[OpenJDK 2D-Dev] [9] Review Request- JDK-8152971- JNI Warning with -Xcheck:jni
Prahalad Kumar Narayanan
prahalad.kumar.narayanan at oracle.com
Mon Jul 11 19:02:44 UTC 2016
Hello Everyone
Good day to you.
A quick follow-up on the bug fix for : JDK-8152971 JNI Warning with -Xcheck:jni
First, Thanks to Phil for the time in review and detailed explanation.
To answer the question -
"The issue with local refs is that you want to avoid too many so there is no need to do that release before return explicitly since the VM will do it for us. same for the direct calls to deletelocalref. Do you get any JNI warnings if you remove these ? "
Answer :
Yes., I see the warnings if DeleteLocalRef is not invoked on objects created in the callback functions.
It is in these callback functions that many objects get created without a corresponding release api.
No warnings were seen when releaseGdiFontMapReferences() was taken out.
This function is removed in the latest change.
The behavior could be attributed to two points
a. FontMap having most of its references to jObjects that are on stack (as arguments to function).
b. Very few references to local objects are created in the parent JNI function ( 2 - 3 classIDs )
To Summarize the Changes
. In all callback functions, the DeleteLocalRef calls have been reorganized to avoid cluttering before return statements.
. releaseGdiFontMapReferences() function has been removed.
The only font map reference that is released in latest code - is the reference to jClass object along with other classID references. This could be avoided but has been retained as a safe measure.
The changes are available under webrev link:
http://cr.openjdk.java.net/~pnarayanan/8152971/webrev.04/
Kindly review at your convenient time and provide your suggestions.
Thanks for your review & feedback
Have a good day
Prahalad N.
-----Original Message-----
From: Phil Race
Sent: Friday, July 08, 2016 1:52 AM
To: Prahalad Kumar Narayanan
Cc: Prasanta Sadhukhan; 2d-dev at openjdk.java.net; Jayathirth D V; Praveen Srivastava
Subject: Re: [9] Review Request- JDK-8152971- JNI Warning with -Xcheck:jni
So far as I can see we can eliminate at least some of the clutter of the calls to deletelocalreference by simply doing it as soon as we are done with the reference
eg
309 familyLC = (*env)->CallObjectMethod(env, fmi->family,
310 fmi->toLowerCaseMID, fmi->locale);
can be immediately followed by :
DeleteLocalReference(env, familyLC);
.. whether or not there is a pending exception.
and
343 (*env)->CallObjectMethod(env, fmi->familyToFontListMap,
344 fmi->putMID, familyLC, fmi->list);
can be immediately followed by :
DeleteLocalReference(env, fmi->family);
DeleteLocalReference(env, fmi->list);
.. whether or not there is a pending exception.
and so on.
releaseGdiFontMapReferences() seems to be used right before returning to Java.
The issue with local refs is that you want to avoid too many so there is no need to do that release before return explicitly since the VM will do it for us.
same for the direct calls to deletelocalref.
Do you get any JNI warnings if you remove these ?
I could even argue that some of the ones before return (0) from the enum functions which will cause us to then exit up the call stack fall into the same bucket but if you can simplify the change as I suggest it'll be easier to sort that out as there'll be less noise.
-phil.
On 07/04/2016 02:43 AM, Prahalad Kumar Narayanan wrote:
> Hello Everyone
>
> Good day to you.
>
> A quick follow-up on the fix for
> Bug ID : JDK-8152971
> Title : -Xcheck:jni - WARNING in native method
>
> Thanks to Phil for his feedback.
> The feedback was not only detailed.. but contained important links for reference.
>
> I 've incorporated the review suggestions and have updated the webrev.
> Brief on changes from previous revision:
>
> 1. Once an exception has been found in an Up-call, the exception is ' not ' consumed or cleared now.
> . As Phil rightly said, this will pave way for the Java methods to know of the exceptions and handle them.
>
> 2. In addition to point No.1, The GDI's callback functions return zero '0' during an exception.
> . This will stop the subsequent enumeration of fonts &
> calls to registered callbacks through the APIs
>
> 3. Since exceptions are not consumed, all GDI 's callback functions are checked for pending exceptions at the entry.
> . On pending exceptions, the GDI's callback functions return immediately.
> . This prevents the functions from invoking any other JNI calls that could be un-safe during a pending exception
> . The callback functions merely release any local reference held before returning to caller.
>
> 4. Since we no longer support Win98, the obsolete ANSI versions of functions (~A functions) have been removed.
> . Functions that work with Unicode strings invoking Unicode versions of Windows APIs (~W functions) are retained.
>
> 5. The jtreg shell script has been modified to execute on all platforms after testing its execution on Windows, Linux, Mac and SunOS.
> . Just to re-iterate: The jtreg script fails on windows and solaris but in java.lang codebase.
