To oscardsmith and other fdlibm developers,
openlibm strikes me as being the perfect correction and replacement for fdlibm, which is what java.lang.StrictMath presently uses. Java’s floating point errors involving float and double occur in the parts of float and double number values between one and zero (1 and 0) than one, which means that the logic already present for the whole numbers can be copied and reapplied for one and zero (1 and 0) float or double number values, despite a small numbers range change. Floating point errors can occur via fdlibm inside java.lang.StrictMath, which is why we seek for a group to replace fdlibm with openlibm inside StrictMath. openlibm never returns floating point error convoluted values.
It is both our mission critical need, while also under our own confidentiality, that we need Java OpenJDK floating point math fixed, just for the OpenJDK FOSS version of Java and its present major platforms. This is despites the Decimal and BigDecimal “workarounds”. Do you know of anyone connected to openlibm both capable and willing to apprehend this task for Java? With OpenJDK’s floating point mathematics corrected, and that seen in relation to
meaning that Java’s “Arbitrary Precision” (really High Precision) becomes far better, since the precision variable could be changed and set up to be used once overhead, and not repeated for every High Precision arithmetic method call, Java’s mathematics situation itself for base 10, and hexadecimal, the view to OpenJDK’s mathematics would be far better.
Now, if project Valhalla brings along more operator support too, that could be applied to the afore-linked big-math. It that could be done, in terms of range, operators, and precision use, combined with the repair of OpenJDK’s floating point errors, Java OpenJDK and its mathematics become an entirely different proprosition to look and and utilise.
Is there anyone over there prepared to take the earlier part, floating point and StrictMath corrections, on board as a public release project?