Sollya seems to consider that 0 is signed in a limit operation (from gforge #21680)
Imported issue: Initially reported by @chevilla in https://gforge.inria.fr/tracker/?group_id=1015&aid=21680
On the following example, tried with commit 64bc5e0f, Sollya seems to consider that 0 is positive when evaluating a function which has a singularity at 0, at least for some functions:
evaluate(1/x, [0]);
[-infty;infty]
works as expected, but
evaluate(erf(1/x), [0]);
[0.99999999999999999999999999999999999999999999999998;1]
evaluate(atan(1/x), [0]);
[1.57079632679489661923132169163975144209858469968754;1.5707963267948966192313216916397514420985846996876]
evaluate(erf(1/sin(x)),[0]);
[0.99999999999999999999999999999999999999999999999998;1]
all seem to consider that the division by 0 leads to a (signed) positive infinity.
Zero is normally considered unsigned within Sollya, meaning that 1/x at 0 should be considered as {-Inf, +Inf} or, more likely its convex hull, namely [-Inf, +Inf], which, when passed to a function, should return (an overestimate of) its range.