Conflicting behavior of MS and Borland

M

Minti

I tried the following code on Borland C++ complier [ 5.5.1 ] and
Microsoft VC7.0 both seem to give conflicting results


void foo(const int& x) { std::cout << "In const foo\n"; }
void foo(int& x) {std::cout << "In non-const foo\n"; }

int main(void)
{
foo(5);
int x = 23;
foo( (const int ) x );
return 0;
}

In case of Borland compiler the result is

In const foo
In const foo

While in case of VC7.0

In const foo
In non-const foo

I could not find much in TC++PL
 
G

Gianni Mariani

Minti said:
I tried the following code on Borland C++ complier [ 5.5.1 ] and
Microsoft VC7.0 both seem to give conflicting results


void foo(const int& x) { std::cout << "In const foo\n"; }
void foo(int& x) {std::cout << "In non-const foo\n"; }

int main(void)
{
foo(5);
int x = 23;
foo( (const int ) x );
return 0;
}

In case of Borland compiler the result is

In const foo
In const foo

gcc 3.4.0 and VC7.1 also says this.
While in case of VC7.0

In const foo
In non-const foo

I suspect it's a bug in VC7.0.
 
V

Victor Bazarov

Gianni said:
Minti said:
I tried the following code on Borland C++ complier [ 5.5.1 ] and
Microsoft VC7.0 both seem to give conflicting results


void foo(const int& x) { std::cout << "In const foo\n"; }
void foo(int& x) {std::cout << "In non-const foo\n"; }

int main(void)
{
foo(5);
int x = 23;
foo( (const int ) x );
return 0;
}

In case of Borland compiler the result is

In const foo
In const foo


gcc 3.4.0 and VC7.1 also says this.

Not the VC7.1 I have. It still says the same as VC7.0. What version of
the compiler do you have? Mine says

Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.3077
for 80x86
Copyright (C) Microsoft Corporation 1984-2002. All rights reserved.

Do I need to update it somehow?
 
P

Peter Kragh

Victor said:
Not the VC7.1 I have. It still says the same as VC7.0. What version of
the compiler do you have? Mine says

Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.3077
for 80x86
Copyright (C) Microsoft Corporation 1984-2002. All rights reserved.

Do I need to update it somehow?

My compiler says:

Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.3052 for 80x86
Copyright (C) Microsoft Corporation 1984-2002. All rights reserved.

I.e. it's a bit older than yours. If I turn Microsoft's extension off
(/Za), I get the expected:

In const foo
In const foo

Just my $0.05

Br,
Peter
 
L

Leor Zolman

Gianni said:
Minti said:
I tried the following code on Borland C++ complier [ 5.5.1 ] and
Microsoft VC7.0 both seem to give conflicting results


void foo(const int& x) { std::cout << "In const foo\n"; }
void foo(int& x) {std::cout << "In non-const foo\n"; }

int main(void)
{
foo(5);
int x = 23;
foo( (const int ) x );
return 0;
}

In case of Borland compiler the result is

In const foo
In const foo


gcc 3.4.0 and VC7.1 also says this.

Not the VC7.1 I have. It still says the same as VC7.0. What version of
the compiler do you have? Mine says

Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.3077
for 80x86
Copyright (C) Microsoft Corporation 1984-2002. All rights reserved.

Do I need to update it somehow?
I suspect it's a bug in VC7.0.

I suspect it is a compatibility "feature" of 7.0 (don't ask me what it is
supposed to be compatible /with/, since VC6 does it "right" by default...)

Using both MSVC 7.0 and 7.1 with /Za, I get both const. Without /Za, I get
the one non-const.

A revealing diagnostic comes out when you compile it with Comeau, though:

test.cpp(10): warning: type qualifier is meaningless on cast type
foo( (const int ) x );

This makes sense, as casts always produce a temporary that is not
assignable, I think. So clearly we have a case of the parameter needing to
be a reference-to-const so that it can bind to a temporary.

None of this explains why VC7.x does it the other way by default. MSVC
works in mysterious ways when it comes to its default compatibility mode...
-leor
 
M

Minti

Leor Zolman said:
Gianni said:
Minti wrote:

I tried the following code on Borland C++ complier [ 5.5.1 ] and
Microsoft VC7.0 both seem to give conflicting results


void foo(const int& x) { std::cout << "In const foo\n"; }
void foo(int& x) {std::cout << "In non-const foo\n"; }

int main(void)
{
foo(5);
int x = 23;
foo( (const int ) x );
return 0;
}

In case of Borland compiler the result is

In const foo
In const foo


gcc 3.4.0 and VC7.1 also says this.

