How to update pipeline steps created with add_function_step?

[kiranzo]

Not sure if this is really a bug, or I'm doing something wrong.

Describe the bug / To reproduce

I inherited some ClearML pipelines from the previous dev. I made some changes to pipeline step (saving artifact in parquet instead of csv and a parameter that lets you choose which), that required to change both code of the step and code of the controlling pipeline, then I committed them to a test branch on Git. I cloned the existing pipeline in ClearML Web UI and tried to change source branch in the pipeline parameters, but it didn't affect behaviour of the pipeline step. I tried to go to 'configuration objects' menu and change the branch for that particular step, but it seems it ignored Git commit and just took the code from the 'uncommitted changes' in the step configuration.

I tried to clone my pipeline again and clear configuration object for that step altogether, or just set "diff" to null, but in both cases, I got this error message: /usr/bin/python3: can't open file '/root/.clearml/venvs-builds/task_repository/pipelines.git/create_clearml_dataset.py': [Errno 2] No such file or directory It seems it just can't generate new code from the function when there's no diff.

Expected behaviour

Was I supposed to use Pipeline.run() and create everything completely anew instead of cloning (I'm not sure I can do that easily, given that pipeline is supposed to run in speacially prepared Docker container)? It's really counterintuitive what we should do when pipeline steps are created from functions. What's the easiest way to modify the code of the pipeline step (easiest = best for lazy researchers)?

Environment

• Server type - self hosted
• ClearML SDK Version - 1.16.3
• ClearML Server Version - 1.14.1
• ClearML Agent Version - 1.8.1
• Python Version - 3.8
• OS - Linux