* Update PROTOCOL.md

* Update PROTOCOL.md

Co-authored-by: Jacky Zhao <j.zhao2k19@gmail.com>

* Update PROTOCOL.md

* half-close semantics

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* fmt

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Update PROTOCOL.md

* Include reader-writer semantics

---------

Co-authored-by: Jacky Zhao <j.zhao2k19@gmail.com>

* ReadStream implementation

* remove tee for now

* Make the stream itself an iterable

* Implement return for iterator

* comment

* Client uses ReadStream

* Client uses WriteStream

* delete port config lol

* Disable everything for handler.test.ts

* no void 0

* Server uses streams

* Bruh

* I -> Input, O -> Output, E -> Err

* Update types

* Update implementation

* Update tests

* Misc changes

* fix checking init instead of input

…le onClose (#160)

…back style onCloseRequest (#164)

* [protocolv2] Implement close requests

* Add some more checks

* Add comments on for control flags

* Revert "Add some more checks"

This reverts commit 20e1c10.

* [protocolv2] Make ReadStream always emit Result values

* [protocolv2] Adapt to stream changes from #172 PR (#173)

…175)

Old stream closures are more akin to what abort is supposed to do
so it's better to swap them to maintain better compatibility
#175 (comment)

- Used extra instead of extras
- Didn't inforce literal values
- No way of validating error results properly

…ntation

## Why

Clients can send invalid requests for many reasons, most commonly due to
backwards incompatible server changes, server should handle those and
send back a stream abort with it.

## What changed

- Introduced `INTERNAL_RIVER_ERROR` code since some of the errors are
invariant violations on the server
- Split out stream request validation from stream handling
- When we see a bad request, we send `INVALID_REQUEST` code with an
abort bit
- Made tracing `createHandlerSpan` accept tracing fields explicitly
instead of a transport message

some more changes that I'll note inline

…s it (#204)

## Why

When the server aborts a stream (due to invalid request or explicit
abort) the client send more messages before it gets the abort. We should
make sure this doesn't lead to errors and unnecessary follow up
responses.

## What changed

Keep an LRU of aborted streams, if a new message comes in for a stream
in our set, we simply ignore the message.

#202 (comment)

#217)

#202 (comment)

#203 (comment)

## Why

I ended up needing this in river-babel, it's probably a useful API to
have

## What changed

- Added `WriteStream.onClose`
- Unrelated: Made `WriteStream.close` lose reference to `onCloseRequest`
listeners and unset `write` function

## Why

It's easier to roll out this way.

## What changed

- Added `protocolVersion` to transport
- Added allow-list for accepted protocol versions
- Use `protocolVersion` in the server router to determine how we should
deal with closes and aborts

…patible (#236)

## Why

In v1.1 sometimes the first message is not `init`, but instead it's the
`input`. Let's handle that.

## What changed

If we get an `init` that's not matching what we expect, we'll check if
the client is on v1.1 and sent us an `input` as a first message, then
we'll pass the first message as `input` and treat `init` as an empty
object (which is what the first pass of requiring `init` for `stream`
and `upload` be).

## Versioning

- [ ] Breaking protocol change
- [ ] Breaking ts/js API change

<!-- Kind reminder to add tests and updated documentation if needed -->

Got some comments on streams being complicated, so this is a take on
simplified API.

- Less methods and states that people need to understand
- Merge interfaces into a single object (i.e. a `stream` call would have
both read and write capabilities on the same object)

## Why

Follow up on #249 to actually use the new interfaces. Removed extra
refactor done in #249 that merges the interfaces as that proved to be a
challenging API (un-yak-shave 🙅 🐃)

## What changed

- Removed close requests (mostly cherry-picked from #248)
- Otherwise a simple swapping out of the interfaces

## Versioning

- [ ] Breaking protocol change
- [ ] Breaking ts/js API change

<!-- Kind reminder to add tests and updated documentation if needed -->

Just a simple follow up to #250

Partially to match gRPC nomeclature and to avoid confusion with AbortSignal/AbortController