Jump to content

  • Log In with Google      Sign In   
  • Create Account

A Proposal to Add Strong Type Aliases to the Standard Language


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
29 replies to this topic

#1 Servant of the Lord   Crossbones+   -  Reputation: 21046

Like
0Likes
Like

Posted 28 December 2012 - 12:48 PM

Hey, I want strong type aliases in C++. I've never written a standard proposal before, so here's my attempt.

 

View PDF online

Download: Word document (20.7kb)

 

Any suggestions, additions, rewordings, and etc... you are able to offer would be much appreciated. After the community reviews and improves it, I'll email it to the standards committee (which are currently accepting proposals for C++ TR2), so it can hopefully be read and discussed at a committee meeting for addition to the C++ standard.

 

Please download the document, annote it in red, and re-post it here. Alternatively, post suggestions in the thread itself.

 

I would really appreciate it! smile.png

Aside from non-static data member initializers, which was added into C++11 (thank you, whoever submitted that proposal!), strong typedefs are one of my most desired C++ features.


Edited by Servant of the Lord, 28 December 2012 - 12:49 PM.

It's perfectly fine to abbreviate my username to 'Servant' rather than copy+pasting it all the time.
All glory be to the Man at the right hand... On David's throne the King will reign, and the Government will rest upon His shoulders. All the earth will see the salvation of God.
Of Stranger Flames - [indie turn-based rpg set in a para-historical French colony] | Indie RPG development journal

[Fly with me on Twitter] [Google+] [My broken website]

[Need web hosting? I personally like A Small Orange]


Sponsor:

#2 santa01   Members   -  Reputation: 307

Like
0Likes
Like

Posted 28 December 2012 - 01:26 PM

Hm... Can I ask why this feature is vital? Conversion between

typedef unsigned int Centimeters;
typedef unsigned int Inches;

are legal as for me, as the real types are the same.

 

For incompatible types gcc (at least) drops an error:

╓─[santa@yukio ~ $]
║
╙─> g++ test.cpp -o test -pedantic -Wall
test.cpp: In function ‘int main()’:
test.cpp:9:8: error: cannot convert ‘MyFloat {aka float}’ to ‘MyInt {aka void*}’ for argument ‘1’ to ‘void test(MyInt)’
╓─[santa@yukio ~ $]
║
╙─> cat test.cpp 
typedef void* MyInt;
typedef float MyFloat;

void test(MyInt) {
}

int main() {
        MyFloat f;
        test(f);
        return 0;
}

For compatible types you won't get an error even without typedefs:

╓─[santa@yukio ~ $]
║
╙─> g++ test.cpp -o test -pedantic -Wall
test.cpp: In function ‘int main()’:
test.cpp:9:8: warning: ‘f’ is used uninitialized in this function [-Wuninitialized]
╓─[santa@yukio ~ $]
║
╙─> cat test.cpp 
void test(int) {
}

int main() {
        float f;
        test(f);
        return 0;
}


#3 SiCrane   Moderators   -  Reputation: 9670

Like
5Likes
Like

Posted 28 December 2012 - 01:29 PM

You probably should read and reference previous opaque typedef proposals such as N1706 and N1891.

#4 Servant of the Lord   Crossbones+   -  Reputation: 21046

Like
3Likes
Like

Posted 28 December 2012 - 01:40 PM

Thanks SiCrane, I assumed there were prior proposals but didn't find them. I'll read over them.
Hm... Can I ask why this feature is vital? Conversion between
typedef unsigned int Centimeters;
typedef unsigned int Inches;
are legal as for me, as the real types are the same.
 
For incompatible types gcc (at least) drops an error: ...

For compatible types you won't get an error even without typedefs: ...
 



That's precisely the point. I don't want Centimeters and Inches to be convertible, except explicitly. Strong aliases should refuse to compile aliases made even from the same base type. You need a function to convert from Centimeters to Inches, because 1 Centimeter is not 1 Inch, and doing myCentimeters = myInches is easy to do and almost always a bug.

Edited by Servant of the Lord, 28 December 2012 - 01:42 PM.

