Foreign.getInstance().longDoubleFromString returns garbage

[aploese]

The failing of NumberTest.returnDefaultF128HighPrecision is buried here.
To reveal the reason, add the following test case to ForeignTest:
@Test public void longDoubleFromString() { String strValue = "1.234567890123456789"; BigDecimal value = new BigDecimal(strValue); assertEquals(value.toEngineeringString(), value.toString()); byte[] ld = new byte[Type.LONGDOUBLE.size()]; Foreign.getInstance().longDoubleFromString(value.toEngineeringString(), ld, 0, Type.LONGDOUBLE.size()); String strRetValue = Foreign.getInstance().longDoubleToString(ld, 0, 0); // String strRetValueE = Foreign.getInstance().longDoubleToEngineeringString(ld, 0, 0); BigDecimal retValue = new BigDecimal(strRetValue); BigDecimal delta = value.subtract(retValue).abs(); assertTrue("Not equals, expected: " + value.toEngineeringString() + " but was: " + retValue.toEngineeringString(), delta.compareTo(new BigDecimal("0.0000000000000000001")) < 0); }

I will poke in the c sources to figure out whats wrong...

[aploese]

The problem is, that the POSIX strtold uses the current locale to convert a string to long double, whereas BigDecimal.toEngeneeringString does not.
I could think of two solutions:

1. use a java.text.Formatter to format and parse the BigDecimal
2. set the locate thread local to "C" in the natve code with uselocale.
   doing this, it would be adviable to pass the BeigDecimal itself and do toEngeneeringString in the native part.

All other issues I saw I will file as separate issue.

[aploese]

Sorry, this is a duplicate of #51

[aploese]

Fixed this in https://github.com/aploese/jffi/tree/fix-mips (hotfix) currently it creates a new locale "C" for every call.
@headius Move this to the global JNI_OnLoad and JNI_OnUnLoad functions?