<AWT Dev> RFR [13] 8227919: 8213232 causes crashes on solaris sparc64

Phil Race philip.race at oracle.com
Wed Jul 31 17:45:01 UTC 2019 

Ok, that looks a lot more sane.

You (Ichiroh) will need to follow the JDK 13 fix request process here :
http://openjdk.java.net/jeps/3#Fix-Request-Process

An example of a recent case is 
https://bugs.openjdk.java.net/browse/JDK-8217676

Also remember you fix this ONLY in JDK 13.
The fix will be forwarded synced by integrators here into JDK 14.

-phil.

On 7/31/19 8:40 AM, Vladimir Kempik wrote:
> Hello
>
> In fact such minimalistic change only affects code generation on solaris sparc64
>
> We have tested such fix on openjdk-11.0.4 with solaris sparc64 and it passed tck just fine.
>
> Vladimir.
>
>> 31 июля 2019 г., в 16:43, Ichiroh Takiguchi <takiguc at linux.vnet.ibm.com> написал(а):
>>
>> Hello Phil.
>>
>> I appreciate your comment.
>>
>> Just in case, I created another fix.
>> It's Vladimir's idea.
>>
>> Bug:    https://bugs.openjdk.java.net/browse/JDK-8227919
>> Change: https://cr.openjdk.java.net/~itakiguchi/8227919/webrev.02/
>>
>> Just added "defined(__linux__)"
>>
>> I think it can minimize the side effect for OpenJDK testing.
>> I'll open another bugid for Linux s390x issue.
>>
>> I'm not sure which is the best or better on this situation...
>>
>> Thanks,
>> Ichiroh Takiguchi
>>
>> On 2019-07-31 05:23, Phil Race wrote:
>>> Hi,
>>> I've stared at this fix and all this bit shifting inside ifdefs is
>>> hard to believe
>>> this is the way you have to use XIM.
>>> Also JDK 13 testing is already well under way and a lot of the testing for Linux
>>> has already been finished. So *you* are going to have to run (1) all regression
>>> tests on Linux & Solaris and that there are no new failures (2) JCK on Linux and
>>> Solaris and certify it all passes else we could have a DOA release or
>>> a release stopper.
>>> I think given that, and my observations about the fix I think that the
>>> safe option for JDK 13 is to completely back out 8213232 and try again
>>> for JDK 14.
>>> But even then I'm first going to want to see pointers to documentation
>>> for XIM which
>>> say why what you are doing is correct.
>>> -phil.
>>> On 7/30/19 10:02 AM, Ichiroh Takiguchi wrote:
>>>> Thanks, Vladimir.
>>>> I think it's urgent issue for JDK13.
>>>> Another reviewer is required.
>>>> And I'd like to obtain a sponsor for this issue.
>>>> Thanks,
>>>> Ichiroh Takiguchi
>>>> On 2019-07-29 21:49, Vladimir Kempik wrote:
>>>>> Looks good to me.
>>>>> Now need some reviewers to take a look at it.
>>>>> Thanks, Vladimir.
>>>>>> 29 июля 2019 г., в 15:44, Ichiroh Takiguchi <TAKIGUC at jp.ibm.com> написал(а):
>>>>>> Hello Vladimir.
>>>>>> I appreciate your suggestion.
>>>>>> And I'm very sorry for bad response.
>>>>>> Our mail server for external mailing list is downed in these days.
>>>>>> So I'd like to use another email address.
>>>>>> Please see following 2 outputs.
>>>>>>> $ XMODIFIERS=@im=local ./xim_root
>>>>>> ......
>>>>>>> ButtonPress:
>>>>>>> [1]w=0x2000002
>>>>>>> [2]w=0x2000002
>>>>>>> [1]state=0x2
>>>>>>> [2]state=0x2
>>>>>>> $ XMODIFIERS=@im=kinput2 ./xim_root
>>>>>> ......
>>>>>>> ButtonPress:
>>>>>>> [1]w=0x200000200000000
>>>>>>> [2]w=0x2000002
>>>>>>> [1]state=0x200000000
>>>>>>> [2]state=0x2
>>>>>> These mean
>>>>>> @im=local is used low 32bit on 64bit long  <= The data comes from Xlib