It's perfectly fine to abbreviate my username to 'Servant' rather than copy+pasting it all the time.
All glory be to the Man at the right hand... On David's throne the King will reign, and the Government will rest upon His shoulders. All the earth will see the salvation of God.
Of Stranger Flames - [indie turn-based rpg set in a para-historical French colony] | Indie RPG development journal

[Fly with me on Twitter] [Google+] [My broken website]

[Need web hosting? I personally like A Small Orange]


#5 santa01   Members   -  Reputation: 307

Like
0Likes
Like

Posted 28 December 2012 - 02:01 PM

I don't want Centimeters and Inches to be convertible

Well and I don't understand why. As I wrote earlier built-in types with similar (in fact equal) semantics _are_ convertible. In your example you are trying to introduce a new type (not an alias in fact, alias is usually just another typename). Simple constructions such as `fancy_typedef old_type new_type' aren't able to describe the type semantics at all (both Centimeters and Inches are just lexems for compiler, you don't define what can be done to them). If you suggest semantics should be copied from the old_type than new_type would be just a typename, not a new type.



#6 Álvaro   Crossbones+   -  Reputation: 13913

Like
3Likes
Like

Posted 28 December 2012 - 02:05 PM


I don't want Centimeters and Inches to be convertible

 
Well and I don't understand why. As I wrote earlier built-in types with similar (in fact equal) semantics _are_ convertible.



You seriously don't understand why it's desirable to get a compiler error if someone has a quantity in centimeters and tries to use it where a quantity in inches is expected? I am not sure how else to explain it, since SOTL has been very clear.

#7 santa01   Members   -  Reputation: 307

Like
0Likes
Like

Posted 28 December 2012 - 02:42 PM

You seriously don't understand why it's desirable to get a compiler error if someone has a quantity in centimeters and tries to use it where a quantity in inches is expected?

This is not my point. My point is thats not a compiler (c++ grammar) problem. Its worth implementing as a part of an STL.



#8 Álvaro   Crossbones+   -  Reputation: 13913

Like
0Likes
Like

Posted 28 December 2012 - 02:56 PM


You seriously don't understand why it's desirable to get a compiler error if someone has a quantity in centimeters and tries to use it where a quantity in inches is expected?

 
This is not my point. My point is thats not a compiler (c++ grammar) problem. Its worth implementing as a part of an STL.



Well, it's a problem where the compiler could help, if it implemented what we are discussing.

I have felt the need for something like this when I have classes Vector3D and Point3D, which are essentially the same thing, but I need to define them separately if I want the type system to help me make sure my operations make sense (e.g., you are not allowed to add points, but adding vectors is fine, and so is adding a point and a vector).

Since we are dealing with units, I just wrote this little test that seems to work fine:
#include <iostream>

namespace units {
  template <int m_pow, int kg_pow, int s_pow>
  struct unit {
    double value;
    explicit unit(double value) : value(value) {
    }
  };
  
  template <int m_pow, int kg_pow, int s_pow>
  unit<m_pow,kg_pow,s_pow> operator+(unit<m_pow,kg_pow,s_pow> u1, unit<m_pow,kg_pow,s_pow> u2) {
    return unit<m_pow,kg_pow,s_pow>(u1.value+u2.value);
  }
  
  unit<0,0,0> operator+(unit<0,0,0> u, double d) {
    return unit<0,0,0>(u.value+d);
  }
  
  unit<0,0,0> operator+(double d, unit<0,0,0> u) {
    return unit<0,0,0>(d+u.value);
  }
  
  template <int m_pow, int kg_pow, int s_pow>
  unit<m_pow,kg_pow,s_pow> operator-(unit<m_pow,kg_pow,s_pow> u1, unit<m_pow,kg_pow,s_pow> u2) {
    return unit<m_pow,kg_pow,s_pow>(u1.value-u2.value);
  }

  unit<0,0,0> operator-(unit<0,0,0> u, double d) {
    return unit<0,0,0>(u.value-d);
  }

