Name: mgC56079 Date: 11/01/98



The notion of default cursor should be clarified.
In particular, the spec for Component.getCursor() should clearly state
whether the method should return null or the Cursor object of a particular type
when invoked on a newly-created component.

The JDK 1.1.7 returns a Cursor object of a DEFAULT_CURSOR type.
The JDK 1.1.8 is proposed to return null in this case (see 4111379)

This change will cause JCK test failure:
JCK1.1a-JCK1.1.1a:
api/java_awt/Comonent/Assertion.html#Component2026

JCK1.1.2a-JCK1.2a:
api/java_awt/Comonent/manual.html#getCursor

See message below for more details:

> From ###@###.### Sat Oct 31 01:44:09 1998
> X-Sun-rcpt-to: ###@###.###
> Date: Fri, 30 Oct 1998 14:31:30 -0800 (PST)
> From: "Kevin A. Smith" <###@###.###>
> Subject: Re: Fw: [Fwd: getCursor tests]
> To: "Mikhail A. Gorshenev" <###@###.###>
> Cc: ###@###.###, ###@###.###, ###@###.###
> MIME-Version: 1.0
>
> Misha,
>
> I agree that this is a change in the behavior of this particular method. I
> also agree that we should exclude the test because the API specification is
> vague enough to allow such a change. Please write a bug report against the
> specification for Component.getCursor(). Please exclude the test because of
> this specification problem in JCK 1.1.x and JCK 1.2.
>
> Thank-you,
> Kevin
>
>
> >----- Begin Included Message -----<
>
> Date: Fri, 30 Oct 1998 18:43:16 +0300 (MSK)
> From: "Mikhail A. Gorshenev" <###@###.###>
> Subject: Re: Fw: [Fwd: getCursor tests]
> To: martin.lister@Eng, Kevin.A.Smith@Eng
> Cc: ###@###.###, ###@###.###
>
> Martin, Kevin,
>
> Yes, we consider this an API change as the
> new implementation contradicts the existing specification
> and causes the existing JCK test failure.
>
> > The change to the initial value of the (non-public) cursor field in a
> > Component did not fall into the definition of an API change, according
> > to guidance that we have been given.
> The changing of non-public field is indeed not an API change.
> However, there is a public method Component.getCursor() which returns this
> value.
>
> In fact this change may break applications which assume
> that Component.getCursor() method always returns a non-null
> Cursor object.
>
> We have discussed the problem with St-Petersburg JDK-AWT team
> and agreed that the fix for 4111379 is a reasonable one as it
> makes the semantics of default cursor consistent across platforms.
> We also understand that changing the behaviour of Component.getCursor()
> is the only acceptable way to fix the 4111379 problem.
>
> >From our point of view, the best way to resolve the problem is to
> apply this fix to JDK 1.2 and document this change as known incompatibility.
> However, we understand that it may be to late to do this for JDK 1.2
> (as far as we see, this fix is not supposed to go into JDK 1.2).
>
> Another way to resolve this is to treat the javadoc of jdk1.1.7
> for Component.getCursor, Component.setCursor and Cursor class itself
> as incorrect (it doesn't define the notion of default cursor precisely enough).
> This will allow us to exclude the test from JCK1.1.x. However, we are not sure
> whether this way of handling an incompatibility is acceptable for JCK.
>
> Kevin, could you please clarify?
>
> thanks,
> Misha
>
> > From ###@###.### Thu Oct 29 21:45:54 1998
> > X-Sun-rcpt-to: ###@###.###
> > From: Martin Lister <###@###.###>
> > Date: Thu Oct 29 10:27:41 1998
> > To: ###@###.###
> > Subject: Fw: [Fwd: getCursor tests]
> > Cc: ###@###.###
> > Mime-Version: 1.0
> > Content-Transfer-Encoding: 7bit
> >
> > Misha,
> >
> > Do we class this as an API change? I assume we class the described behaviour as
> > part of the API specification other wise the test would not exist. If so we will
