corr.m_eff() does not check for division by zero about pyerrors HOT 6 CLOSED

I last change to Corr.m_eff appears to have happened 5 months ago, what version did you use before and can you provide a minimal example for the error you are seeing?

from pyerrors.

I do not know, which version the script was previously running on. I think it was the first 2.0 release.
I have a correlator, which has values up to a certain timeslice and is then set to zero.
So you can see, where the problem occurs.
The error is:

m_eff=corr.m_eff(variant="cosh")
File "/usr/local/lib/python3.8/site-packages/pyerrors/correlators.py", line 631, in m_eff
newcontent.append(np.abs(find_root(self.content[t][0] / self.content[t + 1][0], root_function, guess=guess)))
File "/usr/local/lib/python3.8/site-packages/pyerrors/obs.py", line 744, in truediv
return derived_observable(lambda x, **kwargs: x[0] / x[1], [self, y], man_grad=[1 / y.value, - self.value / y.value ** 2])
ZeroDivisionError: float division by zero

from pyerrors.

Would you mind, if i change this?

from pyerrors.

Ah okay, I understand the problem, and there should be an easy fix, but I do not understand why it did not appear in earlier versions. I think we never explicitly checked whether the denominator is zero.

from pyerrors.

Feel free to fix this. A general way including other kind of arithmetic errors might be to check for an ArithmeticError

try:
    newcontent.append(self.content[t] / self.content[t + 1])
except ArithmeticError:
    newcontent.append(None)

from pyerrors.

Yes, you are right, we only checked for None not for zero. So maybe something that i did to the correlator with the previous version set it to None on these timeslices, but this behaviour changed. We might never know.

from pyerrors.