Notably changes the behavior of __bool__ to always raise. Before there was a hack based on whether a variable had been compared to something before.

This should avoid surprising behavior such as:

import pytensor.tensor as pt
x = pt.zeros(())
if x:
  print("Variable is nonzero!")  # Always triggered
else:
  print("Variable is zero")  # Never triggered

After this PR it raises

TypeError: TensorVariable cannot be converted to Python boolean. Call `.astype(bool)` for the symbolic equivalent.

The correct behavior as per the API standard is to raise in these cases: https://data-apis.org/array-api/latest/design_topics/lazy_eager.html

There is still surprising behavior with (in)equality checks, but we can't do anything about those as they are needed for Python hashing behavior. The default is x == y: True if x is y else False

import pytensor.tensor as pt
x = pt.zeros(())
if x == 0:
  print("Variable is zero!")  # Only triggered when x == x
else:
  print("Variable is nonzero")  # Always triggered

Related to #1189

📚 Documentation preview 📚: https://pytensor--1190.org.readthedocs.build/en/1190/