fix overflow bug in prevpow

[oscardssmith]

Previously prevpow(10, 1234567890123456789) gave -8446744073709551616 due to integer overflow.

[vchuravy]

Needs a test

[simeonschaub]

I think this is still not fully correct since x isn't necessarily the same type as a, so x could be greater than typemax(a) which means that p*a overflowing isn't necessarily sufficient here. So perhaps we should still promote a and x first?

[oscardssmith]

That would be a somewhat breaking change since it would change the return type.

[oscardssmith]

TLDR is I think that the correct behavior here is either this or to throw an error, which we can't do easily (since there is no Base.checked_pow). As such, I think this is what we should merge since it is a strict improvement.

[oscardssmith]

@simeonschaub do you think this is mergable as is?

[KristofferC]

Might a well do a pkgeval?

[KristofferC]

@nanosoldier runtests(ALL, vs = ":master")

[oscardssmith]

@nanosoldier runtests(ALL, vs = ":master")

[nanosoldier]

Your package evaluation job has completed - possible new issues were detected. A full report can be found here.

[KristofferC]

Looks ok.

[oscardssmith]

Merging monday sans objections