> . A bug is already Open indicating JNI warnings in java.lang package. Hence this change does not introduce any new issues.
>
> As with any proposed webrev:
> . The changes were run through internal build system to ensure safe build on all supported OS versions.
> . JCK and JTREG test cases were run to ensure no conformance or regression failures occur.
>
> Kindly review the changes at your convenience and provide your feedback.
> . Updated webrev link:
> http://cr.openjdk.java.net/~pnarayanan/8152971/webrev.03/
>
> Thank you for your time & effort in review Have a good day
>
> Prahalad N.
>
> ----------------------------------------------------------------------
> ------
> From: Philip Race
> Sent: Wednesday, June 29, 2016 10:48 PM
> To: Prasanta Sadhukhan
> Cc: Prahalad Kumar Narayanan; 2d-dev at openjdk.java.net; Jayathirth D V;
> Praveen Srivastava
> Subject: Re: [9] Review Request- JDK-8152971- JNI Warning with
> -Xcheck:jni
>
> https://docs.oracle.com/javase/8/docs/technotes/guides/jni/spec/design
> .html#exception_handling
>
> lists DeleteLocalRef as safe to call with a pending exception.
>
> So here ...
> 74 fullnameLC = (*env)->CallObjectMethod(env, fullname,
> 175 fmi->toLowerCaseMID, fmi->locale);
> 176 if ((*env)->ExceptionCheck(env)) {
> 177 (*env)->ExceptionClear(env);
> 178 /* Delete the created references */
> 179 DeleteLocalReference(env, fullname);
> 180 return 1;
> 181 }
>
> ... perhaps what we should do is not clear the exception and with the
> goal that after returning from this function we can check in the
> caller for a pending exception and return from JNI to Java *without
> clearing it* - so that Java code gets that exception propagated. I
> suggest this because I think we would have a correctness issue which
> should be reported - not swallowed - if there ever really was one.
>
> I think this means that the GDI callbacks need to check for a pending
> exception on entry and bail since once we hand off to windows it may
> be called repeatedly.
>
> But also we should stop enumeration if we detect an error hence we
> should return 0 in this case, not 1 per the docs for EnumFontFamExProc
>
> ---
> https://msdn.microsoft.com/en-us/library/windows/desktop/dd162618(v=vs
> .85).aspx
>
>
> Return value
>
> The return value must be a nonzero value to continue enumeration;
> to stop enumeration, the return value must be zero.
>
> ---
>
> Also the "A" functions are now obsolete. No win 98 support any more
> :-) We should just delete them instead of updating them.
>
> And I'd prefer the test be verified on Solaris rather than excluding
> it
>
> -phil
>
>
> On 6/27/16, 10:24 PM, Prasanta Sadhukhan wrote:
> Looks good to me. Only thing is in test script, you need to add copyright 2015, 2016 as the script is originally written in 2015 and you cannot omit the created year from the copyright.
>
> Regards
> Prasanta
> On 6/27/2016 4:17 PM, Prahalad Kumar Narayanan wrote:
>
> Hello Everyone
>
> Good day to you.
>
> Quick follow up to fix for the Bug
> Bug ID : JDK-8152971
> Bug Link : https://bugs.openjdk.java.net/browse/JDK-8152971
>
> First, Thanks to Prasanta for his review & suggestions.
>
> I 've incorporated some of the review suggestions & updated the webrev.
> Webrev Link :
> http://cr.openjdk.java.net/~pnarayanan/8152971/webrev.02/
>
> Overview on changes from previous webrev:
>
> 1. Code Changes: As Prasanta mentioned,
> The variable fontStrLC should be used in place of fontStr - The correction has been taken up.
> And releaseGdiFontMapReferences call is not required when GdiFontMapInfo object isn't initialized - The particular invocation has been removed.
>
> 2. With regard to deleting references held within GdiFontMapInfo struture
> JNI creates local references only for objects that are created at the Java side - jobject, jstring and jclass
> The other FMI variables - method IDs are not references to objects. Hence invoking deleteLocalRef.. is not valid (may not compile too).
>
> 3. The Jtreg shell script - LoadFontsJNICheck.sh was initially written to run only on MacOS.
> Since it addresses the bug at hand (which is on windows), I tested the script 's execution on Win and Linux.
> The Jtreg script passes on linux and fails on windows with warnings in java.lang codebase.
> As I understand, there is a JBS bug filed already pertaining to JNI warnings in java.lang package.
> Hence enabling the script on windows, doesn't introduce new bug.
> The only OS that the script doesn't run - Solaris.
> Presently, we don't have a Solaris environment at our facility to test the script.
> Hence the original functionality on Solaris is retained.
>
> Kindly take a look at the changes at your convenience & provide the suggestions.
>
> Thank you for your review
> Have a good day
>
> Prahalad N.
