help wanted: strange / unknown behaviour of gui.create_job

Comienza con la acción registrada mostrada y reproduce la cancelación de dt.gui.create_job tanto con dt.control.sleep como sin él. Rastrea el ciclo de vida de job.valid en torno a stop_job y a la asignación final, utilizando la implementación de la Lua API o los scripts existentes mencionados por el issue. La tarea estará terminada cuando el comportamiento de cancelación y el uso seguro del ciclo de vida del job estén aclarados o corregidos sin que se produzca el crash informado.

I'm currently writing (or, trying to write 🙄) a lua script that will generate the preview cache for selected images (without leaving dt to start the external program...). Basically, it seems to work but I see two strange things that I don't understand right now.

First, the essential parts of my script:

local function stop_job( job )
  job.valid = false
end

local function create_previews()
  job = dt.gui.create_job( "generating previews", true, stop_job )
  local images = dt.gui.action_images
  for i,image in pairs(images) do
    if job.valid then
      image:generate_cache( true, 0, 8 )
      job.percent = i / #images
      dt.control.sleep( 1 ) -- without this, the job cannot be aborted
    else
      break
    end
  end
  if job.valid then -- without this, dt crashes with an exception in dt_pthread_mutex_destroy() after aborting the job
    job.valid = false
  end
end

the create_previews function will be called from a registered action.

My two questions:

1. Without the sleep() I cannot stop the running job. After starting the job, the progress bar is shown and when I hit the 'X' nothing happens until the for loop processed all images. Then the stop_job() is called. With the sleep the stop_job is called after processing the current image and the job terminates as desired.

2. When I set the job.valid to false after the loop it only works as long as the job is not terminated via the stop_job function. As this already sets the job to invalid I guess it is already destroyed and in the second job.valid=false dt crashes with an exception in dt_pthread_mutex_destroy.

Both phenomena are completely understandable, but: I see a lot of scripts that execute job.valid=false twice (when aborted) and seem to rely on stopping without a call to sleep(). Hmmm. What I'm missing here?

I hope you can clarify this,
thanks a lot!
Christian