Improve array conversion from Java & add a test for nan and inf values from Java

I couldn't reproduce the reported issue @rdemaria, and have added tests to confirm that it is doing as expected.

Just in case, I also went through the whole Japc machinery with the following test, and also couldn't find the issue:

def test_get_value_double(japc, japc_mock):
    param = 'TEST/TestProperty'
    mock = japc_mock.mockParameter(param)
    japc_mock.whenGetValueThen(
        mock, japc_mock.sel('LHC.USER.TEST'), japc_mock.acqVal(param, np.inf, 0))
    japc.setSelector('LHC.USER.TEST')
    assert np.isinf(japc.getParam('TEST/TestProperty'))

and:

def test_get_value_double(japc, japc_mock):
    param = 'TEST/TestProperty'
    mock = japc_mock.mockParameter(param)
    japc_mock.whenGetValueThen(
        mock, japc_mock.sel('LHC.USER.TEST'), japc_mock.acqVal(param, np.nan, 0))
    japc.setSelector('LHC.USER.TEST')
    assert np.isnan(japc.getParam('TEST/TestProperty'))

I also confirmed this with both float('nan') and float('inf').

is it possible that Fesa defines infinity for integer types?

is it possible that Fesa defines infinity for integer types?

I've found the thread about the original issue (or another issue perhaps), and it seems like something was fixed in JAPC itself. I'll quickly try to run an older version of JAPC to see if the tests above actually fail

added 1 commit

• bb4d582f - Improve the conversion from java to numpy dtypes for arrays. This is a...

Compare with previous version

added 1 commit

• bd45789a - Improve the conversion from java to numpy dtypes for arrays. This is a...

Compare with previous version

added 1 commit

• 365cc98e - Add the benchmarks I used to determine *how* slow the supposed slowdown would...

Compare with previous version

added 1 commit

• 922d2fef - Add the benchmarks I used to determine *how* slow the supposed slowdown would...

Compare with previous version

This MR has morphed a little bit following my investigation. It turns out that float32 values were being converted to float64 in all cases. Under the hood, the JAPC change which fixed the problem, I believe was https://gitlab.cern.ch/acc-co/japc/japc-core/-/compare/japc-value%2Fv7.8.5...japc-value%2Fv7.8.6#e0e59828d4009c8fd6da59e221bc1d986c21e971_1897_1895, but I confess I cannot get the exception to raise in PyJapc using the old version.

In order to fix the further type conversion, I have removed a bunch of code which was an optimisation based on how the old JPype worked. This is now far simpler, and actually results in a faster conversion (benchmark test added).

changed title from Add a test for nan and inf values from Java to Improve array conversion from Java & add a test for nan and inf values from Java

merged

mentioned in commit 715d0df5