I think I'm convinced of the catch Throwable for now, but let's put some comments in the catch block to document that this is being done for RI compatibility right now and that this could be tweaked in the future.
As for the 'contentTypes' field, I'm fine with either changing it to use a Hashtable or just putting synchronized blocks around the get/put/etc invocations and using the field as the lock. Perhaps a code comment about this would be nice as well. I'll take assignment of the bug. Do you mind uploading a new patch, or do you want me to just tweak what's there? Whatever you prefer is fine with me. -Nathan On 4/10/07, Zakharov, Vasily M <[EMAIL PROTECTED]> wrote:
> I guess if the RI is really capturing everything we can do it, but I'd > prefer to go the way of a non-bug difference here, unless we have some > applications that are depending on this behavior. Sure, we can make a NBD here. The question is, what would our behavior be? Catching just Exception is clearly wrong. If not catching Throwable we should at least catch Exception and LinkageError (with subclasses). Do you agree? Do you think we should file a separate NBD JIRA for this? > How do you know it's using a Hashtable? See the stack in the issue description, it's pretty clear. Also this is discussed in comments in http://issues.apache.org/jira/browse/HARMONY-3453 > In this case, I would suggest that we just protect the data structure > itself, such that only the actual read/write of the Map is protected. > This makes sure that the data structure itself doesn't blow up with a > spurious NPE (or the like) during a read/write race within the Map > itself. This is what a Hashtable would prevent because it makes each > method invocation atomic. Ok, I agree. We can really use Hashtable now as the code is restructured and doesn't need null keys/values. Vasily -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Nathan Beyer Sent: Wednesday, April 11, 2007 12:11 AM To: [email protected] Subject: Re: [jira][swing] Could someone please commit HARMONY-3454 ? On 4/10/07, Zakharov, Vasily M <[EMAIL PROTECTED]> wrote: > > Thanks for checking the patch, Nathan! > > 1.1. The problem is the wrapped fragment performs class loading, and > can easily produce LinkageError, ExceptionInInitializerError, > IllegalAccessError > etc. and all these must not disrupt the code operation. > > 1.2. It is a question of compatibility. RI seems to catch everything > at this point, including ThreadDeath. :( I guess if the RI is really capturing everything we can do it, but I'd prefer to go the way of a non-bug difference here, unless we have some applications that are depending on this behavior. > > 2.1. Adding synchronization is not a problem at all. But: > > 2.2. What is a point to make a tread-safe point in non-thread-safe > package? > If user can't and shouldn't rely on internal thread-safety and must > think > of synchronization himself, why waste performance (probably twice, as > user > provides his own synchronization) to provide thread-safety in one > particular > place? Isn't it a drop in the ocean? The synchronization would be a minimal performance hit, especially considering it's protecting the integrity of the data structure. > > 2.3. The only reason I see for adding the synchronization here is RI > uses > Hashtable that is synchronized, and we probably would like to be > compatible > in this manner. How do you know it's using a Hashtable? If it is, then it backs up my understanding of the thread-safety guarantees of Swing. There aren't any guarantees for code that's expected to be executed exclusively in the UI thread and protected via isolation, but shared data that could be updated or queried in any thread needs to be protected for consistency. In this case, I would suggest that we just protect the data structure itself, such that only the actual read/write of the Map is protected. This makes sure that the data structure itself doesn't blow up with a spurious NPE (or the like) during a read/write race within the Map itself. This is what a Hashtable would prevent because it makes each method invocation atomic. -Nathan > > Vasily > > > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Nathan > Beyer > Sent: Tuesday, April 10, 2007 8:42 AM > To: [email protected] > Subject: Re: [jira][swing] Could someone please commit HARMONY-3454 ? > > I have a few comments about the patch - > > 1. I think the "catch(Throwable)" in JEditorPane should be change, at > a minimum, to "catch(Exception)". We don't want to hide things like > OutOfMemoryErrors, which could happen anywhere. > > 2. There's no concurrency protection for the static Map, contentTypes. > I realize that many parts of Swing and AWT aren't thread safe, but > this doesn't seem like one of them. Perhaps this wasn't properly > protected before, but we should add it now. > > -Nathan > > On 4/9/07, Zakharov, Vasily M <[EMAIL PROTECTED]> wrote: > > Hi, > > > > Could someone please commit > > http://issues.apache.org/jira/browse/HARMONY-3454 ? > > > > It has patch available and it looks like the discussion over it is > over. > > > > Thank you! > > > > Vasily Zakharov > > Intel ESSD > > >