  unit<0,0,0> operator-(double d, unit<0,0,0> u) {
    return unit<0,0,0>(d-u.value);
  }

  template <int m_pow1, int kg_pow1, int s_pow1, int m_pow2, int kg_pow2, int s_pow2>
  unit<m_pow1+m_pow2,kg_pow1+kg_pow2,s_pow1+s_pow2> operator*(unit<m_pow1,kg_pow1,s_pow1> u1, unit<m_pow2,kg_pow2,s_pow2> u2) {
    return unit<m_pow1+m_pow2,kg_pow1+kg_pow2,s_pow1+s_pow2>(u1.value*u2.value);
  }

  template <int m_pow, int kg_pow, int s_pow>
  unit<m_pow,kg_pow,s_pow> operator*(unit<m_pow,kg_pow,s_pow> u, double d) {
    return unit<m_pow,kg_pow,s_pow>(u.value*d);
  }

  template <int m_pow, int kg_pow, int s_pow>
  unit<m_pow,kg_pow,s_pow> operator*(double d, unit<m_pow,kg_pow,s_pow> u) {
    return unit<m_pow,kg_pow,s_pow>(d*u.value);
  }

  template <int m_pow1, int kg_pow1, int s_pow1, int m_pow2, int kg_pow2, int s_pow2>
  unit<m_pow1-m_pow2,kg_pow1-kg_pow2,s_pow1-s_pow2> operator/(unit<m_pow1,kg_pow1,s_pow1> u1, unit<m_pow2,kg_pow2,s_pow2> u2) {
    return unit<m_pow1-m_pow2,kg_pow1-kg_pow2,s_pow1-s_pow2>(u1.value/u2.value);
  }

  template <int m_pow, int kg_pow, int s_pow>
  unit<m_pow,kg_pow,s_pow> operator/(unit<m_pow,kg_pow,s_pow> u, double d) {
    return unit<m_pow,kg_pow,s_pow>(u.value/d);
  }

  template <int m_pow, int kg_pow, int s_pow>
  unit<m_pow,kg_pow,s_pow> operator/(double d, unit<m_pow,kg_pow,s_pow> u) {
    return unit<-m_pow,-kg_pow,-s_pow>(d/u.value);
  }

  template <int m_pow, int kg_pow, int s_pow>
  std::ostream &operator<<(std::ostream &os, unit<m_pow,kg_pow,s_pow> u) {  
    os << u.value;
    if (m_pow != 0) {
      os << "m";
      if (m_pow != 1)
	os << "^" << m_pow;
    }
    if (kg_pow != 0) {
      os << "Kg";
      if (kg_pow != 1)
	os << "^" << kg_pow;
    }
    if (s_pow != 0) {
      os << "s";
      if (s_pow != 1)
	os << "^" << s_pow;
    }
    return os;
  }

  unit<1,0,0> meter(1);
  unit<1,0,0> centimeter(0.01);
  unit<1,0,0> inch(0.0254);
  unit<0,0,1> second(1.0);
  unit<0,0,1> minute(60.0);
  unit<0,0,1> hour(3600.0);
  unit<0,1,0> gram(0.001);
  // etc.
}

using namespace units;

int main() {
  std::cout << "100 inches/hour = " << (100.0*inch/hour)/(centimeter/minute) << " centimeters/minute\n";
}

Does anyone know if there is a C++ library that does this type of thing? (Oh, and sorry about the hijack...)

#9 SiCrane   Moderators   -  Reputation: 9670

Like
1Likes
Like

Posted 28 December 2012 - 03:05 PM

boost::units seems to do what you want.

#10 Cornstalks   Crossbones+   -  Reputation: 6991

Like
0Likes
Like

Posted 28 December 2012 - 03:06 PM

It would be cool if you could define casting operators for opaque typedefs too*:

// Obviously this would require a few changes to the Standard
Inches operator Centimeters() (Centimeters cm)
{
    // Estimate
    return (cm * 5) / 2;
    // Note that the above is merely example code; I realize that a) the above isn't necessarily the
    // the right data type, and b) casting it to the right data type could cause recursion without some
    // changes or facilities added to the Standard.
}
 
inline constexpr Centimeters operator"" _cm (int cm)
{
    return (Centimeters)cm;
}
 
Inches i = (Inches)(12_cm); // The point of this code is to show this line

 

Also, what would be the implications with promotions? What if you multiplied Inches (which is typedefd as an int) by a float? Or multiplied it by an int?

