Issue

I'm in the process of setting up a Publishing workflow where some instances will spawn "dependent publishes" if they are set to publish in the current run, e.g. the user did not disable them in the Pyblish UI. More info on what I'm trying to achieve can be found in this Avalon + USD + Pyblish discussion.

As such, after the instances are collected and the user made his pick as to what to publish then when the user clicks validate or publish the dependent instances get created. However, currently these will not themselves have their relevant plug-ins loaded for their family. This is currently due to how after Collecting the irrelevant Plug-ins are removed. For example this second Validator will never trigger:

class CreateDependencies(pyblish.api.ContextPlugin):
    order = pyblish.api.ValidatorOrder - 0.4

    def process(self, context):
        instance = context.create_instance("dependency")
        instance.data["family"] = "dependency"


class ValidateDependencies(pyblish.api.InstancePlugin):
    order = pyblish.api.ValidatorOrder
    families = ["dependency"]

    def process(self, instance):
        # Never runs...
        self.log.info("Running dependency")

This means I cannot add new instances or alter the instances after the User's publishing choices are well defined. Because I can't add it in-between Collecting and Validating.

Changing CVEI?

In the discussion I opted for maybe adding an additional step to CVEI, specifically:

1. Initialize - find and create the Instances for the user to interact with, only with essential information to allow the user to check the contents.
2. Collect - collect any additional data needed for validation, extraction and integration. Including the potential adding of automated dependent instances.
3. Validate - validate all the existing instances
4. Extract - output the publish instances.
5. Integrate - push the extracted data into the correct location on disk and into your production database.

A user would then be able to "change what instances to trigger" (e.g. through Pyblish QML) at the end of Initialize (1). Then on Validate or Publish the publishing will continue onwards through respectively 2-3 or 2-5. The difference in behavior would then be after Collect (2) that all plugins are still checked whether they should be run for the potentially new or changes families.

I would love to see a way where I can trivially add instances as dependencies after the user initiated the publish choices and still allow them to be shown individually to the user as a result. This would then also allow to debug those dependent publishes right in the Pyblish User Interfaces.

This is somewhat related to #346 due to the allowing the Initialize step to become solely the "list which instances are available to publish" and all the post collecting of data to be done in then step two: Collect.