Hi Sergey,
CGLGC.java:
In createCompatVM() I dislike separated tests for "here is early
rejection of the list of things that I can handle" followed by a list of
tests for things it can handle. For one thing we have to test the type
more than once. But mainly it just seems like the two tests can get out
of synch over time. It just seems cleaner to only iterate a set of
allowed types once. Wouldn't it be simpler and more straightforward to
do something like:
if (BITMASK) { return null }
if (type = FBO) {
if (!CAPS_EXT_FBOBJECT) return null;
} else if (type = TEXTURE) {
} else {
return null;
}
which is a list of positive tests for what can be done rather than a
parallel list of tests - one set of tests negative and the other positive.
CGLVSM.java:
In the constructor we claim to accelerate any surface if
CAPS_EXT_FBOBJECT is present, but CGLGC.java has a test to reject
BITMASK. Shouldn't the 2 perform the same test?
CGLSD.m, line 114: The comment still refers to "pbuffer" in the new
version. You changed one reference to WINDOW, but not the other reference.
GLXGC.java - same comment as CGLGC.java
GLXVSM.java - same comment as CGLVSM.java
GLXSD.c - In DestroyOGLSurface() - should we leave the Traceln in even
though the method is empty?
WGLGC.java - same comment as CGLGC.java
WGLVSM.java - same comment as CGLVSM.java
WGLSD.c - In DestroyOGLSurface() - should we leave the Traceln in even
though the method is empty?
...jim
On 6/16/15 6:01 AM, Sergey Bylokhov wrote:
Hello.
Please review the fix for jdk9.
The reason of the removing is a general deprecation of pbuffers, and a
lack of their support in the java2d, because for a long time pbuffers
were not used by default.
Attempts to disable FBO(which are used by default), and enable the
pbufferes will cause a different crashes in different places on all our
platforms(mac,lin,win).
Change description:
- Possibility to create a VolatileImage on top of pbuffers was
removed. FBO and Textures are supported only.
- "-Dsun.java2d.opengl.fbobject" now totally disable accelerated
volatile images(BufSurfaces will be used instead). But it does not
affect manageable images.
- CAPS_STORED_ALPHA is removed because it was used to check is pbuffer
is able to store alpha or not.
Note that pbuffers are still used to get some information from the
opengl on the start. I plan to replace these usage later.
Some related questions:
- Does anybody know about AccelTypedVolatileImage, how it was planned
to use? It seems that the work on it was not completed and its usage can
be simply removed/replaced.
Bug: https://bugs.openjdk.java.net/browse/JDK-7188942
Webrev can be found at: http://cr.openjdk.java.net/~serb/7188942/webrev.01
--
Best regards, Sergey.
