IntelPython / dpnp / 26363031806 / 1

78%

master: 78%

push

github

Resolve gh-2882 (#2930)

This PR adds a fix for gh-2882

When floats were previously converted by `convert_impl` to unsigned
integral types, they would pass through an equally sized, signed
integral type

If the float was in bounds for the unsigned integral type but out of
bounds for the signed type, however, undefined behavior could be
encountered.

This PR proposes a solution of casting to a sufficiently large
intermediate type first, before casting into the unsigned integral type.

1539 of 2874 branches covered (53.55%)

Branch coverage included in aggregate %.

26430 of 32874 relevant lines covered (80.4%)

7901.25 hits per line