Try to workaround race condition for @ run

[amosbird]

Address 38fed8b#commitcomment-142137229

[kovidgoyal]

Rather than mess with the loop in socket-io I chose to just run the
background process after an event loop tick, which should take care of
it, cant be sure as I cant reproduce.

[amosbird]

I chose to just run the background process after an event loop tick

This definitely looks better. However, after some tests, it seems that add_timer(callback, 0, False) doesn't queue the callback properly to make sure AsyncResponse() has been sent.

I've tested add_timer(callback, 0.01, False), 18/20 work fine. With add_timer(callback, 0.1, False), 100% good.

[amosbird]

cant be sure as I cant reproduce.

I'm on a remote ssh connection with kitty domain socket forwarding, which makes it easy to reproduce. I guess you can simply add time.sleep(...) before returning AsyncResponse() to reproduce it :)

[kovidgoyal]

That does not reproduce it. I dont see why it would either. add_timer
schedules the callback it wont actually run till after
response_from_kitty returns as response_from_kitty is running on the
main thread and blocking the event loop. You can trace this sequence of
events. In child-monitor.c around line 500 you can see the return value
being sent. You can add

log_error("%s", msg)

at the top of send_response_to_peer() that will show you everything that
is sent to the peer in order.

[amosbird]

Interesting. Both good and bad cases print the same event log which is indeed in correct order:

[35.176] ␛P@kitty-cmd{"ok": true, "stream": true}␛\
[35.182] ␛P@kitty-cmd{"ok": true, "data": {"stdout": "MQo=", "stderr": "", "exit_code": 0, "exit_status": 0}}␛\
[35.191] (null)

I guess something is wrong at the kitten side.

[kovidgoyal]

You can add

forward_stdio yes

to kitty.conf

then in socket_io.go

you can do

import "kitty/tools/tty"

and then use

tty.DebugPrintln(...)

in simple_socket_io() to see whats going on. The output of the prints
will be in the stderr of the kitty inside which the kitten runs.

[amosbird]

https://github.com/amosbird/kitty/blob/master/tools/cmd/at/socket_io.go#L61

Here bad cases read two responses in one Read call and parse them together, which will swallow the first AsyncResponse().

[kovidgoyal]

Should be fixed by: 703068f

[kovidgoyal]

Or actually no maybe not quite, the parser can call handledcs multiple times in one read

[kovidgoyal]

Next commit fixes it, I hope.

[amosbird]

Unfortunately, kitten @ run gets stuck when packet coalescing happens with a661f00 Fix the previous fix.

[kovidgoyal]

a9924d2

[amosbird]

a9924d2 works! Thank you!