[classlib][net] issue H-1879 needs committer's attention

[Alexei Fedotov <al...@gmail.com>]

Alexey (Petrenko),

I noticed that you assigned the following issue to yourself:

http://issues.apache.org/jira/browse/HARMONY-1879 [drlvm][unit]
org.apache.harmony.tests.internal.net.www.protocol.http.HttpURLConnectionTest
fails

Thank you for taking a care for this important issue. Due to
synchronization bug the test regularly fails on multi-processor
computers and prevents all class library tests to pass.

Let me attract your attention to the fact that Denis submitted a patch
for this issue which to my mind makes a step into the right direction.
Could you please review if the patch worth to be committed?

Thank you in advance, Alexei

On 11/25/06, Alexei Fedotov <al...@gmail.com> wrote:
> Alexey,
>
> I saw that you had assigned the bug
> http://issues.apache.org/jira/browse/HARMONY-1879 to yourself. What do
> you think of the following fix?
>    http://issues.apache.org/jira/secure/attachment/12345485/HARMONY-1879-HttpURLConnectionTest(3).patch
>
> Does it worth to be committed as is? I've got one more idea of what we
> can do about this issue, but it complicates things, and this fix is
> enough to make things work.
>
> All,
> Your comments about the patch are welcome.
>
> With best regards, Alexei
>
> On 11/24/06, Alexei Fedotov <al...@gmail.com> wrote:
> > It was mine request? Oh, guy, it seems I'm getting old... :-)
> >
> > On 11/23/06, Alexey Petrenko <al...@gmail.com> wrote:
> > > 2006/11/23, Alexei Fedotov <al...@gmail.com>:
> > > > Gregory, this is a good question. Let me tell this long story.
> > > >
> > > > Jimmy tried to run the test using J9 VM and the test passed. I beleive
> > > > Alexey Petrenko considered that fact as a justification to move the
> > > > bug to [drlvm] category. Alexey, is my understanding correct?
> > > No. I've changed it by your request.
> > >
> > > SY, Alexey
> > >
> > > > Why the test passed on J9? Denis found out that the issue with test is
> > > > actually server synchronization issue. It is not a surprise that our
> > > > VMs implement threading differently.
> > > >
> > > > Then Denis proposed a fix which made a test less dependent from
> > > > synchronization.
> > > >
> > > >
> > > >
> > > > On 11/22/06, Gregory Shimansky <gs...@gmail.com> wrote:
> > > > > Denis Kishenko wrote:
> > > > > > Dear committers, please take a look on H-1879
> > > > >
> > > > > Is there a reason why this bug has drlvm category, and has [drlvm]
> > > > > prefix? I see from the history of this bug that it was transferred back
> > > > > and forth. The attached patches change only classlib test...
> > > > >
> > > > > > 2006/11/22, Fedotov, Alexei A <al...@intel.com>:
> > > > > >> +1
> > > > > >>
> > > > > >>
> > > > > >> >-----Original Message-----
> > > > > >> >From: Denis Kishenko [mailto:dkishenko@gmail.com]
> > > > > >> >Sent: Wednesday, November 22, 2006 5:40 PM
> > > > > >> >To: dev@harmony.apache.org
> > > > > >> >Subject: Re: [classlib][net] issue H-1879 HttpURLConnectionTest
> > > > > >> >
> > > > > >> >Alexei,
> > > > > >> >
> > > > > >> >Synchronization of accept() looks like test improvement, so I will
> > > > > >> >attach new patch to fix H-1879 and we will try to find solution for
> > > > > >> >accept synchronization.
> > > > > >> >
> > > > > >> >2006/11/22, Fedotov, Alexei A <al...@intel.com>:
> > > > > >> >> Ok, I see. Makes sense for me. I cannot imagine anything better at
> > > > > >> this
> > > > > >> >> moment, and if this works, that's great.
> > > > > >> >>
> > > > > >> >> There is still a problem with accept() call which is not
> > > > > >> synchronized,
> > > > > >> >> but I don't see how we can fix this correctly. accept() call is
> > > > > >> >> synchronized using related ServerSocket object, but this doesn't help
> > > > > >> us
> > > > > >> >> - this blocking doesn't release an object lock.
> > > > > >> >>
> > > > > >> >> With best regards,
> > > > > >> >> Alexei Fedotov,
> > > > > >> >> Intel Java & XML Engineering
> > > > > >> >>
> > > > > >> >> >-----Original Message-----
> > > > > >> >> >From: Denis Kishenko [mailto:dkishenko@gmail.com]
> > > > > >> >> >Sent: Tuesday, November 21, 2006 8:06 PM
> > > > > >> >> >To: dev@harmony.apache.org
> > > > > >> >> >Subject: Re: [classlib][net] issue H-1879 HttpURLConnectionTest
> > > > > >> >> >
> > > > > >> >> >I can suggest add flag to avoid situation described above (see
> > > > > >> attached
> > > > > >> >> >fix).
> > > > > >> >> >
> > > > > >> >> >2006/11/21, Denis Kishenko <dk...@gmail.com>:
> > > > > >> >> >> Alexei,
> > > > > >> >> >>
> > > > > >> >> >> Yep, syncronization is necessary but not such way.
> > > > > >> >> >>
> > > > > >> >> >> As you said, test try to control server/proxy starting using
> > > > > >> >> >> bound.wait(5000). It looks like on linux server/proxy thread
> > > > > >> started
> > > > > >> >> >> eallier then on winxp.
> > > > > >> >> >>
> > > > > >> >> >> In other words, what we are waiting for
> > > > > >> >> >> 1. start server thread
> > > > > >> >> >> 2. bound.wait
> > > > > >> >> >> 3. bound.notify
> > > > > >> >> >> 4. start proxy thread
> > > > > >> >> >> 5. bound.wait
> > > > > >> >> >> 6. bound.notify
> > > > > >> >> >> 7. connect
> > > > > >> >> >> but we have on linux
> > > > > >> >> >> 1. start server thread
> > > > > >> >> >> 2. bound.notify
> > > > > >> >> >> 3. bound.wait
> > > > > >> >> >> ---- wait 5 seconds -----
> > > > > >> >> >> 4. start proxy thread
> > > > > >> >> >> 5. bound.notify
> > > > > >> >> >> 6. bound.wait
> > > > > >> >> >> ---- wait 5 seconds -----
> > > > > >> >> >> 7. connect
> > > > > >> >> >>
>


-- 
Thank you,
Alexei