job_stop cannot kill process tree on windows. #1184
Description
Version:
GVim - 8.0.45 32 bits on Win7 x64
Reproduce:
- Start a batch file by
myjob = job_start(['cmd.exe', '/C', 'abc.cmd'], ...) - Start a long-time-running command in the batch file.
- Stop job by
call job_stop(myjob, 'kill')
Expect:
All the child processes should be killed according to the document.
Result:
Only cmd.exe gets killed, all child processes are still running, close_cb and exit_cb are not invoked.
Channel log:
==== start log session ====
9.211493 : Starting job: C:\Windows\system32\cmd.exe /c C:\Users\Linwei\AppData\Local\Temp\asyncrun.cmd
9.211862 on 0: Created channel
9.276309 : looking for messages on channels
9.377826 RECV on 0: 'Entering dir 'e:/lab/trunk'
python timer 0
'
9.377837 on 0: Invoking channel callback <SNR>83_AsyncRun_Job_OnCallback
9.377921 on 0: Invoking channel callback <SNR>83_AsyncRun_Job_OnCallback
9.393713 : looking for messages on channels
10.428161 RECV on 0: 'python timer 1
'
10.428197 on 0: Invoking channel callback <SNR>83_AsyncRun_Job_OnCallback
10.470649 : looking for messages on channels
11.432531 RECV on 0: 'python timer 2
(call job_stop, and cmd.exe exit here)
'
11.432569 on 0: Invoking channel callback <SNR>83_AsyncRun_Job_OnCallback
11.437469 : looking for messages on channels
12.220074 on 0: Stopping job with 'kill'
12.239247 : looking for messages on channels
12.239365 on 0: Job ended
12.283459 : looking for messages on channels
12.409692 RECV on 0: 'python timer 3
(but child process started by cmd.exe is still running)
'
12.409729 on 0: Invoking channel callback <SNR>83_AsyncRun_Job_OnCallback
12.429221 : looking for messages on channels
13.401916 RECV on 0: 'python timer 4
.............
'
17.432877 on 0: Invoking channel callback <SNR>83_AsyncRun_Job_OnCallback
17.453741 : looking for messages on channels
18.366814 RECV on 0: 'python timer 9
'
19.374784 on 0: Invoking channel callback <SNR>83_AsyncRun_Job_OnCallback
19.376939 : looking for messages on channels
19.390777 ERR on 0: channel_handle_events(): Read error from ch_part[1], closing
19.390886 : looking for messages on channels
19.390914 on 0: Closing channel because all readable fds are closed
19.390935 on 0: Closing channel
19.390969 on 0: Invoking callbacks before closing
19.390991 on 0: Invoking close callback <SNR>83_AsyncRun_Job_OnClose
19.391923 on 0: Freeing job
19.443185 on 0: Closing channel
19.443211 on 0: Clearing channel
19.443219 on 0: Freeing channel
19.443272 : looking for messages on channels
content of asyncrun.cmd:
@echo off
python timer.py
content of timer.py:
import time, sys
print 'Entering dir \'e:/lab/trunk\''
for i in xrange(10):
print 'python timer', i
sys.stdout.flush()
time.sleep(1)
print 'Leaving dir'Escaping and passing arguments correctly on windows are complex. Most plugins I see, including taglist/tagbar, will use a temporary batch to run commands on windows. In this case, the real job can never be killed if job_stop can only stop cmd.exe.
Most jobs have more than one process (eg, a building job, piped commands). If I start a complex grep like program on the root of my filesystem by mistake, which will be running for a long time, I can never stop them on windows ?
If I start a group of processes with the wrong parameters (like tail -f, findstr abc), which continues waiting data from stdin, but in_io is unset, they will never get stopped in vim ?
Most piped commands (dir | grep DIR) require cmd.exe to interprete the command line on windows, will they also unable to be killed by job_stop ?
If a lot of jobs hang in vim, and continue occupying the job slot, we will never able to start a new job until we exit vim and reenter again ?