>>>>>> @im=kinput2 is used high 32bit on 64bit long  <= The data comes from XIM
>>>>>> protocol
>>>>>> So following code changes is required on Linux s390x, it was my mistake.
>>>>>> if (w > 0xffffffffUL) w = w >> 32;
>>>>>> I created another fix, could you review it ?
>>>>>> Bug:    https://bugs.openjdk.java.net/browse/JDK-8227919
>>>>>> Change: https://cr.openjdk.java.net/~itakiguchi/8227919/webrev.01/
>>>>>> It changes:
>>>>>> * 0xffffffffL is changed to 0xffffffffUL
>>>>>> * defined(__linux__) is added into defined(_LP64) line
>>>>>> Thanks,
>>>>>> ------
>>>>>> Ichiro Takiguchi
>>>>>>     IBM Japan Ltd.
>>>>>> "awt-dev" <awt-dev-bounces at openjdk.java.net> wrote on 2019/07/26 21:59:59:
>>>>>>> From: Vladimir Kempik <vkempik at azul.com>
>>>>>>> To: Ichiroh Takiguchi <takiguc at linux.vnet.ibm.com>
>>>>>>> Cc: "awt-dev at openjdk.java.net" <awt-dev at openjdk.java.net>, "i18n-
>>>>>>> dev at openjdk.java.net" <i18n-dev at openjdk.java.net>
>>>>>>> Date: 2019/07/26 22:01
>>>>>>> Subject: [EXTERNAL] Re: <AWT Dev> RFR [13] 8227919: 8213232 causes
>>>>>>> crashes on solaris sparc64
>>>>>>> Sent by: "awt-dev" <awt-dev-bounces at openjdk.java.net>
>>>>>>> Hello
>>>>>>> what still bothers me is this:
>>>>>>> The first Big Endian block:
>>>>>>> 1629 #if defined(_LP64) && !defined(_LITTLE_ENDIAN)
>>>>>>> 1630         // On 64bit BigEndian,
>>>>>>> 1631         // Window value may be stored on high 32bit by
>>>>>>> XGetICValues via XIM
>>>>>>> 1632         if (w > 0xffffffffL) w = w >> 32;
>>>>>>> 1633 #endif
>>>>>>> is inside linux&macos block
>>>>>>> 1622 #if defined(__linux__) || defined(MACOSX)
>>>>>>> however the second Big Endian block is for every platform:
>>>>>>> 1698 #if defined(_LP64) && !defined(_LITTLE_ENDIAN)
>>>>>>> 1699     // On 64bit BigEndian,
>>>>>>> 1700     // XIMPreeditState value may be stored on high 32bit by
>>>>>>> XGetICValues via XIM
>>>>>>> 1701     if (state > 0xffffffffL) state = state >> 32;
>>>>>>> 1702 #endif
>>>>>>> I?m not sure if it?s ok or not.
>>>>>>> Also, since XIMPreeditState is unsigned 64-bit int, I think we need
>>>>>>> to compare it against 0xffffffffUL instead of 0xffffffffL
>>>>>>> typedef unsigned long int XIMPreeditState;
>>>>>>> Thanks, Vladimir
>>>>>>> 26 июля 2019 г., в 15:53, Ichiroh Takiguchi <takiguc at linux.vnet.ibm.com
>>>>>>>> написал(а):
>>>>>>> Hello.
>>>>>>> Could you review the fix ?
>>>>>>> Bug:    https://bugs.openjdk.java.net/browse/JDK-8227919
>>>>>>> Change: https://cr.openjdk.java.net/~itakiguchi/8227919/webrev.00/
>>>>>>> It's crash issue for Solaris SPARC platform.
>>>>>>> If possible, please push this fix into 13.
>>>>>>> XIMPreditState is changed to unsigned long instead of unsigned int
>>>>>>> on 64bit Big endian.
>>>>>>> It's same as Java8.
>>>>>>> I think Solaris SPARC's issue can be fixed by this fix.
>>>>>>> Vladimir,
>>>>>>> I appreciate your great help.
>>>>>>> Thanks,
>>>>>>> Ichiroh Takiguchi
>>>>>>> IBM Japan, Ltd.
>>>>>>> On 2019-07-23 20:00, Vladimir Kempik wrote:
>>>>>>> Hello
>>>>>>> it still crashes with  "Bus Error"
>>>>>>> I guess I may miss something on X server side.
>>>>>>> Thanks, Vladimir.
>>>>>>> 22 июля 2019 г., в 21:08, Ichiroh Takiguchi <takiguc at linux.vnet.ibm.com
>>>>>>>> написал(а):
>>>>>>> Hello Vladimir.
>>>>>>> I really appreciate your help.
>>>>>>> "XMODIFIERS=@im=local" is related Multi key Compose feature [1]
