graph: edge style

[oliver-sanders]

In the present GUIs there are two different edge styles:

• Solid arrows for and type dependencies (e.g. foo => baz or foo & bar => pub)
• Empty arrows for or type dependencies (e.g. foo | bar => pub)

Some ideas to consider for edge style in the new GUI:

• Dependency type (e.g. and or or)
• Qualifier (e.g. :succeed or :fail)
• Satisfaction (see bellow)

In this GUI we would also like the edge style to make it clear if a task has been manually triggered before the completion of its upstream dependencies (see also cylc/cylc-flow#3223 (comment)). Options:

1. Make edge style change depending on whether a dependency is satisfied or not
   • + simple, can be done client side with little to no change to Cylc Flow
   • - doesn't leave a permanent record that the dependency was ignored
2. If a task is manually triggered and some of its upstream dependencies are not met change their edge style to reflect this
   • + leaves a permanent record that the dependency was ignored
   • - more involved change, this state will need to be held at the workflow
3. Some hybrid of the above?

[oliver-sanders]

Design ideas please...

[matthewrmshin]

What are the choices? I suppose we have:

• Solid/empty arrow heads?
• Arrow head style?
• Solid/dotted lines?
• Colours?

On qualifier. Can we use colours for the edge to corresponding task state that provides the qualifier? (Obviously this will not work for custom qualifiers/outputs.) If there are multiple downstream paths, I guess we should use dotted lines?

Do we need to consider clock and xtriggers? (Or in another issue?)

[oliver-sanders]

What are the choices?

• We could also use text or symbols.
• An extreme option would be to represent conditional triggers as an intermediate node (say a diamond symbol like an if in a workflow diagram).

Do we need to consider clock and xtriggers? (Or in another issue?)

At present xtriggers triggers are special graph nodes so the edge style isn't effected. This could change though...

[oliver-sanders]

Some ideas:

1. Edges use a grey colour and turn darker when the dependency is satisfied.
2. If a task is manually triggered before it's dependencies are satisfied the arrow heads of the "ignored" dependencies turn to crosses.
3. Edges representing dependencies other than succeeded should have the qualifier (e.g. :failed) written next to the arrow.

[hjoliver]

At present xtriggers triggers are special graph nodes so the edge style isn't effected.

My thinking was: xtriggers (including clock triggers) are just other entities that tasks can depend on, so they should be represented by other nodes in the graph, with dependence represented by edges (just as for dependence on other tasks).

The complication with xtriggers is how to show their uniqueness (in terms of unique/shared calls to the trigger function). Unlike tasks they don't really have "instances" at each cycle point (although they may or may not be depend on the cycle point of their dependent tasks) ... so I took the easy way out and put in a separate xtrigger node for every dependent task (even if in reality multiple xtrigger nodes might actually be "the same xtrigger"). In practice I think that works quite well though.

This could change though...

I can still see "the method in my madness" above, but if you have other ideas about how to represent xtriggers, plz explain and we can consider them...

[oliver-sanders]

if you have other ideas about how to represent xtriggers

Nope, I'm happy to keep them this way!

BTW there is another issue for xtriggers #131