 

*This is, I would say, related to opaque typedefs, but is also a bit of a separate issue (because declaring a casting operator makes implicit conversions possible, and AFAIK you can't make a casting operator that requires explicit casting). I suppose you could also say it could work for non-opaque typedefs/implicit conversions, but I don't like the ambiguity of Inches i = 12_cm;


Edited by Cornstalks, 28 December 2012 - 03:15 PM.

[ I was ninja'd 71 times before I stopped counting a long time ago ] [ f.k.a. MikeTacular ] [ My Blog ] [ SWFer: Gaplessly looped MP3s in your Flash games ]

#11 Matt-D   Crossbones+   -  Reputation: 1469

Like
0Likes
Like

Posted 28 December 2012 - 03:06 PM

There's already a related discussion on std-proposals:

 



#12 Madhed   Crossbones+   -  Reputation: 3134

Like
0Likes
Like

Posted 28 December 2012 - 03:38 PM

It may not be exactly what you're after but here is something interesting using the new c++11 user-defined-literal syntax.

 

http://www.codeproject.com/Articles/447922/Application-of-Cplusplus11-User-Defined-Literals-t



#13 santa01   Members   -  Reputation: 307

Like
0Likes
Like

Posted 28 December 2012 - 03:49 PM

I have felt the need for something like this when I have classes Vector3D and Point3D, which are essentially the same thing, but I need to define them separately if I want the type system to help me make sure my operations make sense (e.g., you are not allowed to add points, but adding vectors is fine, and so is adding a point and a vector).

 

Ok, I got the idea. Still thats more like a copy of the type, rather than a typedef, so i suggest something like:

 

typecopy unsigned int Centimeters;
typecopy unsigned int Inches;

which should make an exact type copy under the specified name.

 

Also its not clear what to do with types hierarchy. Lets say:

class A {};
class B: public class A {};
typecopy B MyB; // should MyB also be a subclass of A or just copy the public part of an A interface?


#14 wack   Members   -  Reputation: 1348

Like
2Likes
Like

Posted 29 December 2012 - 05:29 PM

I can see how this would be useful, but there was one (non technical) thing about the paper that didn't feel right.

 

"(The author of this proposal does not seriously endorse the use of macros for type aliases, and the 

above example was merely used for dramatic effect to trigger the gag reflex of any committee 
members reading this)"
 
I'm not sure if papers like these are an appropriate place for smug jokes, but I would venture they're not.


#15 Paradigm Shifter   Crossbones+   -  Reputation: 5436

Like
0Likes
Like

Posted 29 December 2012 - 05:36 PM

I agree - know your audience.
"Most people think, great God will come from the sky, take away everything, and make everybody feel high" - Bob Marley

#16 0r0d   Members   -  Reputation: 819

Like
0Likes
Like

Posted 29 December 2012 - 08:28 PM


I don't want Centimeters and Inches to be convertible

 
Well and I don't understand why. As I wrote earlier built-in types with similar (in fact equal) semantics _are_ convertible.

 


You seriously don't understand why it's desirable to get a compiler error if someone has a quantity in centimeters and tries to use it where a quantity in inches is expected? I am not sure how else to explain it, since SOTL has been very clear.

 

How about this:

 

class Inch
{
	float mValue;
};

class Centimeter
{
	float mValue;
};

int main(void)
{
	Inch i;
	Centimeter c = i;

	return 0;
}
test.cpp : error C2440: 'initializing' : cannot convert from 'Inch' to 'Centimeter'
        No constructor could take the source type, or constructor overload resolution was ambiguous
 

 

Am I missing something?



