Cannot handle Float correctly on Windows

[SaiCateDoan]

pylivemaker/src/livemaker/lpb.py

Line 174 in ceeb129

lambda obj, ctx: numpy.frombuffer(obj.rjust(16, b"\x00"), dtype=numpy.longdouble),

Two reasons:

1. numpy.longdouble is alias of float64 on Windows, but not float96 or float128, so I run it on WSL
2. numpy.frombuffer returns ndarray, but not a particular number, so I just patched it

I think this might be work, but I only test the function of dump file, so you know...

Test file: 0000001C.zip

[pmrowla]

Thanks for the sample LSB. I need to revisit the float/double handling since the 80-bit extended precision doubles are also unavailable on ARM64 (apple silicon macs). This probably means handling the 80-bit serialization ourselves and accepting some loss of precision, but I don't have time to work on this at the moment.

In WSL on an x86 CPU, your patch should be fine as a temporary fix. If you also need writing/editing the LSB, it looks like there's another bug where tobytes() should be called using ndarray. You may also want to explicitly change numpy.longdouble to numpy.float96, although I think it should work either way on WSL or native x86 Linux.

So an updated patch to handle both dump + edit would look something like:

"Float": construct.ExprAdapter(
    construct.Bytes(10),
    lambda obj, ctx: numpy.frombuffer(obj.rjust(16, b"\x00"), dtype=numpy.float96, count=1)[0],
    lambda obj, ctx: numpy.array([obj], dtype=numpy.float96).tobytes()[-10:],
),