* feat: allow msgIdProvider to fail
Closes: #642.
Changes the return type of the msgIdProvider to `Result[MessageID, string]` so that message id generation can fail.
String error type was chosen as this `msgIdProvider` mainly because the failed message id generation drops the message and logs the error provided. Because `msgIdProvider` can be externally provided by library consumers, an enum didn’t make sense and a object seemed to be overkill. Exceptions could have been used as well, however, in this case, Result ergonomics were warranted and prevented wrapping quite a large block of code in try/except.
The `defaultMsgIdProvider` function previously allowed message id generation to fail silently for use in the tests: when seqno or source peerid were not valid, the message id generated was based on a hash of the message data and topic ids. The silent failing was moved to the `defaultMsgIdProvider` used only in the tests so that it could not fail silently in applications.
Unit tests were added for the `defaultMsgIdProvider`.
* Change MsgIdProvider error type to ValidationResult
Adds a pin-file for CI. When a dependency break something in the libp2p, we don't want for the CI to be
broken on every branch, just on a "bump" branch that will update the pin-file & fix the eventual interop issue.
Currently, `ecnist`'s `toBytes` and `getBytes` methods operate only on
properly initialized keys. If an un-initialized key is given, an
`IndexError` may be raised if the key's `xlen` / `qlen` property is
larger than the maximum buffer size. This patch hardens those functions
to report a proper error in that case.
Note that the library functions called by `init` and `initRaw` already
reject data that does not have the expected length, so these new checks
should not be reachable in practice.
* fix: remove returned Futures from switch.start
The proc `start` returned a seq of futures that was mean to be awaited by the caller. However, the start proc itself awaited each Future before returning it, so the ceremony requiring the caller to await the Future, and returning the Futures themselves was just used to handle errors. But we'll give a better way to handle errors in a future revision
Remove `switch.start` return type (implicit `Future[void]`)
Update tutorials and examples to reflect the change.
* Raise error during failed transport
Replaces logging of error, and adds comment that it should be replaced with a callback in a future PR.
When closing a connection of clientA(using the command “/disconnect”), the connection is then closed on clientA’s side. However, the connection remained open on clientB and clientB could continue sending message to clientA and clientA would receive those messages.
This PR listens for a closing connection on clientB, then also closes clientB’s connection.
* add test for multiple local addresses
* allow transports to listen on multiple addrs
* fix tcp transport accept
* check switch addrs are correct
* switch test to port 0
* close accepted peers on close
* ignore CancelledError in transport accept
* test ci
* only accept in accept loop
* avoid accept greedyness
* close acceptedPeers
* accept doesn't crash on cancelled fut
* add common transport test
* close conn on handling failure
* close accepted peers in two steps
* test for macos
* revert accept greedyness
* fix dialing cancel
* test chronos fix
* add ws
* ws cancellation
* small fix
* remove chronos blocked test
* fix testping
* Fix transport's switch start (like #609)
* bump chronos
* Websocket: handle both ws & wss
Co-authored-by: Tanguy Cizain <tanguycizain@gmail.com>
Co-authored-by: Tanguy <tanguy@status.im>
Tanguy
Eric Mastro
Ștefan Talpalaru
Dmitriy Ryajov
Jacek Sieka
Etan Kissling
Eric Mastro
Tanguy
Zahary Karadjov