>>>>>>> One of document said "XMODIFIERS=@im=none" [2]
>>>>>>> Could you try "XMODIFIERS=@im=none" ?
>>>>>>> [1] https://www.x.org/releases/current/doc/man/man5/Compose.5.xhtml <
>>>>>>> https://www.x.org/releases/current/doc/man/man5/Compose.5.xhtml>
>>>>>>> [2] https://fedoraproject.org/wiki/I18N/InputMethods <https://
>>>>>>> fedoraproject.org/wiki/I18N/InputMethods>
>>>>>>> Thanks,
>>>>>>> Ichiroh Takiguchi
>>>>>>> On 2019-07-22 23:58, Vladimir Kempik wrote:
>>>>>>> Hello
>>>>>>> I?m getting Bus Error on startup with
>>>>>>> XMODIFIERS=@im=local ./xim_root
>>>>>>> nothing yet in log at that moment.
>>>>>>> Thanks, Vladimir
>>>>>>> 22 июля 2019 г., в 15:12, Ichiroh Takiguchi <takiguc at linux.vnet.ibm.com
>>>>>>>> написал(а):
>>>>>>> Hello Vladimir.
>>>>>>> I appreciate your reply.
>>>>>>> I did not expect "[1]w=0x0", it means xim_root could not receive
>>>>>>> focus window via XIC.
>>>>>>> I checked ibus XIM server on Linux s390x platform again, it also
>>>>>>> returned "[1]w=0" ...
>>>>>>> I was confused.
>>>>>>> I'd like to add new instruction
>>>>>>> 5. Run xim_root with XMODIFIERS=@im=local environment variable on
>>>>>>> desktop, like
>>>>>>> $ XMODIFIERS=@im=local ./xim_root
>>>>>>> 6. Move input focus to XIM demo window
>>>>>>> 7. Press Ctrl key and Press Right mouse button
>>>>>>> I tested xim_root on Linux s390x platform.
>>>>>>> $ XMODIFIERS=@im=ibus ./xim_root
>>>>>>> fevent = 0x3
>>>>>>> topwindow = 0x1e00001, window = 0x1e00002
>>>>>>> ClientMessage is received
>>>>>>> WM_TAKE_FOCUS is received, window=0x1e00002, timestamp=1470262945
>>>>>>> XSetInputFocus() is called
>>>>>>> FocusOut: window=0x1e00002(Internal Window)
>>>>>>> XUnsetICFocus
>>>>>>> FocusOut: window=0x1e00001(XIM demo - root-im window - 64bit)
>>>>>>> FocusIn: window=0x1e00001(XIM demo - root-im window - 64bit)
>>>>>>> FocusIn: window=0x1e00002(Internal Window)
>>>>>>> XSetICFocus
>>>>>>> KeyPress:
>>>>>>> ButtonPress:
>>>>>>> [1]w=0x0
>>>>>>> [2]w=0x0
>>>>>>> [1]state=0x0
>>>>>>> [2]state=0x0
>>>>>>> State: Unknown
>>>>>>> ClientMessage is received
>>>>>>> I tried it with XMODIFIERS=@im=local
>>>>>>> $ XMODIFIERS=@im=local ./xim_root
>>>>>>> fevent = 0
>>>>>>> topwindow = 0x2000001, window = 0x2000002
>>>>>>> ClientMessage is received
>>>>>>> WM_TAKE_FOCUS is received, window=0x2000002, timestamp=1470356372
>>>>>>> XSetInputFocus() is called
>>>>>>> FocusIn: window=0x2000001(XIM demo - root-im window - 64bit)
>>>>>>> FocusIn: window=0x2000002(Internal Window)
>>>>>>> XSetICFocus
>>>>>>> KeyPress:
>>>>>>> ButtonPress:
>>>>>>> [1]w=0x2000002
>>>>>>> [2]w=0x2000002
>>>>>>> [1]state=0x2
>>>>>>> [2]state=0x2
>>>>>>> State: IM Off
>>>>>>> ClientMessage is received
>>>>>>> I was expected this result.
>>>>>>> $ XMODIFIERS=@im=kinput2 ./xim_root
>>>>>>> fevent = 0x3
>>>>>>> topwindow = 0x2000001, window = 0x2000002
>>>>>>> ClientMessage is received
>>>>>>> WM_TAKE_FOCUS is received, window=0x2000002, timestamp=1470386128
>>>>>>> XSetInputFocus() is called
>>>>>>> FocusIn: window=0x2000001(XIM demo - root-im window - 64bit)
>>>>>>> FocusIn: window=0x2000002(Internal Window)
>>>>>>> XSetICFocus
>>>>>>> KeyPress:
>>>>>>> ButtonPress:
>>>>>>> [1]w=0x200000200000000
>>>>>>> [2]w=0x2000002
>>>>>>> [1]state=0x200000000
>>>>>>> [2]state=0x2
>>>>>>> State: IM Off
>>>>>>> ClientMessage is received
>>>>>>> I'm sorry, it seemed current code did not work on even if Linux
