[Octave-bug-tracker] [bug #59088] Failing java.lang.Float BIST test in ov-java.cc

[Philip Nienhuis]

URL:
  <https://savannah.gnu.org/bugs/?59088>

                 Summary: Failing java.lang.Float BIST test in ov-java.cc
                 Project: GNU Octave
            Submitted by: philipnienhuis
            Submitted on: Wed 09 Sep 2020 07:50:50 PM CEST
                Category: Libraries
                Severity: 3 - Normal
                Priority: 5 - Normal
              Item Group: Unexpected Error or Warning
                  Status: Patch Submitted
             Assigned to: None
         Originator Name: Philip Nienhuis
        Originator Email: 
             Open/Closed: Open
                 Release: dev
         Discussion Lock: Any
        Operating System: Any

    _______________________________________________________

Details:

Running  __run_test_suite__.m  in dev Octave (hg id eb35003c3851) on a Win10
system with Java 10 I see the following failing test in fntests.log:

>>>>> processing
L:\philip\devel\octdev\mxe\mxe_64w_20200818D\dist\octave7\mingw64\share\octave\7.0.0\etc\tests\libinterp\octave-value\ov-java.cc-tst
***** testif HAVE_JAVA; usejava ("jvm")
 n = javaObject ("java.lang.Float", 1.35);
 assert (n.compareTo (1.0), 1);
 assert (n.compareTo (1.35), 0);
 assert (n.compareTo (10), -1);
 assert (n.doubleValue (), 1.35, 1e7);
!!!!! test failed
[java] java.lang.ClassCastException: class java.lang.Double cannot be cast to
class java.lang.Float (java.lang.Double and java.lang.Float are in module
java.base of loader 'bootstrap')


This is probably due to Java 10 being stricter than Java 8 (on my Linux build
systems: OpenJDK Runtime Environment, build 1.8.0_262-b10); the "compareTo()"
method now only accepts arguments of equal type. So for said test it can only
compare variable "n" to a Java float value, or an Octave single.

Even though I initially wrote the test (8b5bc5e5f74b) I'm unsure what it was
meant to check exactly. Assuming it is meant to check that javaObject can
actually produce a Java float with numeric value 1.35, the following
(admittedly fairly convoluted) code seems to work better:

n = javaObject ("java.lang.Float", 1.35);
n.compareTo (single (javaObject ("java.lang.Integer", 1).floatValue ()))    # 
1
n.compareTo (single (javaObject ("java.lang.Double", 1.35).floatValue ()))  # 
0
n.compareTo (single (javaObject ("java.lang.Double", 10).floatValue ()))    #
-1


Patch along those lines attached.

Note that I got this with dev Octave, but I expect it to also fail on stable
Octave (6.0.90) with Java 10 as ov-java.cc is essentially the same w.r.t.
numerics processing for stable and dev Octave.

FWIW, the above code won't run on Matlab as that complains about "Undefined
variable 'javaObject' or class 'javaObject'". Maybe Octave's boxing/unboxing
works differently or Matlab's parser gets confused.
In Octave the "cast" to single is required because in ov-java.cc all numeric
scalar Java types are converted to Octave doubles; and an ensuing ".floatValue
()" method produces a scalar numeric Java value as well:

>> class (javaObject ("java.lang.Double", 10).floatValue ())
ans = double





    _______________________________________________________

File Attachments:


-------------------------------------------------------
Date: Wed 09 Sep 2020 07:50:50 PM CEST  Name: ov-java_cc_float-BIST.patch 
Size: 764B   By: philipnienhuis

<http://savannah.gnu.org/bugs/download.php?file_id=49772>

    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?59088>

_______________________________________________
  Message sent via Savannah
  https://savannah.gnu.org/