>
> ----------------------------------------------------------------------
> ------
> From: Prasanta Sadhukhan
> Sent: Monday, June 27, 2016 11:51 AM
> To: Prahalad Kumar Narayanan; 2d-dev at openjdk.java.net
> Cc: Philip Race; Jayathirth D V; Praveen Srivastava
> Subject: Re: [9] Review Request- JDK-8152971- JNI Warning with
> -Xcheck:jni
>
> I guess this method(s) should take "fontStrLC" instead of "fontStr"
> 650 (*env)->CallObjectMethod(env, fontToFileMap, fmi->putMID,
> 651 fontStr, fileStr);
>
> 692 (*env)->CallObjectMethod(env, fontToFileMap, fmi->putMID,
> 693 fontStr, fileStr);
>
> 762 (*env)->CallObjectMethod(env, fontToFileMap, fmi->putMID,
> 763 fontStr, fileStr);
>
> 805 (*env)->CallObjectMethod(env, fontToFileMap, fmi->putMID,
> 806 fontStr, fileStr);
> It seems this line is not needed as we have not populated fmi structure yet.
> 882 releaseGdiFontMapReferences(env, &fmi);
>
> Why aren't we deleting fmi->env,fmi.arrayListCtr,fmi.addMID,fmi.putMID in releaseGdiFontMapReferences()?
>
> Also, it seems earlier the testscript was supposed to execute only on Mac but now you are extending the execution platform to windows and linux as well excluding only solaris.
> Is there any particular need to restrict solaris too?
>
> Regards
> Prasanta
>
>
> On 6/24/2016 7:34 PM, Prahalad Kumar Narayanan wrote:
> Hello Everyone on Java2D Forum
> Good day to you.
> A Quick follow-up on webrev to fix the following issue
> Bug ID : JDK-8152971 / -Xcheck:jni - warning in native method
> Bug Link :
> https://bugs.openjdk.java.net/browse/JDK-8152971?filter=-1
> Updated webrev link:
> http://cr.openjdk.java.net/~pnarayanan/8152971/webrev.01/
> Description on Changes
> . The previous webrev contained changes to additional files which are not related to the current bug in concern.
> . Henceforth, the updated webrev limits the changes to only fontpath.c and a Jtreg test script to verify the change.
> Regarding Build & Test
> . Though the changes pertain to windows specific code, internal build system was triggered to ensure safe build on all supported platforms.
> . In addition, no new Jtreg failures were found with the proposed changes.
> Kindly review the changes at your convenience & provide your feedback.
> Thank you for your time in review
> Have a good day
> Prahalad N.
>
> -----Original Message-----
> From: Prahalad Kumar Narayanan
> Sent: Wednesday, June 22, 2016 3:20 PM
> To: 2d-dev at openjdk.java.net
> Cc: Philip Race; Prasanta Sadhukhan; Jayathirth D V; Praveen
> Srivastava
> Subject: [9] Review Request- JDK-8152971- JNI Warning with -Xcheck:jni
> Hello Everyone, on Java2D Group Good day to you.
> Please find herewith, webrev changes to fix the bug-
> Bug ID : JDK-8152971 / -Xcheck:jni - warning in native method
> Bug Link : https://bugs.openjdk.java.net/browse/JDK-8152971?filter=-1
> Webrev :
> http://cr.openjdk.java.net/~pnarayanan/8152971/webrev.00/
> Description on Bug:
> 1. Too many JNI warnings are reported in the native functions when test-code is run with VM Option: -Xcheck:jni
> 2. The warnings can be classified into 2 categories
> a. JNI warnings that are thrown due to the missing exception checks after an Up call ( JNI function invoking Java method )
> b. JNI warnings that are thrown due to increased usage of Local Reference objects.
> Description on the Fix:
> 1. File : FontPath.c
> Added JNIEnv->ExceptionCheck() and ExceptionClear() where Up call is invoked.
> The checks are added only to the valid Up calls as per JNI spec.
> Added JNIEnv->DeleteLocalRef where the native functions created Java side objects but did not delete their local reference after usage.
> A few of the native functions get invoked as Callbacks to Windows APIs following the font enumeration.
> Therefore at each callback, the count of active local references would increase.
> Over time, the active local references would exceed the planned number of local references set by JVM.
> 2. File : awt_Component.cpp
> Added JNIEnv->ExceptionCheck() and ExceptionClear() where an Up call displayed a JNI warning while running the Jtreg test script.
> Information on Jtreg test script is given below.
> 3. File : LoadFontsJNICheck.sh
> The shell script is already a part of JTreg test case & invokes LoadFontsJNICheck with -Xcheck:jni VM option
> However, the script was configured to run only on Mac OS. Now, the script is modified to run on windows, linux and mac systems.
> This way, the code changes can be checked for proper execution with test-case.
> Kindly review the changes at your convenience and share your feedback.
> Thank you for your time in review
> Have a good day
> Prahalad N.
More information about the 2d-dev
mailing list