Hannu Hartikainen
Hannu Hartikainen
I'm hoping you who have been active with this PR could review and test #3028.
This seems interesting indeed, though I'm pretty sure the performance will be awful on most platforms. For individual cases there are probably much more performant workarounds/hacks using 64-bit floats (but...
I guess the instructions that may change the control word are - `FINIT` / `FNINIT` (set CW to `037F`, ie. extended precision) - `FLDCW` / `FLDENV` - `FRSTOR` / `FXRSTOR`...
At a glance it seems like compiling software that uses `__float128` is not trivial even on platforms supporting it. It will probably only be possible using GCC + libquadmath. And...
Right, so the occasional person claiming to have compiled float128 code for Apple M1 is just plain mistaken. :joy: At least that makes the implementation easier and more portable.
Adding a dependency seems to also be the correct fix per [JEP 320](http://openjdk.java.net/jeps/320), though `[javax.activation/javax.activation-api "1.2.0"]` is suggested. The JavaMail API still uses javax.activation.DataHandler legitimately.
We've been using the dependency and postal in production since May, and have had no issues. I'm not sure if adding the dependency in postal itself causes issues in JDK...
I've got this same error with OpenCV 2.4. More specifically, I have installed the ubuntu packages `python-opencv2.4`, `libopencv-core2.4` etc.
I'd like this too. I live near the arctic circle so in the winter it's always dark but in the summer it never is. :)
See pull request #37 for an experimental implementation.