#718 OverflowErrors with float types in 8.1.6

[Georg Brandl]

With PyTango 8.1.6 (built without numpy due to bug #704) I have problems with float arguments that have values <= 0. With the PowerSupply server example from the quick tour doc, and this client:

import PyTango
dp = PyTango.DeviceProxy('test/test/test')
dp.current = 1
print dp.current
dp.current = 0
print dp.current

I get an OverflowError on the second "print dp.current":

Traceback (most recent call last):
  File "client.py", line 5, in <module>
    dp.current = 0
  File "/usr/lib64/python2.7/site-packages/PyTango/device_proxy.py", line 223, in __DeviceProxy__setattr
    self.write_attribute(name, value)
  File "/usr/lib64/python2.7/site-packages/PyTango/green.py", line 165, in greener
    ret = submit(fn, self, *args, **kwargs)
  File "/usr/lib64/python2.7/site-packages/PyTango/green.py", line 81, in synch_submit
    return get_synch_executor().submit(fn, *args, **kwargs)
  File "/usr/lib64/python2.7/site-packages/PyTango/green.py", line 73, in submit
    return fn(*args, **kwargs)
  File "/usr/lib64/python2.7/site-packages/PyTango/device_proxy.py", line 981, in __DeviceProxy__write_attribute
    return self._write_attribute(*args, **kwargs)
OverflowError: Value is too small.

The same error also appears when calling commands that take or return float or float array arguments. It does not appear when building with numpy.

[Georg Brandl]

Looking at the source of the exception, I noticed that in src/boost/cpp/fast_from_py.h, in the definition of DEFINE_FAST_TANGO_FROMPY_NUM there is a difference between the Numpy and non-Numpy cases: the non-Numpy case does not check for TangoScalarTypeLimits::is_integer.

[Vincent Michel]

I can confirm this bug (same version).
Here is a simple device server to reproduce it:

1 2 3 4 5 6 7 8 9 10 11 12 13

#!/usr/bin/python from PyTango.server import Device, DeviceMeta, attribute class Dummy(Device): __metaclass__ = DeviceMeta Test = attribute(dtype=float) def read_Test(self): return -0.1 if __name__ == "__main__": PyTango.server.run((Dummy,))

[Tiago Coutinho]

Confirmed.

When not using numpy, the range check is done for double types.

Unfortunately, the std::numeric_limits<t>::min() returns the minimum integer when T=int/long but in case of float/double returns the smallest positive.</t>

The test is now similar to what is done in case of having numpy and the problem should be fixed.
Will become visible in next PyTango 8.1.7

Thanks for reporting

[Tiago Coutinho]

Just one more thing:

Bug #704 should now be fixed so you when you build next PyTango 8.1.7 I really recommend enabling numpy.

[Georg Brandl]

Thanks!