`test_extremes` failing with 64 bit python 3 on Windows about btrees HOT 11 CLOSED

@tseaver: you wrote that test, do you have any opinions on this?

My gut feeling is that this exposes a real bug in OLBTree implementation: it can hold 64-bit values on 32-bit Windows, but not on 64-bit Windows? Most illogical!

from btrees.

This test is skipped for 32 bit architecture https://github.com/zopefoundation/BTrees/blob/master/BTrees/tests/test_OLBTree.py#L112

from btrees.

I believe this error is due to a Windows oddity: long is 32-bit on 64-bit Windows.

from btrees.

Which is maybe the stupidest design choice I can think of: likely due to some crazy notion of preserving ABI with Windows 95. :(

from btrees.

That is weird!

from btrees.

Shall we declare Win64 to be a 32-bit architecture and skip this test too?

from btrees.

What does platform.architecture() return on Win64? [1]

[1] https://docs.python.org/2/library/platform.html#platform.architecture

Maybe we should be using the workaround from that API reference:

def _skip_on_32_bits(wrapped):
    if sys.maxsize == 2**32:
        def _dummy(*args):
            pass
        return _dummy
    return test_method

from btrees.

+1 for the sys.maxsize check, but I'm tempted to make it sys.maxsize <= 2**32 -- just in case someone ports Python to a 16-bit machine ;)

from btrees.

LOL

from btrees.

I think I have the culprit: the longlong_check function in BTrees.BTreeModuleTemplate was not using PyLong_LongLongOverflow (as is used in longlong_convert). See #35 for my attempt at a fix.

from btrees.

Fixed in #36.

from btrees.