Killing a background process while building a bash prompt

Question

I have managed to make my bash prompt lazy loads some components to have the feeling of an async rendering. The lazy loading function runs as a background process with (render_async &).

From the screencast below you can see how the prompt actually loads part of it right away and the other "lazy loads". However, what I noticed is that if you change the directory before the async part is loaded the new prompt gets overwritten with wrong context.

My thought process of fixing this is the following:

• Check for background process using the jobs command
• I discovered that the jobs command shows a wd:* next to processes that are not in the background of the current working directory
• Find all and kill all background processes that have that pattern to avoid this overwriting

I added a sleep to my async command to simulate that and I can see the background process hanging.

[1]   Running                 render_async &
[2]-  Running                 render_async &  (wd: ~/Projects/My Personal Space/blog-core)
[3]+  Running                 render_async &

I went ahead and wrote:

  jobs | grep 'render_async.*wd:' | cut -d "[" -f2 | cut -d "]" -f1 | while read -r line ; do
    kill "%$line"
  done

Which in theory should parse the id of the background jobs in other directories and kill them. However in practice I keep getting for the example above kill: %2: no such job.

When I execute the same kill command in the shell itself it works perfectly.

Would appreciate any help here.

Other thoughts

I have tried naming my forked process (name each async process by appending the CWD to the function name) inspired from https://askubuntu.com/questions/153900/how-can-i-start-a-process-with-a-different-name where:

bash -c "exec -a MyUniqueProcessName <command> &" replaces the current shell, no new process is created, that's why I'm starting a new shell to call exec.

Then you can kill the process with:

pkill -f MyUniqueProcessName You can start more than one process under the same name, then pkill -f will kill all of them.

but that kept telling me that my async function that I was trying to pass to the <command> is not found and I suspect it has something to do with the fact that it is a custom function and that is forking a new bash process but I am not an expert here.

Thanks

Answer 1

Piping creates an implicit subshell, so jobs from the parent are not visible inside the piped-to while loop where the kill is running. How about this, which will at least kill the first job?

job="$(jobs | grep 'render_async.*wd:' | cut -d "[" -f2 | cut -d "]" -f1)"
kill "%$job"

In my tests, the $(...) happens in the current shell, at least initially, so the job table is visible. Example:

$ cat &
[1] 8296
$ jobs
[1]+  Stopped                 cat
$ echo "$(jobs| cut -d "[" -f2 | cut -d "]" -f1|head -1)"
1

(By the way, does the job you want to kill always have the same job number? Can you just hardcode %2, or whatever it may be?)

Edit Multijob:

joblist="$(jobs| sed -E 's/^[^0-9]*([0-9]+).*$/\1/'|tr '\n' '@')"
    # E.g., 1@2@
IFS='@'     # Split on @
for job in $joblist     # No double-quotes!
do
    kill "%$job"
done

Working example from a shell prompt:

$ cat&
[1] 8824
$ cat&
[2] 6452

[1]+  Stopped                 cat
$ jobs
[1]-  Stopped                 cat
[2]+  Stopped                 cat
$ joblist="$(jobs| sed -E 's/^[^0-9]*([0-9]+).*$/\1/'|tr '\n' '@')"
$ IFS=@
$ for job in $joblist; do kill "%$job" ; done

[1]-  Stopped                 cat

[2]+  Stopped                 cat
$ jobs
[1]-  Terminated              cat
[2]+  Terminated              cat
$ jobs
$

I chose @ as a separator because I don't think it means anything particular to the shell.

I don't know if it makes a difference in the PS1-function environment.