tspan argument should be checked to be a proper time interval

[schillic]

Passing tspan=(1., 0.) to solve should throw an error instead of computing a normal-looking but invalid result.

EDIT: According to the documentation, this is actually allowed. But I think at least the TMJets algorithm does not support this properly.

[mforets]

yes, the interval tspan=(1., 0.) was meant to compute backwards approximation.
iirc DiffEq has a similar behavior.