I feel that the definition of the behavior of the Error intermediate Event is unclear in the BPMN Adopted Specification. We implemented its behavior to closely model the <throw>/<catch> in BPEL, but upon discussion with one of our customers and then with Steve it appears that our interpretation of the spec is not what was intended by its authors.
Error intermediate Events are discussed with any detail in the following places in the spec:
Table 9.6 (p. 41) describes the End event and says - "This type of End indicates that a named Error should be generated. This Error will be caught by an Intermediate Event within the Event Context."
Table 9.8 (p. 45) describes the Error intermediate event - "This is used for error handling--both to set (throw) and to react to (catch) errors. It sets (throws) an error if the Event is part of a Normal Flow. It reacts to (catches) a named error, or to any error if a name is not specified, when attached to the boundary of an activity."
Table 9.9 (p. 46, 47) has Error intermediate event attributes described. The behavior described there relates to the ErrorCode.
Text on p 59-60 describes its role in modeling a Hazard in a Business Transaction.
Text on p 76 mentions it can be a target of an Event-based Gateway, although on p. 78 under "Changes since the 1.0 draft version" it says that Error was removed as a possible target. I believe the text on p
76 is an editing error.
Text on page 130-131 describes the "Event Context", mentioned in table 9.6. Text on that page in the last paragraph explicitly compares Error to the BPEL4WS fault, and goes on in that paragraph to describe the role of the ErrorCode.
Table 11.5 (p. 141) mentions that Error end events map to BPEL throw elements, again tying it to the notion of faults.
Table 11.10 (p. 145) describes the mapping to BPEL for intermediate Error event, which proposes that a boundary event be mapped to a <catch> within a <scope> that encompasses the activity.
From this somewhat limited description, we chose to implement Error as a strictly hierarchical faulting mechanism, as it is in BPEL as well as most programming languages, and which seems a direct consequence of the mapping in Table 11.10. However, in subsequent discussion I have learned it was the intention of the authors that Errors be "subscribed" to in a global scope, and that they will be responded to from any activie location in the process, but that the highest "parent" activity in a subprocess hierarchy will supersede (and terminate) any lower-level Error intermediate events. (This is my interpretation of an email thread with Steve on this subject that is not crisp enough to include verbatim in this issue. I may have interpreted it wrongly, however.)
If the intention of BPMN is to allow Error to be used as a parallel- thread signaling mechanism (supporting a "first one done wins" pattern, for example) as well as a hierarchical faulting mechanism (where the scope in which a thrown Error can be caught is limited to the hierarchy of parents of the activity that causes it, including the activity itself) then the behavior of a number of ambiguous topologies need to be clarified in the spec. Also, I believe the mapping to BPEL is incorrect for the signaling use, and that for some configurations a correct mapping may not exist.
In any event, clarifying the scope of the Event Context with respect to the behavior of the Error event seems necessary.
My proposed resolution will be for the text to make it clear that Error can only be caught by the activity that causes it or one of its parents. There are several other alternatives to use for signaling between parallel sequence flows, including our suggestion to our customers that the Rule event be used.