Reduce usage of SCRIPT_NAME

[marc-hb]

Just simplifying and reducing the usage of the SCRIPT_NAME indirection, see commit messages.

v1 was fixing 1398cf4 which broke logging by moving the logs of both tests into the same logs/multiple-pipeline/ subdirectory but @xiulipan and @aiChaoSONG said we don't need to move the logs back to where they were, so this PR does not move anything anymore, it's just cleanups now.

For v1 https://sof-ci.01.org/softestpr/PR546/build487/devicetest/ was all green

[marc-hb]

v2 https://sof-ci.01.org/softestpr/PR546/build488/devicetest is all green

[xiulipan]

I do not see any need to use a tmp file here, this will bring more trouble with sub-test
why not use readarray -t cmd_lst < <(pgrep -P $$ -a)

And the whole logic is changed to me, the parent will still be killed by itself now.

How is # can't run pgrep in any subshell because the latter would pollute the list take effect here

[xiulipan]

I checked pgrep -P $$ -a it will show the script it self, how do you remove it from the list

[marc-hb]

Wrong, demo:

hostname:~$ pgrep -P $$ -a
hostname:~$ sleep 1000 &
[1] 140297
hostname:~$ 
hostname:~$ ps f
    PID TTY      STAT   TIME COMMAND
 140238 pts/3    Ss     0:00 -bash
 140297 pts/3    S      0:00  \_ sleep 1000
 140298 pts/3    R+     0:00  \_ ps f
 139197 pts/2    Ss+    0:00 -bash
 138772 pts/1    Ss+    0:00 /bin/sh -i
hostname:~$ echo $$
140238
hostname:~$ pgrep -P $$ -a
140297 sleep 1000

[marc-hb]

I do not see any need to use a tmp file here,

The new file is not temporary, it's part of the error logs. Most tests should fail when a process is unexpectedly left behind... but off-topic for now.

How is # can't run pgrep in any subshell because the latter would pollute the list take effect here

Try removing just grep -v "$SCRIPT_NAME" from the existing code (before this PR) and you will see that the <( ) subshell is caught every time. What other reason would the grep -v "$SCRIPT_NAME" be there for?

this will bring more trouble with sub-test

Good point, I think I missed that.

And the whole logic is changed to me

The logic is exactly the same, the only difference is removing the subshell to avoid the awkward filtering.

I checked pgrep -P $$ -a it will show the script it self, how do you remove it from the list

I wonder what you tested because if you run justpgrep -P $$ -a without anything else then it prints nothing. This is expected because a process is not its own parent.

FYI a lot of actual testing went into this PR (except for the subtests, my bad)

[xiulipan]

@marc-hb Try with

cat test_a.sh
#!/bin/bash

aplay -Dhw:0,0 -f dat /dev/zero &
readarray -t cmd_lst < <(pgrep -P $$ -a)

if [ ${#cmd_lst[@]} -gt 0 ]; then
        for line in "${cmd_lst[@]}"
        do
            echo "Catch pid: $line"
            echo "Kill cmd:'${line#* }' by kill -9"
            kill -9 "${line%% *}"
        done
fi

The output is interesting

./test_a.sh
Playing raw data '/dev/zero' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo
Catch pid: 16164 aplay -Dhw:0,0 -f dat /dev/zero
Kill cmd:'aplay -Dhw:0,0 -f dat /dev/zero' by kill -9
Catch pid: 16165 /bin/bash ./test_a.sh
Kill cmd:'/bin/bash ./test_a.sh' by kill -9
./test_a.sh: line 11: kill: (16165) - No such process

See the log try to kill test_a.sh Kill cmd:'/bin/bash ./test_a.sh' by kill -9
This will why the old code used grep -v to remove the scripts itself.

 cat test_b.sh
#!/bin/bash

aplay -Dhw:0,0 -f dat /dev/zero &
readarray -t cmd_lst < <(pgrep -P $$ -a | grep -v $0)

if [ ${#cmd_lst[@]} -gt 0 ]; then
        for line in "${cmd_lst[@]}"
        do
            echo "Catch pid: $line"
            echo "Kill cmd:'${line#* }' by kill -9"
            kill -9 "${line%% *}"
        done
fi

Above test_b.sh will only kill the aplay.

./test_b.sh
Playing raw data '/dev/zero' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo
Catch pid: 16261 aplay -Dhw:0,0 -f dat /dev/zero
Kill cmd:'aplay -Dhw:0,0 -f dat /dev/zero' by kill -9

And that is why I said the logic is different here as I see different behavior with above test scripts.

[plbossart]

@marc-hb is this still relevant or needs to be closed?

[marc-hb]

I've been using this locally and successfully for months. I got confused by some Xiuli's review, will take another look. No one else reviewed unfortunately.

[marc-hb]

This will why the old code used grep -v to remove the scripts itself.

No, the old code used grep -v to exclude the <( ) SUBSHELL (with the same name), not the main shell which pgrep -P already knows how to exclude. That's why it says "no such process": because the subshell has already completed by the time the main script tries to kill it.

So the only issue left is to make sure sub-tests don't collide their respective children_left.txt files.

You can see the PID is NOT the main shell if you add ps f

--- xiuli-pgrep.sh	2021/06/24 23:45:51	1.1
+++ xiuli-pgrep.sh	2021/06/24 23:46:10
@@ -4,6 +4,8 @@
 aplay -Dhw:0,0 -f dat /dev/zero &
 readarray -t cmd_lst < <(pgrep -P $$ -a)
 
+ps f
+
 if [ ${#cmd_lst[@]} -gt 0 ]; then
         for line in "${cmd_lst[@]}"
         do

You cannot see the subshell <( ) because you cannot ass ps f inside it.

[marc-hb]

https://stackoverflow.com/questions/392022/whats-the-best-way-to-send-a-signal-to-all-members-of-a-process-group

stale