Possible websocket connection closing bug

[Tectu]

Using malloy-example-server-websocket and connecting to the /timer endpoint does correctly send the 10 messages with 1 second interval. However, it seems like the connection closing is not working as expected.

I am using websocat as a client. On first glance it seems like the connection is not closed properly.

This needs some investigation and possibly fixing.

[0x00002a]

Yeah I ran into the same issue during testing, its an issue with websocat (issue) you need to pass -E

[Tectu]

There still seems to be an issue so I am reopening this.

When running the malloy-example-server-websocket example and using websocat -E the server crashes due to an uncaught exception after sending the 10th message:

C:\Users\joel\Documents\projects\malloy\cmake-build-debug\examples\server\websocket\malloy-example-server-websocket.exe
[2021-07-02 16:37:06.431] [router] [debug] adding websocket endpoint at /
[2021-07-02 16:37:06.432] [router] [debug] adding websocket endpoint at /echo
[2021-07-02 16:37:06.433] [router] [debug] adding websocket endpoint at /timer
[2021-07-02 16:37:06.433] [debug] starting server.
[2021-07-02 16:37:06.434] [debug] starting i/o context.
[2021-07-02 16:40:31.448] [listener] [info] accepting incoming connection from 127.0.0.1:7026
[2021-07-02 16:40:31.449] [http_connection] [debug] launching plain connection.
[2021-07-02 16:40:31.449] [http_connection] [info] upgrading HTTP connection to WS connection.
[2021-07-02 16:40:31.449] [router] [debug] handling WS request: GET /timer
terminate called after throwing an instance of 'boost::wrapexcept<boost::system::system_error>'
  what():  End of file

Process finished with exit code 3

[0x00002a]

Found the bug: https://github.com/Tectu/malloy/blob/9eecc9cc9da9ca22d28323fcaec832df9a974ac0/lib/malloy/websocket/stream.hpp#L75 this calls the non error_code version meaning it will throw on failure. I'm currently writing up an issue meant to be a discussion of how we deal with stuff like this (how to report errors, whether to use error code or exceptions mostly, etc) but this should probably be async anyway.

[Tectu]

40 is definitely something very vital for the future of this library and needs attention by itself.

I am currently unable to check boost.beast sourced/API but I do think that there is an overload that takes an error_code and/or async.

[0x00002a]

Yes there is, I misunderstood the API there, I though the close_reason was filled and reported any errors but its meant to actually have a value (not sure if that might be part of the issue with websocat too). So yeah that code needs just plain fixing :p

[Tectu]

hehe :p are you going to submit a patch or should I put this on my ToDo list?

[0x00002a]

I should have time to do it tomorrow. I'll fix it in the log style and we can potentially change it after #40 is resolved

[Tectu]

Personally I'd put out a fix for this bug before tackling #40 given that at least on the surface this looks like an easy fix.