Hi Petr,
src/macosx/native/sun/awt/CImage.m:
452 NSBitmapImageRep* imageRep = CImage_CreateImageRep(env, buffer, width,
height);
453 if (imageRep) {
454 NSData *tiffImage = [imageRep TIFFRepresentation];
455 jsize tiffSize = (jsize)[tiffImage length];
456 result = (*env)->NewByteArray(env, tiffSize);
457 CHECK_NULL_RETURN(result, nil);
458 jbyte *tiffData = (jbyte *)(*env)->GetPrimitiveArrayCritical(env,
result, 0);
459 CHECK_NULL_RETURN(tiffData, nil);
460 [tiffImage getBytes:tiffData];
461 (*env)->ReleasePrimitiveArrayCritical(env, result, tiffData, 0);
462 [imageRep release];
I see that we're managing the imageRep manually. So it is not a part of
the auto-release pool created with the JNF_COCOA_ENTER macro. Which
means that the return statements at lines 457 and 459 will lead to a
memory leak. I suggest to use @finally to ensure that the imageRep gets
released (this is what the JNF_COCOA_EXIT does, too).
--
best regards,
Anthony
On 3/7/2014 10:59 AM, Petr Pchelko wrote:
Hello,
Please review the updated version of the fix:
http://cr.openjdk.java.net/~pchelko/9/8030710/webrev.02/
I've moved the image handling code from DataTransferer to CImage. This let us
reuse the existing code that creates an NSBitmapImageRep from the int array and
let us get rid of some JNI upcalls.
Tested with all SQE datatransfer tests and out reg tests. No new failures.
With best regards. Petr.
On 06.03.2014, at 19:38, Petr Pchelko <[email protected]> wrote:
Hello, Sergey.
To retrieve the image from native we are using NSDeviceRGBColorSpace (see
CImage.m), so we need to use the same color space to put the image into native.
If i revert this line no calibrated space the test would fail because the image
we get from native is different from the one we put.
I’ve just got an idea. CImage API is able to put the Java Image into native, so
we could probably reuse it here. Let me investigate this, I’ll write you back
when I succeed or fail.
With best regards. Petr.
06 марта 2014 г., в 7:22 после полудня, Sergey Bylokhov
<[email protected]> написал(а):
Hi, Petr.
How NSCalibratedRGBColorSpace change affects the fix?
On 3/6/14 5:30 PM, Petr Pchelko wrote:
Hello again.
I’ve updated the fix. The new version is here:
http://cr.openjdk.java.net/~pchelko/9/8030710/webrev.01/
1. I’ve modified the test so it now doesn’t need .html file, also copyright was
added.
2. I’ve replaced the public.jpeg mime with a predefined kUTTypeJPEG (which is
the same).
No such constant in Cocoa, only in Carbon.
With best regards. Petr.
06 марта 2014 г., в 2:56 после полудня, Petr Pchelko <[email protected]>
написал(а):
Hello, Sergey.
I guess JFIF is not a typo in CDataTransferer?
Nop. http://en.wikipedia.org/wiki/JPEG_File_Interchange_Format
Also please see the macosx flavormap.properties type. We call a native as a
JFIF and it’s left as is because
it’s called like this on other platforms.
Do we really need ImageTransferTest.html in the test?
The test is for interprocess DnD, so we run the first process as an applet from
jtreg.
The applet then starts a second processes main(). I’ll check if I could use 2
main() in one file and if it’s fine for jtreg.
Also copyright header is missing.
Thank you. I’ll update the fix shortly.
With best regards. Petr.
06 марта 2014 г., в 2:49 после полудня, Sergey Bylokhov
<[email protected]> написал(а):
Hi, Petr.
I guess JFIF is not a typo in CDataTransferer? Do we really need
ImageTransferTest.html in the test?
Also copyright header is missing.
On 3/5/14 8:23 PM, Petr Pchelko wrote:
Hello, AWT Team.
Please review the fix for the issue:
https://bugs.openjdk.java.net/browse/JDK-8030710
The fix is available at:
http://cr.openjdk.java.net/~pchelko/9/8030710/webrev/
The problem was that on Mac we thought that re support PNG and JPEG image
formats, but in reality we did not have mappings for these formats. I've added
the missing mappings.
The test is being open sourced. The fix was checked with out tests and with
native-Java DnD.
With best regards. Petr.
--
Best regards, Sergey.
--
Best regards, Sergey.
