(Issue title and text were AI-generated from the mentioned PR)
The libparallel scheduler currently catches all exceptions raised during task execution and treats them as tolerated task failures. This is too broad: programming errors and other unexpected exceptions can be silently swallowed, making failures harder to diagnose and potentially allowing execution to continue in an invalid state.
The scheduler should only catch exceptions that represent expected HADDOCK task-execution failures, such as CNS execution errors. Unexpected exceptions should propagate normally so they remain visible and fail fast.
PR #1654 addresses this by introducing a dedicated HaddockTaskExecutionError base exception, making CNSRunningError inherit from it, and restricting the scheduler's exception handler to that class. It also adds tests confirming that expected task failures remain tolerated while unrelated exceptions such as ValueError propagate.
(Issue title and text were AI-generated from the mentioned PR)
The
libparallelscheduler currently catches all exceptions raised during task execution and treats them as tolerated task failures. This is too broad: programming errors and other unexpected exceptions can be silently swallowed, making failures harder to diagnose and potentially allowing execution to continue in an invalid state.The scheduler should only catch exceptions that represent expected HADDOCK task-execution failures, such as CNS execution errors. Unexpected exceptions should propagate normally so they remain visible and fail fast.
PR #1654 addresses this by introducing a dedicated
HaddockTaskExecutionErrorbase exception, makingCNSRunningErrorinherit from it, and restricting the scheduler's exception handler to that class. It also adds tests confirming that expected task failures remain tolerated while unrelated exceptions such asValueErrorpropagate.