v12.6.5

Patch Changes

• #238 5310b93 - Reject an aborted request with the reason of its signal, and stop reporting it to the onError hook

  An aborted send() used to reject with a CcRequestError carrying the ABORTED code. It now rejects with signal.reason, as fetch() does. That is a DOMException named AbortError, unless abort() was given another reason. ABORTED is removed from CC_REQUEST_ERROR_CODES.

  A stream whose request signal aborts while it connects rejects start() the same way, instead of reconnecting when it has a retry configuration.

  To detect an abort, check the signal you passed rather than the error:

   try {
     await client.send(command, { signal });
   } catch (error) {
  -  if (isCcRequestErrorWithCode(error, 'ABORTED')) {
  +  if (signal.aborted) {
       return;
     }
     throw error;
   }

• #238 fc35d6a - Stop failing event streams opened at the same time on the same URL

  Both streams used to fail before receiving any event, one with SSE_SERVER_ERROR and the other with a locked ReadableStream error.

• #238 cb1ee89 - Stop rejecting a request because an identical request sent at the same time was aborted

• #238 e6929ae - Make CcNetworkError.isWorthRetrying() answer for the command of each caller when identical requests sent at the same time fail together

  A caller used to get the answer computed for the first one, even when its command declared a different idempotence.

• #238 55c7ccb - Stop setting a cause on the errors raised without one

  CcClientError and its subclasses used to carry a cause set to undefined, which util.inspect() printed. The cause of a DomainParseError is no longer enumerable either, so JSON.stringify() and the spread operator leave it out, as they do for native errors.

• #238 55c7ccb - Name the errors the client throws after their class

  CcClientError, CcRequestError, CcNetworkError, CcHttpError, DomainParseError, PollingInterruptedError and PollingTimeoutError used to report Error as their name. Their stack traces and String(error) started with Error: too.

• #240 ffc6d3b - Streams now reconnect in Firefox when the server cuts the connection mid-response

  Firefox words that failure differently from the other engines, and the client did not recognise it as a
  network failure. A stream hit it on every server restart, failed for good instead of retrying, and
  rejected with a bare TypeError instead of a CcNetworkError.

• #238 7b2b896 - Stop failing a command because another command resolving the same owner or add-on id at the same time was aborted

• #238 2122126 - Call the onError hook once per failed send(), and never for an error a composite command recovers from

  A composite command used to report an inner error twice. It also reported the errors it tolerates itself, such as the 404 ListDomainCommand gets for an application without a primary domain, or the 404 responses polled while waiting for a network group or a Kubernetes cluster.

• #238 7d1febc - Call the onError hook once per error, even when it rejects several send() calls

  Resolving the owner or add-on id of a command used to report a failure twice. Identical GET requests sent at the same time share one request, and used to report a network failure or a timeout once per call. An HTTP error response is still reported once per call.

• #238 815a84a - Treat an option set to undefined in a request or stream configuration as an absent option

  It now keeps the value it would have without it. { signal: controller?.signal } without a controller used to drop the signal of defaultRequestConfig, so aborting that signal no longer aborted the request. Likewise, timeout: undefined made a request fail at once with TIMEOUT_EXCEEDED, and isCorsEnabled: undefined overrode the value a command needs.

• #238 7da0732 - Keep a stream from connecting once close() was called

  A stream closed right after start(), or while it reconnected, could still connect and stay open with no listeners.

• #238 d41c56e - Disable the retries of a stream when its configuration sets retry to null

  Both client.stream(command, { retry: null }) and defaultStreamConfig: { retry: null } used to keep retrying with the retry configuration they were meant to override.

• #238 2e2b2f0 - Close a stream when the signal of its request aborts, whenever it aborts

  start() now rejects with signal.reason as soon as the signal aborts: before the stream starts, while it reads events, while it is paused, or while it waits to reconnect. The stream used to ignore a signal aborted before it started. An abort while reading failed with SSE_SERVER_ERROR, or reconnected with a retry configuration. An abort while paused or waiting to reconnect only closed the stream at the next attempt.

• #238 7cb3fb2 - Keep resume() from opening a second connection while a stream waits to reconnect

  Both connections delivered their events, and close() left one of them open.