View mode: basic / threaded / horizontal-split · Log in · Help
January 24, 2007
postincrement behaviour (differences between dmd and gdc)
Hello again,

I have a huge arguing going on with a friend of mine, which behaviour of
x++ is the right one. This has in fact to do with D, so read on :)

This is our example application:

int x = 5;
x = x++;
x = x++;
print(x);


I argue: the result printed is 7.
He argues: the result printed is 5.

the "++" operator is defined as: "returns the value and then increments
it by 1". so I guess, the execution should be as follows:


int x = 5;
-> x "gains" the value 5.


x = x++;
-> first, (x = x) is executed. afterwards, x is incremented by 1. as x
and x are the same variables, x now has the value 6.

x = x++;
-> the same as above, so x is now 7.


Allright, now let's look what the compilers do:

C/C++ compiled with GCC:	7
java compiled with SUN's:	5
D compiled with DMD:		5
D compiled with GDC:		7

Allright, so please explain this behaviour to me?

Best regards,
Nicolai Waniek
January 24, 2007
Re: postincrement behaviour (differences between dmd and gdc)
Nicolai Waniek wrote:
> Hello again,
> 
> I have a huge arguing going on with a friend of mine, which behaviour of
> x++ is the right one. This has in fact to do with D, so read on :)
> 
> This is our example application:
> 
> int x = 5;
> x = x++;
> x = x++;
> print(x);
> 
> 
> I argue: the result printed is 7.
> He argues: the result printed is 5.
> 
> the "++" operator is defined as: "returns the value and then increments
> it by 1". so I guess, the execution should be as follows:
> 
> 
> int x = 5;
> -> x "gains" the value 5.
> 
> 
> x = x++;
> -> first, (x = x) is executed. afterwards, x is incremented by 1. as x
> and x are the same variables, x now has the value 6.
> 
> x = x++;
> -> the same as above, so x is now 7.
> 
> 
> Allright, now let's look what the compilers do:
> 
> C/C++ compiled with GCC:	7
> java compiled with SUN's:	5
> D compiled with DMD:		5
> D compiled with GDC:		7
> 
> Allright, so please explain this behaviour to me?

You forgot this:

> C compiled with GDC:          5

So at least it's consistent D and C for each backend - this looks like a DMC vs GCC problem. :/

If you use a different test, you'll see that DMC/DMD treats the post-increment as truly following the evaluation of the 
incremented term, within the expression.

x = 5;
x = x++ + 5; // x is now 10

or

x = 5;
x = x++ + x; // x is now 11

-- 
- EricAnderton at yahoo
January 24, 2007
Re: postincrement behaviour (differences between dmd and gdc)
(Edit) fixed compiler comparison

Nicolai Waniek wrote:
> Hello again,
> 
> I have a huge arguing going on with a friend of mine, which behaviour of
> x++ is the right one. This has in fact to do with D, so read on :)
> 
> This is our example application:
> 
> int x = 5;
> x = x++;
> x = x++;
> print(x);
> 
> 
> I argue: the result printed is 7.
> He argues: the result printed is 5.
> 
> the "++" operator is defined as: "returns the value and then increments
> it by 1". so I guess, the execution should be as follows:
> 
> 
> int x = 5;
> -> x "gains" the value 5.
> 
> 
> x = x++;
> -> first, (x = x) is executed. afterwards, x is incremented by 1. as x
> and x are the same variables, x now has the value 6.
> 
> x = x++;
> -> the same as above, so x is now 7.
> 
> 
> Allright, now let's look what the compilers do:
> 
> C/C++ compiled with GCC:	7
> java compiled with SUN's:	5
> D compiled with DMD:		5
> D compiled with GDC:		7
> 
> Allright, so please explain this behaviour to me?

You forgot this:

> C compiled with DMC:          5

So at least it's consistent D and C for each backend - this looks like a DMC vs GCC problem. :/

If you use a different test, you'll see that DMC/DMD treats the post-increment as truly following the evaluation of the
incremented term, within the expression.

x = 5;
x = x++ + 5; // x is now 10

or

x = 5;
x = x++ + x; // x is now 11

-- 
- EricAnderton at yahoo
January 24, 2007
Re: postincrement behaviour (differences between dmd and gdc)
> 
>> C compiled with DMC:          5
> 
> So at least it's consistent D and C for each backend - this looks like a
> DMC vs GCC problem. :/
> 

I don't have dmc installed anywhere here, so thank you :)
January 24, 2007
Re: postincrement behaviour (differences between dmd and gdc)
Nicolai Waniek wrote:
> Hello again,
> 
> I have a huge arguing going on with a friend of mine, which behaviour of
> x++ is the right one. This has in fact to do with D, so read on :)
> 
> This is our example application:
> 
> int x = 5;
> x = x++;
> x = x++;
> print(x);
[snip]
> Allright, now let's look what the compilers do:
> 
> C/C++ compiled with GCC:	7
> java compiled with SUN's:	5
> D compiled with DMD:		5
> D compiled with GDC:		7
> 
> Allright, so please explain this behaviour to me?

