Armin Rigo <[email protected]> added the comment: My point is that on CPython, programs that do str() on the result and are unhappy with the trailing L are broken --- because they will get a trailing L on some inputs. Ryu's regression tests are possibly an exception, in that they are not broken by relying explicitly on the absence of a trailing L, for a specific input value chosen for the test.
We have to decide if it's worth taxing the jitted performance of all programs that use unpack() by returning mixed types, in order to support this very special compatibility requirement; or if we say that Ryu should just fix its test to not rely on this detail (similar to how you need to add "gc.collect()" in some tests that verify that __del__() was called). If you have another example where the problem is not in a test, I'd be interested. :-) ________________________________________ PyPy bug tracker <[email protected]> <https://bugs.pypy.org/issue1738> ________________________________________ _______________________________________________ pypy-issue mailing list [email protected] https://mail.python.org/mailman/listinfo/pypy-issue
