Hi Paulex,
2006/5/25, Yang Paulex <[EMAIL PROTECTED]>:
Mikhail,
Sorry to response so late, seems both IBM's smtp server and gmail is not
stable so that I have to send this mail several times from today before
yesterday (please pardon me if the mailing list get several duplicate mails
after hundreds of hours! ).
Basically, I'm afraid I cannot agree with you on this patch.
1. I have no strong objection on the patch for HttpURLConnection, but...If I
understand correctly, Harmony-482 is about the internal contract between
java.net.HttpURLConnection.setRequestMethod() and
o.a.h.l.internal.net.www.protocol.http.HttpURLConnection.getOutputStream(),
Not exactly. The second problem is that the code makes assumption that
two String constant-pool constants in two different classes will be the same
object. That is an assumption about VM that is met in J9 and RI.
Other VMs might not have this kind of optimization and the code will not work
on those VMs. I've fixed the code to not depend on that assumption.
Then I had to develop a test that would
1. passes on RI
2. fails on unfixed Harmony
3. passes on fixed Harmony
The code you suggested below passes on unfixed Harmony, so
it is not a good regression test.
and the internal contract may be broken only if some of the codes refactored
(no chance for user application), say, modification for the
setRequestMethod() like this:
- this.method = methodTokens[i];
+ this.method = method;
But this can be easily monitored by test codes below:
public void testGetOutputStream() throws Exception {
HttpURLConnection c = (HttpURLConnection) new
URL("http://127.0.0.1:";<http://127.0.0.1/>
+ port).openConnection();
c.setDoOutput(true);
c.setRequestMethod(new String("GET"));
c.getOutputStream();
...//other for POST/PUT
}
This test is enough to enforced the internal contract, it will shout if some
refactory breaks the internal contract. So IMHO, I think the patch for
HttpURLConnection is a little over-designed.
2. Even the patch is necessary, I think the introduction of injected
java.net.HttpURLConnectionAccessor is debatable. I suggest to avoid the
injected helper class as long as possible, because it introduces complexity
to understand/manage and is sometimes fragile because it may highly coupled
with implementation details. For this specific case, a mocked subclass of
o.a.h....HttpURLConnection with overridden getRequestMethod can be used
instead.
I partially agree. If you suggest a test that would be easier to understand,
I'll be happy to replace the current one.
Thanks,
Mikhail
Your comments?
2006/5/23, Mikhail Loenko <[EMAIL PROTECTED]>:
>
> I've created regression test and patch for
> http://issues.apache.org/jira/browse/HARMONY-482
>
> I had to make some changes in the luni's build.xml,
> that are intended to be fixed once we all agree with
> proposed test suite layout.
>
> Please review the changes.
>
> Thanks,
> Mikhail
>
> ---------------------------------------------------------------------
> Terms of use : http://incubator.apache.org/harmony/mailing.html
> To unsubscribe, e-mail: [EMAIL PROTECTED]
> For additional commands, e-mail: [EMAIL PROTECTED]
>
>
--
Paulex Yang
China Software Development Labotary
IBM
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