#17 Servant of the Lord   Crossbones+   -  Reputation: 21046

Like
0Likes
Like

Posted 29 December 2012 - 08:52 PM

 
I can see how this would be useful, but there was one (non technical) thing about the paper that didn't feel right.
 
"(The author of this proposal does not seriously endorse the use of macros for type aliases, and the 
above example was merely used for dramatic effect to trigger the gag reflex of any committee 
members reading this)"
 
I'm not sure if papers like these are an appropriate place for smug jokes, but I would venture they're not.

Good point - I was trying to inject a spot of humor, but yeah, probably not the right place to do so.
How about this:

class Inch
{
	float mValue;
};

class Centimeter
{
	float mValue;
};

int main(void)
{
	Inch i;
	Centimeter c = i;

	return 0;
}
 
Am I missing something?



Yes, that makes the entire language very clunky. I would hate to have to do this:
myInch.mValue = 17;
myCentimeter.mValue = 35;

It's perfectly fine to abbreviate my username to 'Servant' rather than copy+pasting it all the time.
All glory be to the Man at the right hand... On David's throne the King will reign, and the Government will rest upon His shoulders. All the earth will see the salvation of God.
Of Stranger Flames - [indie turn-based rpg set in a para-historical French colony] | Indie RPG development journal

[Fly with me on Twitter] [Google+] [My broken website]

[Need web hosting? I personally like A Small Orange]


#18 0r0d   Members   -  Reputation: 819

Like
0Likes
Like

Posted 29 December 2012 - 09:11 PM

How about this:

class Inch
{
	float mValue;
};

class Centimeter
{
	float mValue;
};

int main(void)
{
	Inch i;
	Centimeter c = i;

	return 0;
}
 
Am I missing something?

 


Yes, that makes the entire language very clunky. I would hate to have to do this:
myInch.mValue = 17;
myCentimeter.mValue = 35;

 

 

First, that code will not compile because mValue is a private member.  Second, you can define the constructors and assignment operators so that you dont need to directly access mValue at all.  Yes, there's some code you need to write for the class, but using it would be no different from using a built-in type, and you actually now have more flexibility in what operations you can do on your types and how to convert them, which get automatically converted and which result in compile errors, etc.

 

So I still dont understand why simply using a class is not an acceptable solution.  You want a new type with your own defined data and behaviors... which is exactly what classes are meant for.



#19 Álvaro   Crossbones+   -  Reputation: 13913

Like
0Likes
Like

Posted 29 December 2012 - 09:33 PM

So I still dont understand why simply using a class is not an acceptable solution.  You want a new type with your own defined data and behaviors... which is exactly what classes are meant for.


The point is that we don't want to have to write identical classes if we can help it.

Here's another example:
struct Vector3D {
  double x, y, z;
  
  Vector3D(double x, double y, double z) : x(x), y(y), z(z) {
  }
  
  void print(std::ostream &os) const {
    os << '(' << x << ',' << y << ',' << z << ')';
  }
};

struct Point3D {
  double x, y, z;
  
  Point3D(double x, double y, double z) : x(x), y(y), z(z) {
  }
  
  void print(std::ostream &os) const {
    os << '(' << x << ',' << y << ',' << z << ')';
  }
};

It would be nice to say "a Point3D works exactly as a Vector3D". A typedef would allow that, but then you won't actually get two separate types.

#20 Servant of the Lord   Crossbones+   -  Reputation: 21046

Like
0Likes
Like

Posted 29 December 2012 - 09:58 PM

First, that code will not compile because mValue is a private member.  Second, you can define the constructors and assignment operators so that you dont need to directly access mValue at all.  Yes, there's some code you need to write for the class, but using it would be no different from using a built-in type, and you actually now have more flexibility in what operations you can do on your types and how to convert them, which get automatically converted and which result in compile errors, etc.
 
So I still dont understand why simply using a class is not an acceptable solution.  You want a new type with your own defined data and behaviors... which is exactly what classes are meant for.
 
No, I want a new type with identical data and identical behavior, with the only change being no implicit conversion between the original type and the derivative.
It'd be a huge amount of unnecessary boilerplate code for virtually identical classes.
 
