Closed gchiu closed 4 years ago
What was happening is that an erroring-but-unrelated request was able to make it appear that other reads were failing, when really it was raising an error for a stale previous request in the polling loop.
This seems to help the issue somewhat for the case I could reproduce:
https://github.com/metaeducation/ren-c/commit/31d0e5f4d5deea9aeea9edbd78a12a59277ca055
But as we discussed in chat, the entire question of how asynchronous failures are handled needs to get some review...though I have not wanted to be the one doing that design, myself. (!)
if a httpd server crashes it leaves ports open
so the client now thinks it's still got a connection to localhost which it tries to use again
where the server code is at https://gitlab.com/Zhaoshirong/rebol-chat/-/blob/master/src/s.reb commit 57c4b415e8f652beed97b7703af5cf2cf1d9f33b