Task-value references ({{tasks.P.values.*}}) not added as dependencies → job creation rejected

[matthewmoorcroft]

Summary

File: src/flowx/bundler/dab_writer.py (new _backfill_task_value_dependencies, wired into _build_job_resource)

flowx rewrites ADF @variables('X') / @activity('Y').output into {{tasks.P.values.*}} task-value references (synthesised _init_<var> tasks and IfCondition/Switch/ForEach bridge tasks), but does not always add the matching depends_on edge. Databricks requires the producer task to be a (transitive) dependency of any task referencing its value.

Observed failure (deploy-time)

INVALID_PARAMETER_VALUE: Task 'P' must be a dependency of task 'T'

Impact

Job creation is rejected at deploy time whenever a bridge/init task references a task value without an explicit ordering edge. Blocks deployment of pipelines that use ADF variables or cross-activity output references.

Proposed fix

After _strip_dangling_task_value_refs, scan each task's condition_task.{left,right} and notebook_task.base_parameters for {{tasks.P.values.*}} refs; for each referenced producer P that exists in this job's task list but isn't already a direct dependency of the consumer, append {"task_key": P} to its depends_on. Operates within a single job scope; recurses into for_each_task.task bodies against their own sibling set.

Note

Intra-job / intra-pipeline dependency correctness — independent of the cross-pipeline lineage work in #23/#24.

Test gap

No test asserts the task-value ⟹ dependency invariant. Fix PR adds coverage.