Non-constant numbers in Boolean context are parsed to constants #581

sympify_petab gets the result wrong when a non-constant numeric expression is used as a Boolean, i.e. as an operand of &&, ||, !, or as a piecewise condition. The PEtab v2 spec ("Boolean ↔ float conversion") says 0.0 is converted to false and any other value to true. Instead, the operand is always treated as true. This happens with both evaluate=True and evaluate=False:

from petab.v1.math import sympify_petab

sympify_petab("(x + 1) && (y > 2)")      # y > 2.0         (expected: (x + 1 != 0) && (y > 2))
sympify_petab("!(x + 1)")                # 0.0             (expected: x + 1 == 0)
sympify_petab("piecewise(1, x + 1, 0)")  # 1.0             (expected: 1 if x + 1 != 0 else 0)
sympify_petab("piecewise(1, x > 0, 0) && true")  # 1.0     (expected: x > 0)

The cause is the fallback in num2bool (petab/v1/math/SympyVisitor.py), sp.Piecewise((True, x != 0.0), (False, True)). There, x != 0.0 is Python's structural inequality, not sp.Ne, so it evaluates to True for any symbolic x.

In the same function, a bare symbol is passed through unchanged, because sp.Symbol is a subclass of sp.logic.boolalg.Boolean. As a result, x && true is parsed to x instead of x != 0, and x && y is parsed to And(x, y), which fails once numbers are substituted. Boolean-valued piecewise operands such as !piecewise(true, x > 0, false) are not handled correctly either.

petab.v2.math re-exports petab.v1.math, so both versions are affected.

The same bug in sbmlmath: dweindl/sbmlmath#52

🤖 Generated with AI assistance (Claude Code)