SIGCHLD side effects

[jeroen]

Our SIGCHLDhandler cleans child processes before the parent can get to the exit code. This affects all subprocesses:

res <- system("ping sdffsdsfsdfasdf.xyz", T)
attr(res, "status")
# [1] 68

However after sys is loaded the exit code is always 0:

library(sys)
res <- system("ping sdffsdsfsdfasdf.xyz", T)
attr(res, "status")
# NULL

This is bad. We should find a way to mark a child spawned by sys and only clean up these processes in our SIGCHLD handler.

[jeroen]

Alternatively we should abandon SIGCHLD and make the user manually clean children, or do it in some handle finalizer.

[gaborcsardi]

Yeah, that's bad.

Alternatively we should abandon SIGCHLD and make the user manually clean children, or do it in some handle finalizer.

That's fine with me, I can write a wrapper.

[jeroen]

Actuall are you sure zombies are a problem in the first place? At least on OSX:

for(i in 1:100) print(exec_background('whoami'))

I am not seeing zombies in top, even when disabling the signal handler.

[jeroen]

How does the parallel SIGCHLD handler avoid this problem?

[jeroen]

Ah I see it now. mcparallel keep track of a list of children that were spawned and only cleans up those.

[jeroen]

I have added a exec_status function to check the exit status of a pid, and optionally wait for it. The SIGCHLD handler is now commented-out in master until we find a solution without side effects.

@gaborcsardi can you check if exec_background() + exec_status() suffices for your use cases?

I still need to implement this for windows btw.