Details
-
Type:
Feature Request
-
Status: Closed (View Workflow)
-
Priority:
Major
-
Resolution: Done
-
Affects Version/s: None
-
Fix Version/s: 3.0.0.Beta1
-
Component/s: Core
-
Labels:None
-
Estimated Difficulty:Low
Description
There appears to be some confusion about what the traversal path means. We've come up with some better terms that should help users understand how to use it.
When one of the exceptions in the stack trace is being handled, the handlers for that exception are notified in a particular order. That order is based on the type hierarchy of the exception. The traversal seeks to address this scenario:
Assume that a group of exceptions have a common super class. You may choose to write a handler that catches all of those exceptions by handling that super type. We'll call that a category handler, where the super class represents a category (i.e., SQLException) So the question you have to ask yourself, then, is:
Do you want your category handler to be notified before or after the handler for the subtype gets notified?
You specify you want the category handler to be notified before by adding the attribute during = TraversalPath.DESCENDING to the @Handles exception.
However, this descending/ascending isn't catching on. Tree traversal is more commonly referred to using the terms breadth-first or depth-first. Catch notifies breadth-first handlers before depth-first handlers, in the order of the tree traversal (walking the exception type hierarchy).
Therefore, these attribute values would make more sense:
during = TraversalMode.BREADTH_FIRST
during = TraversalMode.DEPTH_FIRST (default)
If you write a SQLException handler with during = TraversalMode.BREADTH_FIRST, then it will be notified before a handler for BatchUpdateException.
Gliffy Diagrams
Issue Links
- incorporates
-
SOLDER-171
Descending traversal path not working as expected
-
- Closed
-
- relates to
-
SOLDER-180
untangle processing of exceptions in stack
-
- Closed
-
Activity
- All
- Comments
- Work Log
- History
- Activity
- Links Hierarchy
I realize my proposal will change how things are done (when considering both SEAMCATCH-31 and SEAMCATCH-32).
What I would like is for each exception in the chain to be handled in turn. We see if any handlers are interested in the inner exception (the cause). If we get to the end of processing that exception and no handler marked it as handled, then we need to move to the wrapper, and so on.
When handling a single exception in the stack, we then notify the handlers in a particular order. That order works breadth-first for handlers marked as such, then depth-first, which is the default phase (or mode) in which handlers get registered.
One way to think of the handler notification order is to think of them as JSF phase listeners, where the "before" is a super-type handler marked TraversalMode.BREADTH_FIRST and the "after" is a super-type handler marked TraversalMode.BREADTH_FIRST. Obviously, the actual event is the invocation of the handler for the specific type (which may not exist, btw).
Dan Allen
added a comment - - edited I realize my proposal will change how things are done (when considering both SEAMCATCH-31 and SEAMCATCH-32 ).
What I would like is for each exception in the chain to be handled in turn. We see if any handlers are interested in the inner exception (the cause). If we get to the end of processing that exception and no handler marked it as handled, then we need to move to the wrapper, and so on.
When handling a single exception in the stack, we then notify the handlers in a particular order. That order works breadth-first for handlers marked as such, then depth-first, which is the default phase (or mode) in which handlers get registered.
One way to think of the handler notification order is to think of them as JSF phase listeners, where the "before" is a super-type handler marked TraversalMode.BREADTH_FIRST and the "after" is a super-type handler marked TraversalMode.BREADTH_FIRST. Obviously, the actual event is the invocation of the handler for the specific type (which may not exist, btw). Everyone good with this proposed change?
Jason Porter
added a comment - Everyone good with this proposed change? 
)I'm fine with it.
Just wanted to point out that with these changes we are dropping the "2-way pass" of the exception chain so that everybody is aware. We can probably live without that.
Jozef Hartinger
added a comment - I'm fine with it.
Just wanted to point out that with these changes we are dropping the "2-way pass" of the exception chain so that everybody is aware. We can probably live without that. Yep. This is also consistent w/ the behavior in Seam 2 (though Catch is far more thorough).
Implemented change
Jason Porter
added a comment - Implemented change 
IMHO these are two different things.
Currently, the ASCENDING/DESCENDING setup controls the order of handlers based on the exception chain. According to the docs there are two phases during exception handling (http://docs.jboss.org/seam/3/catch/latest/reference/en-US/html/client_usage.html#client_usage.traversal)
DESCENDING - where the chain is being traversed from the outer most exception to the root cause
ASCENDING - where the same chain is being traversed from the root cause to the outer most exception
for every exception in the chain, handlers for supertypes of the exception are called (depth first)
IMHO your proposal changes the semantics of traversal mode. It removes the chain traversing completely (it is traversed only once) and it uses the traversal mode to control the order in which handlers for different types of an exception type tree are called.
To ilustrate:
3SS
3S
3 --> 2 --> 1
3SS is a superclass of 3S
3S is a superclass of 3
3 is caused by 2
2 is caused by 1
So far the traversal mode controlled the horizontal axis. I think that what you propose modifies it to control the vertical axis.