[swift-evolution] [Discussion] Breaking precedence
Xiaodi Wu
xiaodi.wu at gmail.com
Tue Aug 2 16:46:55 CDT 2016
It's not that << will overflow and / will not. Substitute * for / and the
argument would be the same. The difference is that << traps when you shift
more than the total number of bits but does *not* trap when you shift
numbers off as would arithmetic exponentiation; * traps on overflow. Thus,
what << is concerned about is the bits (as it should), but * is concerned
about the max representable value.
On Tue, Aug 2, 2016 at 16:41 Anton Zhilin <antonyzhilin at gmail.com> wrote:
> Here is the proposal draft:
>
> https://github.com/Anton3/swift-evolution/blob/remove-precedence-bitwise-arithmetic.md/proposals/NNNN-remove-precedence-bitwise-arithmetic.md
>
> Motivation will be the most difficult section here.
>
> 2016-08-02 23:30 GMT+03:00 Xiaodi Wu <xiaodi.wu at gmail.com>:
>
>> Let me give you theoretical basis for why I'm +0.5 on branching off
>> binary operators but not these other ones. FĂ©lix is absolutely right that
>> `a << b / c` mixes two things. It's not merely that they're in two
>> "different" domains. It's that these two operators take the same values of
>> the same type and operate on them in fundamentally disparate ways.
>>
>> This is a bad way of phrasing it, I know--let me try to clarify: the
>> operator `<<` operates on an integer as a fixed-length collection of bits;
>> the operator `/` operates on an integer as a number, an element in the set
>> of all integers. The practical consequence is that overflow behavior can be
>> subtly different; the overflow behavior of << is 'obvious' if you're
>> thinking about an integer as a fixed-length collection of bits but
>> surprising if you think of it as an integer being multiplied by an exponent
>> of two. Thus, it is best to separate operators that work on integers as a
>> collection of bits from the other numeric operators.
>>
>> In no other of your proposed branches do I find the same fundamental
>> conflict.
>>
>
> And that argumentation seems insufficient to me. Assuming that everyone
> knows precedence table, 'a << b / c` should not be ambiguous. I'd argue
> that it's easy to understand that << operation will overflow here and /
> will not.
> Instead, I suggest that these operations are so different that any
> precedence relationship between them is meaningless (another wording of
> explanation with domains).
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20160802/76fa1e11/attachment.html>
More information about the swift-evolution
mailing list