> > need to discuss the validty of the fix with the AWT team.
> >
> > Martin
> >
> > -----Begin Included Message -----
> >
> > Date: Thu, 29 Oct 1998 10:30:40 -0800
> > From: NEIL RICHARDS <neilrich@Eng>
> > Subject: [Fwd: getCursor tests]
> > To: martin.lister@Eng
> > Cc: stuartl@Eng
> >
> >
> >
> > Hi Martin,
> > This change has had a detailed (code) review by Fred Ecks from the awt
> > team, & the design of the change has been reviewed & discussed with the
> > members of the awt-core distribution list.
> > The change to the initial value of the (non-public) cursor field in a
> > Component did not fall into the definition of an API change, according
> > to guidance that we have been given.
> > Maybe you could clarify what you consider an API change to be?
> > Cheers,
> > Neil Richards
> >
> > stuartl wrote:
> > >
> > > > From: Martin Lister <###@###.###>
> > > > Date: Thu Oct 29 09:24:58 1998
> > > > To: stuartl@netscapesrv
> > > > Subject: Re: getCursor tests
> > > > Mime-Version: 1.0
> > > > Content-Transfer-Encoding: 7bit
> > > >
> > > > Stuart,
> > > >
> > > > Has this fix been evaluated by development?
> > > >
> > > > The test assumes the cursor of newly-created component
> > > > (Canvas in this case) is always set to DEFAULT_CURSOR.
> > > >
> > > > The jdk1.1.7 spec for Component.getCursor/setCursor says nothing about
> > > > initial cursor value.
> > > >
> > > > However, the spec for Cursor.DEFAULT_CURSOR says
> > > > /**
> > > > * The default cursor type (gets set if no cursor is defined).
> > > > */
> > > > public static final int DEFAULT_CURSOR = 0;
> > > >
> > > > So this should mean the initial cursor value for any
> > > > component for which cursor is not defined is a Cursor object of
> > > > DEFAULT_CURSOR type.
> > > >
> > > > The test correctly verifies whether this condition is true.
> > > >
> > > > So the proposed fix for 4111379 is an incompatible API change. Surely we
> > have
> > > > not changed the API in 1.1.8?
> > > >
> > > > Martin
> > > >
> > > > > Martin,
> > > > > In the implementation of cursors in 1.1.8, the initial cursor for a Canvas
> > > > > (and
> > > > > many other Components) is null.
> > > > > This fix requires exclusion of JCK getCursorTests ("Component2026"), which
> > > > > includes the lines:
> > > > >
> > > > > Canvas c=new Canvas(); //step Create some Component
> > > > > Cursor cs;
> > > > > cs = c.getCursor(); //step Get the initial Cursor
> > > > >
> > > > > In the new implementation, this throws a NullPointerException.
> > > > > See bug 4111379 if you need more details.
> > > > >
> > > > > Thanks a lot,
> > > > > Stuart
> > > > >
> > > > > ---------------------------
> > > > > Stuart Lawrence (x51533)
> > > > > Tel: 408 343-1533
> > > > > ###@###.###
> > > > >
> > > >
> > > >
> > > >
> > > >
> > > > Martin Lister
> > > > Java Technology Licensee Services
> > > > 901 San Antonio Road MS UCUP02-202
> > > > Palo Alto, CA 94303-4900
> > > > TEL : 408-863-3544 FAX : 408-343-1815
> > > > EMAIL: ###@###.###
> > > >
> > > >
> > >
> > > ---------------------------
> > > Stuart Lawrence (x51533)
> > > Tel: 408 343-1533
> > > ###@###.###
> >
> >
> > ---- End Included Message ----
> >
> > Martin Lister
> > Java Technology Licensee Services
> > 901 San Antonio Road MS UCUP02-202
> > Palo Alto, CA 94303-4900
> > TEL : 408-863-3544 FAX : 408-343-1815
> > EMAIL: ###@###.###
> >
> >
>
> >----- End Included Message -----<
>


======================================================================

Name: saC57035 Date: 02/08/99

Another JCK1.2a test that fails because of the 4111379 bug fix:
api/java_awt/serialization/manual.html#Component

======================================================================