QUERY · ISSUE

MICROPY_PY_BUILTINS_ROUND_INT==0 breaks round(<int>, <n>)

openby dhalbertopened 2019-05-16updated 2019-08-20

Before https://github.com/micropython/micropython/commit/b318ebf1015ced6354f8bbaf035308214b3f5c5d, which was a rework of #3557, round(<int>, <n>) worked. Now, if MICROPY_PY_BUILTINS_ROUND_INT is set to 0, round(2, 2) or similar throws NotImplementedError. round(2) still works.

@dpgeorge Was it your intention to disable rounding of integers in this way completely, instead of just not handling negative values for the second arg if it was an int?

A user discovered this after ending up with an integer zero while doing some floating arithmetic.

CANDIDATE · ISSUE

round(x, -1) is not implemented for integers, whereas it is in CPython

closedby stroobandtopened 2018-01-03updated 2018-05-22

round(x, -1) is not implemented for integers, whereas it is both in CPython and for floats in MicroPython.

MicroPython v1.9.2-87-gda8c4c26 on 2017-09-13; PYBv1.1 with STM32F405RG
>>> x = 98836
>>> round(x, -1)
98836
>>> x = 98836.
>>> round(x, -1)
98840.0

In CPython:

In [1]: x = 98836
In [2]: round(x, -2)
Out[2]: 98800

Some more information: https://stackoverflow.com/a/3349102/2192488

Assessment

S Sonnet · high

Both issues involve round() with integer arguments and a second precision parameter in py/builtins.c. However, they describe distinct problems: #3538 reports that round(int, -n) returns the original value unchanged instead of rounding (wrong result, no error), while #4792 reports that round(int, +n) raises NotImplementedError when MICROPY_PY_BUILTINS_ROUND_INT==0 (regression introduced by the commit that attempted to fix #3538's underlying problem, #3557). #4792 explicitly references commit b318ebf which was a rework of #3557 — meaning #3538 is the upstream cause of the regression in #4792, but neither issue resolves the other. They share the same affected function and code area but describe different failure modes.

Suggested action

No closure warranted. These are distinct bugs in the same code area. When fixing #4792 (the regression), revisit whether #3538 (negative precision for integers) is also addressed, as both stem from the same round() integer handling logic. Link the issues for cross-reference.

MICROPY_PY_BUILTINS_ROUND_INT==0 breaks round(<int>, <n>)

round(x, -1) is not implemented for integers, whereas it is in CPython

Keyboard