harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Nathan Beyer" <ndbe...@apache.org>
Subject Re: [jira][swing] Could someone please commit HARMONY-3454 ?
Date Tue, 10 Apr 2007 20:10:48 GMT
On 4/10/07, Zakharov, Vasily M <vasily.m.zakharov@intel.com> 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.


>  Vasily
> -----Original Message-----
> From: nbeyer@gmail.com [mailto:nbeyer@gmail.com] On Behalf Of Nathan
> Beyer
> Sent: Tuesday, April 10, 2007 8:42 AM
> To: dev@harmony.apache.org
> 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 <vasily.m.zakharov@intel.com> 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
> >

View raw message