying a single count: 1 request into many requests. It now performs exactly
  one request and filters command records out of that single batch client-side, matching the
  Rust, Go, and Python SDKs.

  The returned batch may be empty when every record in it was a command record. Use
  readSession() to transparently keep reading until data records are found — the streaming
  session already tracks the remaining budget correctly across reconnects.