I'd go with option 2 (in the release branch) to get rid of the
regression with practically no risk for the release. Performance-wise,
nothing is lost compared to 0.92beta.
On 20.12.2006 09:08:38 Simon Pepping wrote:
I would like to know what to do for the release:
1. Leave as is, a known problem.
2. Do the quick fix as jeremias suggests: putting the Commons codec
before the ImageIO variant in ImageFactory.
3. Test Jeremias' patch and apply it.
We can do the fix in the release branch (to be created) only, in the
interest of an errorless release.
Simon
On Tue, Dec 19, 2006 at 11:14:13PM +0100, Jeremias Maerki wrote:
Actually, this is so simple, I've created a patch. I'm hesitant to apply
it without much testing with various PNGs for which I have no time right
now. But maybe one of you can take a look. If it's good JAIImage and
JimiImage could be similarly patched.
On 19.12.2006 23:01:36 Jeremias Maerki wrote:
It turns out the following revision is responsible for the regression:
http://svn.apache.org/viewvc?rev=424349view=rev
Putting the Commons codec before the ImageIO variant in ImageFactory is
a quick fix for this. I originally switched because of speed reasons but
obviously I didn't test PNGs with alpha channel.
The reason why Martin's PNG doesn't work with the ImageIOImage class is
the lack of support for java.awt.Transparency.TRANSLUCENT. If it were a
BITMASK it would work. The same problem exists for JAIImage and
JimiImage, BTW. Again, this is something a complete redesign of the
image package would have addressed. It's on my higher priority list but
I still haven't got to it, yet. If someone wants to try to implement
that little code that is missing as a temporary fix, please do. It
shouldn't be difficult. You can maybe use XmlGraphicsCommonsImage for
hints.
--
Simon Pepping
home page: http://www.leverkruid.eu
Jeremias Maerki