Fix net.openhft.chronicle.salt.EasyBoxTest

[RobAustin]

please fix this test, net.openhft.chronicle.salt.EasyBoxTest, for now, it has been ignored because it fails on TeamCity, see http://teamcity.higherfrequencytrading.com/viewLog.html?buildId=309860&buildTypeId=OpenHFT_ChronicleSalt_Snapshot&tab=buildLog&_focus=14105

[RobAustin]

we updated EasyBox to use sodium 1.0.16 but this test is still failing, It works for @peter-lawrey on his laptop.

[rogersimmons]

This appears to be an issue with the current version of JNR FFI - specifically, it is corrupting the 6th argument in the crypto_box_(open)_easy calls (the secret key). These are the only functions in the Sodium interface which take so many arguments, hence why only the EasyBox tests failing.

A stand-alone C version running on the same libsodium version works fine.
Intercepting the JNR-FFI call with a manual C lib/function exposing the same interface shows the corruption of the 6th arg. Interestingly the corruption is of a constant form - looks like the last 4 bytes or so of the 6th argument (which is a pointer/address) is being clobbered with a constant value.

Continuing to look into options. This may need a bug report raised to JNR-FFI.

[rogersimmons]

Confirmed. Amending the Java code to wrap all the addresses into a long[] array, and just forwarding this one argument via JNR FFI through an intermediate C .so which unpacks and calls on to sodium works fine (EasyBox tests pass). Using this as a stopgap for now while a fix from JNR FFI is pending.

[rogersimmons]

This is now fixed.
The 2 problematic EasyBox calls which are being corrupted by JNR-FFI has been switched to JNI for now following the same model as Chronicle-Ticker

[rogersimmons]

JNR-FFI issue raised:
jnr/jnr-ffi#181

[hft-team-city]

Released in Chronicle-Salt-2.17.2, BOM-2.17.119

[hft-team-city]

Released in Chronicle-Salt-2.17.2, BOM-2.17.196

[hft-team-city]

Released in Chronicle-Salt-2.17.2, BOM-2.17.225

[hft-team-city]

Released in Chronicle-Salt-2.17.2, BOM-2.17.241