For C & C++, IIRC this is one of the traditional examples of undefined 
behavior (although typically with i instead of x, and only one such 
statement :p).
The technical reason has to do with sequence points, one of the more 
annoying parts of the respective standards.

Basically, the above means the compiler is allowed to emit code that 
does either, or even something else entirely. So 42 or 2235023 would be 
an equally acceptable answer as 5, 6 or 7 here. (Typically though, since 
compilers aren't actively *trying* to screw their users, one of the 
latter 3 will result)

I believe D was designed to be as familiar as possible to C programmers 
(and therefore, to some degree, those of other C-like languages).
IIRC one guideline to achieve this was that code that looks the same 
should have the same behavior. Therefore, this doesn't surprise me at 
all. It's in fact exactly what I would expect to happen.

If above reasoning is correct, you can't rely on any particular result; 
it's likely to be different between compilers and can even be different 
for new versions of the same compiler, or different runs of the same 
version of a compiler...

Conclusion: just don't do stupid **** like above code[1].



[1]: That is, in anything other than such short test programs as the 
above trying to figure out what the compiler does in such a situation ;)
January 24, 2007
Re: postincrement behaviour (differences between dmd and gdc)
On Wed, 24 Jan 2007 21:31:44 +0100, Nicolai Waniek wrote:

> Hello again,
> 
> I have a huge arguing going on with a friend of mine, which behaviour of
> x++ is the right one. This has in fact to do with D, so read on :)
> 
> This is our example application:
> 
> int x = 5;
> x = x++;
> x = x++;
> print(x);
> 
> 
> I argue: the result printed is 7.
> He argues: the result printed is 5.
> 
> the "++" operator is defined as: "returns the value and then increments
> it by 1". so I guess, the execution should be as follows:

I think that DMD is correct. The result should be 5. 

Based on the definition above, I think that the example is equivalent to
...

 int x = 5;
 int temp;

 // x = x++;
 temp = x;  // temp is now 5
 x = x + 1; // x is now 6
 x = temp;  // x is now 5

 // x = x++;
 temp = x;  // temp is now 5
 x = x + 1; // x is now 6
 x = temp;  // x is now 5

The key phrase is "returns the value and then increments" which I take it
to mean that it returns the value of the variable that it had prior to it
being incremented.

-- 
Derek Parnell
January 24, 2007
Re: postincrement behaviour (differences between dmd and gdc)
> Based on the definition above, I think that the example is equivalent to
> ...
> 
>   int x = 5;
>   int temp;
> 
>   // x = x++;
>   temp = x;  // temp is now 5
>   x = x + 1; // x is now 6
>   x = temp;  // x is now 5
> 
>   // x = x++;
>   temp = x;  // temp is now 5
>   x = x + 1; // x is now 6
>   x = temp;  // x is now 5
> 
> The key phrase is "returns the value and then increments" which I take it
> to mean that it returns the value of the variable that it had prior to it
> being incremented.
> 

I interpret it that way:

"first do everything related to the 'return' part, afterwards
increment". therefore, 7 would be the right solution, because in the
first step, 5 is the return value and is assigned to x, afterwards and
as the final step, x is incremented...

;)
January 24, 2007
Re: postincrement behaviour (differences between dmd and gdc)
> 
> Conclusion: just don't do stupid **** like above code[1].
> 
> 
> 
> [1]: That is, in anything other than such short test programs as the
> above trying to figure out what the compiler does in such a situation ;)


Hehe, that's exactly what I told my friend ;)
January 25, 2007
Re: postincrement behaviour (differences between dmd and gdc)
"Derek Parnell" <derek@psych.ward> wrote in message 
news:sh2qs4ffr1mg$.1tm119rzt8yyw$.dlg@40tude.net...
> On Wed, 24 Jan 2007 21:31:44 +0100, Nicolai Waniek wrote:
>
>> Hello again,
>>
>> I have a huge arguing going on with a friend of mine, which behaviour of
>> x++ is the right one. This has in fact to do with D, so read on :)
>>
>> This is our example application:
>>
>> int x = 5;
>> x = x++;
>> x = x++;
>> print(x);
>>
>>
>> I argue: the result printed is 7.
>> He argues: the result printed is 5.
>>
>> the "++" operator is defined as: "returns the value and then increments
>> it by 1". so I guess, the execution should be as follows:
>
> I think that DMD is correct. The result should be 5.
>
> Based on the definition above, I think that the example is equivalent to
> ...
>
>  int x = 5;
>  int temp;
>
>  // x = x++;
>  temp = x;  // temp is now 5
>  x = x + 1; // x is now 6
>  x = temp;  // x is now 5
>
>  // x = x++;
>  temp = x;  // temp is now 5
>  x = x + 1; // x is now 6
>  x = temp;  // x is now 5
>
> The key phrase is "returns the value and then increments" which I take it
> to mean that it returns the value of the variable that it had prior to it
> being incremented.

I agree. Take this code:

int f( int i ) { return i; }//nop

int x=5;
x = f(x++);
assert(x==5);
x = f(x++);
assert(x==5);

L.
Top | Discussion index | About this forum | D home