Corretto Java Flexible Path to repair OpenJDK, Corretto Floating Point Arithmetic, StrictMath, bound classes and interfaces?
sminervini.prism
sminervini.prism at protonmail.com
Sun Apr 10 12:04:05 UTC 2022
To OpenJDK core-libs-dev,
I have been part way through a discussion with Amazon Web Services and Corretto on the subject
of a patch to mend OpenJDK's floating point woes. Particularly in a way which will leave
the option of mutual compatability. So far, they seem to take an initial view with their patching
of Java, and have directed us here.
As can be seen and has been outlined through all these discussion threads, it is the case that
IEEE 754 is incomplete in the area of floating point and base 10, and that the JRE can generate
denormal and pronormal values with float and double, which are really range errors.
We have seen a compatability option in the shape of Corretto Java, particularly with their
patching policy to enhance things. We have also thought that mutual compatibility means
no disadvantages, particularly if there were some way to use both approaches in the same
code space, such as a runtime switch would allow, since java can call itself from the command
line, at least.
We have tried to lay down the simplicity and requisite nature of this situation, for ourselves
and others, Java programmers and users, in the following:
https://github.com/corretto/corretto-18/issues/15
https://github.com/corretto/corretto-18/issues/17
https://github.com/corretto/corretto-18/issues/18
https://repost.aws/questions/QU8pXxczMPRdyGypZM2NNX2A/java-floating-point-correction-patch-in-corretto-in-a-compatible-manner
https://repost.aws/questions/QUBDoBC76zQ8KKLwP6UCc7Xg/compatability-enhancement-of-corretto-java-floating-point
Could something like a floating point patch be appropriate for the Corretto space for Java, were
it mutually compatible and efficient, for stable relase versions of Open versions of Java, upcoming
and continuing to the future?
S.M.
More information about the core-libs-dev
mailing list