>>>>>>> s390x platform.
>>>>>>> Thanks,
>>>>>>> Ichiroh Takiguchi
>>>>>>> On 2019-07-19 23:29, Vladimir Kempik wrote:
>>>>>>> Hello
>>>>>>> I?m probably missing something about XIM on X side here, using ssh
>>>>>>> -X from ubuntu(with Xserver) to solaris11 sparc64 machine
>>>>>>> the output is
>>>>>>> fevent = 0x3
>>>>>>> topwindow = 0x3800001, window = 0x3800002
>>>>>>> ClientMessage is received
>>>>>>> WM_TAKE_FOCUS is received, window=0x3800002, timestamp=3262503
>>>>>>> XSetInputFocus() is called
>>>>>>> FocusIn: window=0x3800001(XIM demo - root-im window - 64bit)
>>>>>>> FocusIn: window=0x3800002(Internal Window)
>>>>>>> XSetICFocus
>>>>>>> FocusOut: window=0x3800002(Internal Window)
>>>>>>> XUnsetICFocus
>>>>>>> FocusOut: window=0x3800001(XIM demo - root-im window - 64bit)
>>>>>>> ClientMessage is received
>>>>>>> WM_TAKE_FOCUS is received, window=0x3800002, timestamp=3267238
>>>>>>> XSetInputFocus() is called
>>>>>>> ButtonPress:
>>>>>>> [1]w=0x0
>>>>>>> [2]w=0x0
>>>>>>> [1]state=0x0
>>>>>>> [2]state=0x0
>>>>>>> State: Unknown
>>>>>>> FocusIn: window=0x3800001(XIM demo - root-im window - 64bit)
>>>>>>> FocusIn: window=0x3800002(Internal Window)
>>>>>>> I also had to press left button, not right one.
>>>>>>> I think when you initialy created 8213232 you were targeting linux
>>>>>>> ppc64 BE system in your patch and simply forgot about another 64-bit
>>>>>>> BE system - solaris sparc64
>>>>>>> I would propose this fix for 8227919
>>>>>>> --- a/src/java.desktop/unix/native/libawt_xawt/awt/awt_InputMethod.c
>>>>>>> +++ b/src/java.desktop/unix/native/libawt_xawt/awt/awt_InputMethod.c
>>>>>>> @@ -1671,7 +1671,7 @@
>>>>>>> {
>>>>>>>   X11InputMethodData *pX11IMData = NULL;
>>>>>>>   char * ret = NULL;
>>>>>>> -#if defined(_LP64) && !defined(_LITTLE_ENDIAN)
>>>>>>> +#if defined(__linux__) && defined(_LP64) && !defined(_LITTLE_ENDIAN)
>>>>>>>   // XIMPreeditState value which is used for XGetICValues must be
>>>>>>> 32bit on BigEndian XOrg's xlib
>>>>>>>   unsigned int state = XIMPreeditUnKnown;
>>>>>>> #else
>>>>>>> Such change already allowed to pass our internal tests on solaris
>>>>>> sparc64
>>>>>>> if you think it?s ok then let me take over and finish with 8227919
>>>>>>> Thanks, Vladimir
>>>>>>> 19 июля 2019 г., в 16:57, Ichiroh Takiguchi
>>>>>>> <takiguc at linux.vnet.ibm.com <mailto:takiguc at linux.vnet.ibm.com> <
>>>>>>> mailto:takiguc at linux.vnet.ibm.com <mailto:takiguc at linux.vnet.ibm.com>><
>>>>>>> mailto:takiguc at linux.vnet.ibm.com <mailto:takiguc at linux.vnet.ibm.com><
>>>>>>> mailto:takiguc at linux.vnet.ibm.com <mailto:takiguc at linux.vnet.ibm.com>>>>
>>>>>>> написал(а):
>>>>>>> Hello.
>>>>>>> I need Solaris 64bit SPARC desktop user volunteer for jdk-8227919 [1]
>>>>>>> Please follow the steps:
>>>>>>> 1. Download xim_root.c [2] and compile it on Solaris 64bit SPARC
>>>>>>> 2. Run xim_root on desktop
>>>>>>> 3. Move input focus to XIM demo window
>>>>>>> 4. Press Ctrl key and Press Right mouse button
>>>>>>> Please put the output into JDK-8227919 [1] or post it into mailing list.
>>>>>>> I really appreciate if you are using XIM like IIIMF.
>>>>>>> [1] https://bugs.openjdk.java.net/browse/JDK-8227919
>>>>>>> [2] https://cr.openjdk.java.net/~itakiguchi/8227919/xim_root.c
>>>>>>> Thanks,
>>>>>>> Ichiroh Takiguchi

More information about the awt-dev mailing list