Janet erases stacktrace for an error while it is dying..

[amano-kenji]

When a janet script exits, (file/read stdin :line) returns nil. Thus,

(json/decode (file/read stdin :line))

causes

error: bad slot #0, expected string, symbol, keyword or buffer, got nil

without any stacktrace because janet is busy killing itself. It doesn't have stacktrace anymore.

It took hours to figure this out. What's the solution?

[bakpakin]

The description of the problem doesn't make sense to me, I would need some sample code to debug. It sounds like there is some multithreading involved.

[amano-kenji]

Yes, there are a lot of threads that stall the shutdown process.... Expect at least 7 threads.

file/read is called in a separate thread, too.

[amano-kenji]

I think file/read didn't have time to return nil if the shutdown was fast.

If there are many threads that stall the shutdown, then file/read has enough time to return nil which causes json/decode to throw an error because json/decode doesn't like nil.

Because janet is shutting down, it doesn't have stacktrace for json/decode.

It is basically a race condition.

[bakpakin]

If you apply the link mentioned here : https://github.com/janet-lang/janet/issues/1378#issuecomment-1918257564, does that fix the issue?

Whatever the issue is, I don't think the stacktrace is being "erased" as is suspected.

[amano-kenji]

I suspect janet shut down before stacktrace was printed?

Should I mechanically replace all invocations of exit with _exit and compile janet?

[amano-kenji]

Is it possible for file/read to return nil during shut down?

[bakpakin]

I suspect janet shut down before stacktrace was printed?

Should I mechanically replace all invocations of exit with _exit and compile janet?

Yeah, that is what I'm suggesting. After _exit called, nothing should be returning, as that is the root of all of this strange cleanup behavior

[amano-kenji]

I was able to reproduce this issue partially.

test.janet

(import spork/json)

(ev/spawn-thread
  (with [proc (os/spawn ["sleep" "100000000000"] :p {:out :pipe :err :pipe})]
    (os/proc-wait proc)))

(ev/spawn-thread
  (forever
    (json/decode (file/read stdin :line))))

(forever
  (ev/sleep 1))

Execute

$ sleep 1000 | janet test.janet

If you kill sleep, you will get an error message like

$ sleep 10000 | janet test.janet
error: bad slot #0, expected string, symbol, keyword or buffer, got nil
  in json/decode
  in _spawn-thread [test.janet] on line 9, column 5

So, I should check whether (file/read stdin :line) returns nil anyway.

My suspicion is that if you kill sleep and then test.janet very fast in sequence, you can kill the script before stacktrace is printed. That's probably why I could sometimes see the stacktrace and could not most of the time.

What do you think?

[amano-kenji]

I think I'm just closing this issue.... for now...