[vtkusers] Allowable C++11 Features in VTK

[Andrew Maclean]

I agree with David.

I would argue for strongly typed enums to be allowed. In contrast to
ordinary enums, they don't export their enumerators to the surrounding
scope and they don't have implicit conversion to int. Implicit conversion
to int can dangerous in that the code performs correctly but the results
are unexpected. I don't think their use should be restricted to just global
enumerations.

I actually had the experience of converting old code to C++11 and when I
converted the enums to strongly typed enums a whole lot of subtle faults
were revealed where ints were used instead of enums. The important thing
was that the old code ran perfectly but it was nearly impossible to account
for the anomalous behaviour until strongly typed enums were used.


Here is a contrived example that illustrates the problem.
#include <cstdlib>
#include <iostream>
#include <iomanip>

enum bg1 {r, g, b};

enum class bg2 { r, g, b };

void c0(int intensity)
{
std::cout << "Intensity: " << intensity << "\n";
}

void c1(bg1 bg, int intensity)
{
std::cout << "Background: ";
switch (bg)
{
case r: std::cout << "r\n";   break;
case g: std::cout << "gr\n"; break;
case b: std::cout << "b\n";  break;
}
std::cout << "Intensity: " << intensity << "\n";
}

void c2(bg2 bg, int intensity)
{
std::cout << "Background: ";
switch (bg)
{
case bg2::r: std::cout << "r\n";   break;
case bg2::g: std::cout << "g\n"; break;
case bg2::b: std::cout << "b\n";  break;
}
std::cout << "Intensity: " << intensity << "\n";
}

int main()
{
int intensity = 7;
c0(intensity);
c0(bg1::r); // OK but surely an error!
c1(bg1::r, intensity);
// This call is Ok as bg is implicitly converted to int.
// Is this what we really want?
c1(bg1::r, bg1::g);
// Strongly typed case.
c2(bg2::r, intensity);
// Because of strongly typed enums,
// these will fail as bg2 is strongly typed.
// c0(bg2::r);
// c2(bg2::r, bg2::g);

return EXIT_SUCCESS;
}

Regards
   Andrew

---------- Forwarded message ----------
> From: David Gobbi <david.gobbi at gmail.com>
> To: Robert Maynard <robert.maynard at kitware.com>
> Cc: VTK Developers <vtk-developers at vtk.org>, vtk vtk <vtkusers at vtk.org>
> Bcc:
> Date: Mon, 27 Mar 2017 15:22:29 -0600
> Subject: Re: [vtk-developers] [vtkusers] Allowable C++11 Features in VTK
> The fixed-type enums in C++11 could also be beneficial.  Consider the
> following vtkLookupTable code, which declares a few constants in the header
> (shown below) and then provides the definition in the .cxx file (not shown):
>
> class vtkLookupTable
> {
> public:
>   static const vtkIdType BELOW_RANGE_COLOR_INDEX;
>   static const vtkIdType ABOVE_RANGE_COLOR_INDEX;
>   static const vtkIdType NAN_COLOR_INDEX;
>   static const vtkIdType NUMBER_OF_SPECIAL_COLORS;
> ...
> };
>
> In C++11 these "static const" members can be defined more efficiently as a
> fixed-type enum:
>
> public:
>   enum : vtkIdType {
>     BELOW_RANGE_COLOR_INDEX = 0,
>     ABOVE_RANGE_COLOR_INDEX = 1,
>     NAN_COLOR_INDEX = 2,
>     NUMBER_OF_SPECIAL_COLORS
>   };
>
> Of course VTK already uses enums for most constants (except for the ones
> that are still #define'd), so the benefit is for when you need a guarantee
> that the constant will be evaluated as a specific integral type.
>
>  - David
>
>
>
> On Mon, Mar 27, 2017 at 2:14 PM, Robert Maynard <
> robert.maynard at kitware.com> wrote:
>
>> Hi,
>>
>> Thanks for the feedback.
>>
>> 1. I agree that when doing template metaprogramming, the alias keyword is
>> significantly better than typedef. I will update the document to state that
>> we prefer alias over typedef
>>
>> 2. I am going to update the std::array section to clarify that is
>> preferred over raw fixed size 'C' arrays.
>>
>> 3. This is good information to have. I will add a comment to the scoped
>> enums section while I wait for more feedback
>>
>>
>> On Mon, Mar 27, 2017 at 3:30 PM, B B <baljci at hotmail.com> wrote:
>>
>>> I would suggest the following:
>>>
>>>
>>> - Prefer the use of alias declarations instead of typedef: they do the
>>> same but alias declarations are better because they can be templetized (see
>>> Scott Meyers, Effective Modern C++, 63-67)
>>>
>>> - Replace raw arrays with std::array when possible: they are
>>> copyable, easier to manipulate and have no extra performance costs compared
>>> to raw arrays.
>>>
>>> - For scoped enums, I would restrict their use to global enums,
>>> especially in case of possible name conflicts. For nested enums, I would
>>> suggest to maintain the use of unscoped enums for two reasons: first, you
>>> don't need to write MyEnum::MyEnumValue each time you use them inside
>>> the class implementation; second, from my own experience, their implicit
>>> conversion to int can be useful in many cases.
>>>
>>> Boris
>>>
>>> ------------------------------
>>> *De :* vtkusers <vtkusers-bounces at vtk.org> de la part de Robert Maynard
>>> <robert.maynard at kitware.com>
>>> *Envoyé :* lundi 27 mars 2017 20:03
>>> *À :* VTK Developers; vtk vtk
>>> *Objet :* [vtkusers] Allowable C++11 Features in VTK
>>>
>>> As everyone is aware over the past couple of months we have updated
>>> VTK to require a C++11 compiler, but have not explicitly stated what
>>> C++11 features are usable.
>>>
>>> We do not intend to incorporate all features of the language at this
>>> time because of incompatibilities with the structure of VTK and/or
>>> incomplete support for the features by all of the compilers that VTK
>>> aims to support.
>>>
>>> The current proposed C++11 features, and where they are allowed can be
>>> found at:
>>>
>>> https://docs.google.com/document/d/1h7wIq25d-qimQO8N9sE43fHX
>>> KKlHM2sW2ErohfHiuCg/edit?usp=sharing
>>>
>>> Over the next two weeks please provide feedback, either by commenting
>>> on the google document, or replying on the mailing list. Once the two
>>> weeks are over, we will integrate the result into the existing coding
>>> documentation, and then allow C++11 to be used.
>>>
>>
>
> _______________________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at http://www.kitware.com/opensou
> rce/opensource.html
>
> Search the list archives at: http://markmail.org/search/?q=vtk-developers
>
> Follow this link to subscribe/unsubscribe:
> http://public.kitware.com/mailman/listinfo/vtk-developers
>


-- 
___________________________________________
Andrew J. P. Maclean

___________________________________________
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/vtkusers/attachments/20170329/8c220abc/attachment.html>