[rust-dev] On Copy = POD

Paulo Sérgio Almeida pssalmeida at gmail.com
Sat Jun 21 05:18:38 PDT 2014


It should not need to be said, but I am not advocating (over)using Rc, Gc,
or Arc. One should start by Box<T> and &T. But when this is not enough,
using the other pointer types should not be a pain. More than a pain, I
would not like to have unpleasant surprises when I find out that Box is not
enough, start refactoring code changing Box to, say, Rc, and then finding
out that referents that were cloned when Box was used are now not cloned
anymore, and only the pointer is now "cloned".


On 21 June 2014 11:29, Val Markovic <val at markovic.io> wrote:

> On Fri, Jun 20, 2014 at 11:06 PM, Nick Cameron <lists at ncameron.org> wrote:
>
>> I found all the clone()s in Rust unpleasant, it really put me off using
>> ref counting.
>>
>
> I consider that to be a feature, not a bug.
>
>
>> Given that this is something C++ programmers coming to Rust will be used
>> to using, I believe ergonomics is especially important.
>>
>
> I write C++ for a living in a massive codebase and shared_ptrs are used
> extremely rarely, and it's *not *because of the perf overhead of the
> atomic increment/decrement, but because using shared_ptrs obscures
> ownership. People tend to just put some memory in a shared_ptr and not care
> which part of the system owns what and that ends up producing code that's
> very hard to reason about and maintain.
>
> unique_ptrs have made the transfer of ownership of heap-allocated memory
> super-easy. Damn-nigh every design can be expressed with unique_ptrs owned
> by the logical owners of that memory passing refs or const refs to other
> parts of the system.
>
> So please don't represent that shared_ptrs are commonly used in all good
> C++ code. Experience has thought me and others to look at shared_ptrs as a
> code smell and something to be flagged for extra clarification by the
> author in code review. I hate to quote the Google C++ style guide since it
> has many flaws, but this is one of the things it gets completely right
> <http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml?showone=Ownership_and_Smart_Pointers#Ownership_and_Smart_Pointers>:
> "Do not design your code to use shared ownership without a very good
> reason."
>
> Rust has unique_ptrs in the form of ~ and they're doing their job just
> fine. Rust needs special support for Rc ergonomics as much as it needs such
> support for Gc, which is none at all. In fact, making Rc and Gc pointers
> more difficult to use should steer people away from such poor design
> crutches.
>
>
>>
>> In this case I don't think we need to aim to be more 'bare metal' than
>> C++. Transparent, ref counted pointers in C++ are popular and seem to work
>> pretty well, although obviously not perfectly.
>>
>> zwarich: I haven't thought this through to a great extent, and I don't
>> think here is the right place to plan the API. But, you ought to still have
>> control over whether an Rc pointer is copied or referenced. If you have an
>> Rc<T> object and pass it to a function which takes an Rc<T>, it is copied,
>> if it takes a &Rc<T> or a &T then it references (in the latter case with an
>> autoderef-ref). If the function is parametric over U and takes a &U, then
>> we instantiate U with either Rc<T> or T (in either case it would be passed
>> by ref without an increment, deciding which is not changed by having a copy
>> constructor). If the function takes a U literal, then U must be
>> instantiated with Rc<T>. So, you still get to control whether you reference
>> with an increment or not.
>>
>> I think if Rc is copy, then it is always copied. I would not expect it to
>> ever move. I don't think that is untenable, performance wise, after all it
>> is what everyone is currently doing in C++. I agree the second option seems
>> unpredictable and thus less pleasant.
>>
>> Cheers, Nick
>>
>>
>> On Sat, Jun 21, 2014 at 4:05 PM, Cameron Zwarich <zwarich at mozilla.com>
>> wrote:
>>
>>> I sort of like being forced to use .clone() to clone a ref-counted
>>> value, since it makes the memory accesses and increment more explicit and
>>> forces you to think which functions actually need to take an Rc and which
>>> functions can simply take an &.
>>>
>>> Also, if Rc becomes implicitly copyable, then would it be copied rather
>>> than moved on every use, or would you move it on the last use? The former
>>> seems untenable for performance reasons, since removing unnecessary
>>> ref-count operations is important for performance. The latter seems
>>> unpredictable, since adding a second use of a value in a function would
>>> mean that new code is implicitly executed wherever the first use is.
>>>
>>> Cameron
>>>
>>> On Jun 20, 2014, at 8:49 PM, Nick Cameron <lists at ncameron.org> wrote:
>>>
>>> I think having copy constructors is the only way to get rid of
>>> `.clone()` all over the place when using` Rc`. That, to me, seems very
>>> important (in making smart pointers first class citizens of Rust, without
>>> this, I would rather go back to having @-pointers). The trouble is, I see
>>> incrementing a ref count as the upper bound on the work that should be done
>>> in a copy constructor and I see no way to enforce that.
>>>
>>> So, I guess +1 to spirit of the OP, but no solid proposal for how to do
>>> it.
>>>
>>>
>>> On Sat, Jun 21, 2014 at 8:00 AM, Benjamin Striegel <
>>> ben.striegel at gmail.com> wrote:
>>>
>>>> I'm not a fan of the idea of blessing certain types with a
>>>> compiler-defined whitelist. And if the choice is then between ugly code and
>>>> copy constructors, I'll take ugly code over surprising code.
>>>>
>>>>
>>>> On Fri, Jun 20, 2014 at 3:10 PM, Patrick Walton <pcwalton at mozilla.com>
>>>> wrote:
>>>>
>>>>> On 6/20/14 12:07 PM, Paulo Sérgio Almeida wrote:]
>>>>>
>>>>>  Currently being Copy equates with being Pod. The more time passes and
>>>>>> the more code examples I see, it is amazing the amount of ugliness
>>>>>> that
>>>>>> it causes. I wonder if there is a way out.
>>>>>>
>>>>>
>>>>> Part of the problem is that a lot of library code assumes that Copy
>>>>> types can be copied by just moving bytes around. Having copy constructors
>>>>> would mean that this simplifying assumption would have to change. It's
>>>>> doable, I suppose, but having copy constructors would have a significant
>>>>> downside.
>>>>>
>>>>> Patrick
>>>>>
>>>>> _______________________________________________
>>>>> Rust-dev mailing list
>>>>> Rust-dev at mozilla.org
>>>>> https://mail.mozilla.org/listinfo/rust-dev
>>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Rust-dev mailing list
>>>> Rust-dev at mozilla.org
>>>> https://mail.mozilla.org/listinfo/rust-dev
>>>>
>>>>
>>> _______________________________________________
>>> Rust-dev mailing list
>>> Rust-dev at mozilla.org
>>> https://mail.mozilla.org/listinfo/rust-dev
>>>
>>>
>>>
>>
>> _______________________________________________
>> Rust-dev mailing list
>> Rust-dev at mozilla.org
>> https://mail.mozilla.org/listinfo/rust-dev
>>
>>
>
> _______________________________________________
> Rust-dev mailing list
> Rust-dev at mozilla.org
> https://mail.mozilla.org/listinfo/rust-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.mozilla.org/pipermail/rust-dev/attachments/20140621/0a9ff6e1/attachment.html>


More information about the Rust-dev mailing list