commons-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eduard Papa (Commented) (JIRA)" <>
Subject [jira] [Commented] (DIGESTER-161) Document thread-safety in javadoc of Rule class
Date Mon, 12 Dec 2011 21:45:30 GMT


Eduard Papa commented on DIGESTER-161:

The code I was referring to was actually using digester 2.x, so I don't think the provider
method is there. It was just creating digester and adding all the rules. The rule that was
causing the problem was a static final...hence the thread-safety issue.

I would have liked someone who knows more about Digester to update the javadoc but I'll give
a try....tomorrow hopefully. 
> Document thread-safety in javadoc of Rule class 
> ------------------------------------------------
>                 Key: DIGESTER-161
>                 URL:
>             Project: Commons Digester
>          Issue Type: Improvement
>    Affects Versions: 3.1
>            Reporter: Eduard Papa
>            Priority: Trivial
>              Labels: rule, thread-safe
>   Original Estimate: 1h
>  Remaining Estimate: 1h
> I discovered a problem today with some code that was reusing a custom Rule in multiple
threads, even though each thread was creating its own digester. It seems that Digester.addRule
is calling rule.setDigester and if the rule is shared across multiple threads, the calls to
begin/end can get tangled across threads. 
> It is obvious that Rules are not meant to be shared, but the javadoc <>
seems to be implying the opposite and is confusing at best. It talks about the rules being
stateless, even though the framework itself is changing its state with rule.setDigester(digester).
It further states that since all state is part of the digester, the rule is safe under all
cases, which is very misleading.
> " ... Rule objects should be stateless, ie they should not update any instance member
during the parsing process. A rule instance that changes state will encounter problems if
invoked in a "nested" manner; this can happen if the same instance is added to digester multiple
times or if a wildcard pattern is used which can match both an element and a child of the
same element. The digester object stack and named stacks should be used to store any state
that a rule requires, making the rule class safe under all possible uses. ..."
> I think the statement above should be reworded to be more correct and avoid confusion.
Down the line, maybe the digester accessed by the rule should be a ThreadLocal.

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:!default.jspa
For more information on JIRA, see:


View raw message