>>>round(1.123456789, 9)
1.123456768

should be 1.123456789.
It rounds wrong.

@turbinenreiter I mentioned this on the forum, but I figured I'd mention it here as well.

• You've got some build failures - click on the red x to see the full log. I think you need to add some casts to convert doubles to mp_float_t
• I'd recommend squashing this into a single commit
• The PR title should have a py: at the front

I think I got the casts figured out, it builds on all platforms, but the test fails, because it rounds wrong in some cases.

round(1.123456789, 8) rounds to 1.1234568, but should be ``1.12345679`

I'll try to figure that out and come back with a better pull request.

It works up until 7 decimals, everything bigger than that gives wrong values.

@turbinenreiter That's because the FPU on the pyboard is only 32-bit, so you can only get 6 significant digits.

So I'd remove that test and call it a day :)

@turbinenreiter: Pretty good for the first patch. Few issues: it should not call round or roundf and friends directly, we have a special macro which use function appropriate for a particular port. Rounding by multiplying to power of 10 and then dividing can lead to overflow. It's ok for a first implementation, but there should be "// TODO: Mat lead to overflow" comment.

As for tests, just use "lowest common denominator" for all ports + fuzz factor, which is rounding up to 5.