Here's a copy+paste of real code from my current project, using my macro version (which hides all the boilerplate, but has some flaws):
strong_typedef(cPoint, SubTileLoc, Point); //The position (in tiles, not in pixels) of a tile within a tileset image.
strong_typedef(cPoint, PosInImage, Point); //The pixel position in an image.

strong_typedef(cPoint, PosInMonitor, Point); //The position (in pixels) in the monitor itself.
strong_typedef(cPoint, PosInWindow, Point); //The position (in pixels) in the game window's client area.
strong_typedef(cPoint, PosInVirtualWindow, Point); //The position (in pixels) in the game's window, after accounting for the virtual window size.

strong_typedef(cPoint, PosInWorld, Point); //Position within the world, in pixels.
strong_typedef(cPoint, VisiblePosInWorld, Point); //Position within the loaded portion of the world, in pixels.
//strong_typedef(cPoint, PosInArea, Point); //Might be needed in the future (if I ever allow multiple _areas_ side by side).
strong_typedef(cPoint, PosInCell, Point); //Position within a cell, in pixels.
strong_typedef(cPoint, PosInTile, Point); //Position within a tile, in pixels.

strong_typedef(cPoint, TileLoc, Point); //The position of a tile within a cell, in tiles (not pixels).
strong_typedef(cPoint, CellLoc, Point); //The position of a cell within the area, in cells (not pixels).
strong_typedef(cPoint, VisibleCellLoc, Point); //The position of a cell within the loaded chunks, in cell (not pixels). VisibleCellLoc(0,0) is usually the cell the player is in.

strong_typedef(cPoint, TileOffset, Point); //The offset from the grid, in pixels, at which a tile is drawn. (0,0) is aligned with grid.

strong_typedef(cRect, AreaBounds, Rect); //Bounds of the area, in cells.
Ignore the third parameter, and read it as "typedef cRect AreaBounds;". (The 'c' in my classes are part of an abbreviated namespace, and do not stand for 'class'. Only a few basic times have that prefix)

Notice a common theme there? 13 of the 14 strong typedefs actually are identical classes. You'd have me manually create 13 different classes? No, I wouldn't bother. One or two, maybe. But since it's so easy with a good strong typedef, I can do a dozen with zero extra programming work, and gain the following benefits for free:
1) The code is more self-descriptive. cPoint is generic name, CellLoc tells much better what the variable is, so the variable name can tell its use. Example: "CellLoc centerCell;"
2) Compile-time bug catching of simple mistakes where data of one type (like inches, or tiles) is given when another measurement (like centimeters, or pixels) is expected. This is the primary motivation for me doing this, after several times (over the course of my current multi-year project) I got burned by subtle bugs that the compiler could've caught (and now does!) where I made a simple mistake that went uncaught for quite awhile. After implementing the system and recompiling, the compiler balked and gave a false positive... so I thought, until I looked closer and realized it caught a bug that I didn't yet know existed.

Again, it takes zero programming effort to get these benefits, if strong typedefs are available. Regular typedefs provide the first benefit only, and people only use regular typedefs in less than half the situations where they'd be beneficial, because the benefit they provide is so minor. Mostly people use typedefs for minimizing the typing work they have to do on long class names (which is a valid usage), instead of using the benefits of typedefs to increase code clarity.

My typedef'd items above may have gone a little overboard, and two or three of them probably don't make much sense (but do no harm either). The rest are very significant improvements to the readability and stability of my code.

Edited by Servant of the Lord, 29 December 2012 - 10:01 PM.

It's perfectly fine to abbreviate my username to 'Servant' rather than copy+pasting it all the time.
All glory be to the Man at the right hand... On David's throne the King will reign, and the Government will rest upon His shoulders. All the earth will see the salvation of God.
Of Stranger Flames - [indie turn-based rpg set in a para-historical French colony] | Indie RPG development journal

[Fly with me on Twitter] [Google+] [My broken website]

[Need web hosting? I personally like A Small Orange]





Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS