[Completion bug] Completion only works once

[LStruber]

Minimum steps to reproduce:

• Open your favorite project
• Add a smooth brick in the pipeline manager
• Export all unconnected mandatory plugs
• Select a file in the input node
• Run the brick
• Go to data browser, and see it worked
• Go back to pipeline manager
• Open a new tab
• Add a smooth brick in this tab (or another one)
• Export all unconnected mandatory plugs
• Select a file in the input node
• Run the brick

It fails during initialization with the following message:

Tested on a fresh bv_maker, with populse_mia on master and mia_processes on master, both up-to-date

NOTE: If you run the brick alone (without exporting plugs), it works !

By investigating a bit, I identified the problem: In init_pipeline() function of pipeline_manage_tab.py, the workflow returned by workflow_from_pipeline function (of Capsul) should contains the output file (in example case smoothed_files). This is the case at the initialization of the first brick, and not the second time (see self.workflow.jobs[1].param_dict['smoothed_files'] after creating the workflow in both cases). This make me think that the issue could come from Capsul... @denisri would you have an idea why "sometimes" the output file is not present in the workflow param_dict ?

[LStruber]

Note that you can relaunch the first pipeline, even after the error occured

[denisri]

I'm looking at it. Completion seems to fail for the second pipeline. I don't know why it's different and why it fails (yet), but I can reproduce the problem.

[denisri]

OK I think it's fixed: it did not actually depend on the 1st/second tab in the pipeline manager, but whether you had clicked on the node using the left button before exporting its parameters ;) I explain:

• when you click on a node, it displays its controller GUI, and while doing it it tries to instantiate a completion engine for its process (using the Process instance). Once instantiated, the completion engine is cached and will be reused later.
• When you export the parameters, then complete parameters for the pipeline, the completion system will try to instantiate a completion engine for its nodes, using the Node instance, not the Process instance (the Node here contains a Process but is not one). If the completion engine has been cached it will be reused (with the Process registered with it), otherwise it will be created (but with the Node registered with it). It should work using either the Node or the Process.
• But in Mia the overload of the completion system had a bug using Nodes, it actually expected a Process, thus was failing when instantiated using the Node. This explains the problem. It has been fixed now. You can check.

[servoz]

@LStruber can you check and report from your side, please ?

[LStruber]

It's indeed working now. I close the ticket.