Change on Python Basic Console not transferred to tools about spine-toolbox

[PekkaSavolainen]

This is not a bug but a feature. The Python interpreter (and Julia Executable) settings in File->Settings->Tools are just the default settings for new Tool specifications. This is mentioned in the User Guide and in the File->Settings->Tools tooltips.

However, we could add a button into the Tool Specification Editor that automatically updates the execution settings into the default settings defined in File->Settings->Tools.

[dlr-cjs]

If this is a feature, it makes sharing workflows unpleasant, since each user needs to update a workflow (possible large workflow comprising dozens of tools) manually. I suggest a button on the workflow level allowing to "update basic console" in all comprising tools.

To be more precise what my issue is:
I am using spinetoolbox in TradeRES to share my workflow with others. I require a different environment for my Python tool "AMIRIS", since we have a higher minimum requirement for protobuf. Thus, I need the users to change to another environment for the Python tools in my workflow (luckily its only three tools).

Currently, users need to follow these instructions to accomplish this... which is rather annoying.

Thus, the "update console" process should be as easy as possible.

[PekkaSavolainen]

I suggest a button on the workflow level allowing to "update basic console" in all comprising tools.

This is a good idea.
,
But I'm not sure why you are instructing user's to edit the default python interpreter and then in each Tool spec you switch to Julia and then back to Python? I guess you want to avoid navigating to the same Python interpreter file multiple times, which can be pretty tedious.

There is however a little bit simpler option available. You could try running the Tool Specs with the Jupyter Console option. There is a section in the user guide on how to do this. https://spine-toolbox.readthedocs.io/en/master/setting_up.html#conda

When you have created the AMIRIS Conda environment, you could simply instruct users to:

1. Make sure that the Select Miniconda executable is set in File->Settings->Tools
2. Double click on a Tool in Design View
3. Select Jupyter Console radio button
4. Select the AMIRIS envronment from the dropdown 'kernel' menu. The name in the list is not AMIRIS but something like conda-env-.conda-AMIRIS-py
5. Repeat 2&3&4 for each Tool in project

But this is not ideal and it would be a lot simpler with your suggestion.

[PekkaSavolainen]

I added two buttons to Settings->Tools widget. The upper one sets all Julia Tools in the current project to use the execution settings that are defined on the same page (Basic/jupyter Console, Julia executable path, Julia env, and kernel. The lower button does the same for all Python tools in the current project.

There's also a new button in the Tool Spec Editor, which does the same for individual Tool Specs. The button is disabled (grayed out) if the current selections match the default settings defined in Settings->Tools.

This is in 0.8-dev branch and it will trickle down to master branch eventually.

[dlr-cjs]

Very cool! Thank you, that is super useful for our case.

[soininen]

After being hit with the same issue, I think the behavior and the "broadcast" buttons are confusing. The fact that the Python interpreter in Settings only affects new Tool specifications is not really what I would expect and the "broadcast" button is difficult to discover. I suggest the following behavior:

• If Tool has an interpreter set explicitly, it will be used no matter what.
• Otherwise, if Settings provides an interpreter, Tool uses it.
• Otherwise, use the interpreter provided by resolve_*() functions.

This way we can drop the "broadcast" buttons in Settings and the Restore defaults button in Tool specification editor as clearing the Interpreter field will do the same.

[PekkaSavolainen]

It's not that simple. Consider, what happens in the following case.

1. I open a project with Tool A using Julia Spec_A with no executable set
2. I go to Settings, change the Julia to <somejulia/bin/julia.exe>, click Ok
3. Spec_A now uses <somejulia/bin/julia.exe>
4. Then I open another project, which has a Tool B with Julia Spec_B with no executable set
5. Which executable does Spec_B use now?

There are two options:

1. If it uses 'no executable' (i.e. the one in path), then there's a conflict with the one in Settings->Tools. If a user now executes Tool B, then opens the Settings window and clicks Ok, then executes Tool B again, the first and second executions will happen on a different Julia. Not very user friendly IMO.
2. If it uses <somejulia/bin/julia.exe>, then toolbox has overwritten the Spec_B executable. This means that Tool Specs cannot be saved with the 'no interpreter set' option anymore, because this will always be overwritten if Settings provides an interpreter.

TLDR:
The 'broadcast' buttons are there because the options in Settings->Tools are global settings. If we want to make them as project level settings then that brings another set of problems with consistency and discoverability.

[soininen]

Which executable does Spec_B use now?

B would use the Julia in Settings, i.e. same Julia a A <somejulia/bin/julia.exe>

If it uses <somejulia/bin/julia.exe>, then toolbox has overwritten the Spec_B executable. This means that Tool Specs cannot be saved with the 'no interpreter set' option anymore, because this will always be overwritten if Settings provides an interpreter.

But that is what global settings are supposed to do. If you want to lock B's interpreter, then you set in the specification. You use 'no interpreter set' to say "I am happy with whatever is in Settings."

[PekkaSavolainen]

I guess that makes sense. It's possible that I was overthinking the issue.

[soininen]

There could be a way to set the interpreter/executable/shell for multiple specifications at once but I think it should be separate from the Settings window.

[dlr-cjs]

I would argue differently. Especially for Python, a each project is likely to have its own interpreter path. Therefore, I think if you really use spine-toolbox in mutliple projects, these python and julia defaults should be stored and controllable on a "per project" basis. I, for myelf only have one toolbox project,so I don't mind personally, but other users might benefit from such a "hierarchy":

general default <-> project default <-> tool-specific settings

[soininen]

Especially for Python, a each project is likely to have its own interpreter path.

Agreed. We should have Project-specific settings, too.

[jkiviluo]

In a perfect world, each tool would know what they need (requirements file). We could then check if the environment satisfies the requirements - if not, we could offer to setup a new clean environment and install requirements. Or optionally upgrade current environment, but there one needs to be careful, because that environment might be used for other things too.

We also want all of this hard installation part to be on the shoulders of workflow developers - not workflow users. So, this relates: #2554

As you can see, I'm bit vary of global and project defaults - I don't want to say they are all bad, but I'm afraid they may lead to trouble down the road (e.g. you change your global environment and things break).

[soininen]

I am happy with how the interpreter/executable global settings work.