Not the VC7.1 I have. It still says the same as VC7.0. What version of
the compiler do you have? Mine says

Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.3077
for 80x86
Copyright (C) Microsoft Corporation 1984-2002. All rights reserved.

Do I need to update it somehow?
While in case of VC7.0

In const foo
In non-const foo


I suspect it's a bug in VC7.0.

I suspect it is a compatibility "feature" of 7.0 (don't ask me what it is
supposed to be compatible /with/, since VC6 does it "right" by default...)

Using both MSVC 7.0 and 7.1 with /Za, I get both const. Without /Za, I get
the one non-const.

A revealing diagnostic comes out when you compile it with Comeau, though:

test.cpp(10): warning: type qualifier is meaningless on cast type
foo( (const int ) x );

This makes sense, as casts always produce a temporary that is not
assignable, I think.

I really don't think that is indeed the case, if the cast can be done
at compile time
like in

const int x = 5.5;

then there is no need to produce a temporary.

I am not sure what you mean by "not assignable" do you mean that they
can't be an lvalue.
 
L

Leor Zolman

I really don't think that is indeed the case, if the cast can be done
at compile time
like in

const int x = 5.5;
I don't consider that a "cast"; that's just a conversion. By "cast", I mean
either
(type) expression
type (expression)
or new-style_cast<type>(expression)

and I'm saying you can't make the result of one of those be the left
operand of an assignment operator (or ++/--).
then there is no need to produce a temporary.

Right, because it isn't a "cast".
I am not sure what you mean by "not assignable" do you mean that they
can't be an lvalue.

That's probably better wording, or "not modifiable" perhaps. I began trying
to elaborate on this based on what the Standard says; I've been bouncing
around between 3.10/2, 3.10/6 and writing test programs, and finally
decided I just don't understand some of the subtleties enough to really
follow through with an analysis.

I do, however, have a pet theory as to what the default MSVC 7 behavior is
doing, even though I haven't a clue why Microsoft would find such behavior
to be useful. In the strange behavior, the call

foo ((const int) x);

results in the function call argument being bound to a reference to
non-const (the function parameter). Here's what section 3.10/6 has to say
about the result of casts:

"An expression which holds a temporary object resulting from a cast to
a nonreference type is an rvalue (this includes the explicit creation
of an object using functional notation (5.2.3))."

My suspicion is that the strange behavior is ignoring this rule, and
retaining all the lvalue-ness of the cast's operand. Thus it behaves as we
see when the operand is an lvalue, but not when it is a constant. Consider
this expansion of the original example:

#include <iostream>

void foo(int& x) {std::cout << "In non-const foo\n"; }
void foo(const int& x) { std::cout << "In const foo\n"; }

int main(void)
{
int x = 23;

foo(5);
foo(x);
foo( (const int ) x );
foo( (const int ) 5 );

return 0;
}

The output is:

In const foo
In non-const foo
In non-const foo
In const foo

The third one is the strange one, and can be explained by imagining that
the cast isn't actually producing a "new" temporary value, but rather is
passing the original operand x "as if" it had the desired type. The last
example shows that if you start with a constant (5), it doesn't magically
turn into an lvalue (which is a good thing.)

Another shot in the dark: perhaps Microsoft is interpreting 3.10/6's use of
the term "nonreference" to mean that, when the argument can be bound to a
(non-const) reference in the function call, the result of the cast is no
longer a "nonreference type" and thus exempt from the requirements of being
an rvalue. But if /Za is used, it battens down the hatches. I just have no
idea.

For what it's worth.
-leor
 
G

Gianni Mariani

Victor said:
Gianni said:
Minti said:
I tried the following code on Borland C++ complier [ 5.5.1 ] and
Microsoft VC7.0 both seem to give conflicting results


void foo(const int& x) { std::cout << "In const foo\n"; }
void foo(int& x) {std::cout << "In non-const foo\n"; }

int main(void)
{
foo(5);
int x = 23;
foo( (const int ) x );
return 0;
}

In case of Borland compiler the result is

In const foo
In const foo



gcc 3.4.0 and VC7.1 also says this.


Not the VC7.1 I have. It still says the same as VC7.0. What version of
the compiler do you have? Mine says

Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.3077
for 80x86
Copyright (C) Microsoft Corporation 1984-2002. All rights reserved.

Do I need to update it somehow?

Nope - It was my mistake ... Getting a bit punchy at 9:00AM !
 
V

Victor Bazarov

Gianni Mariani said:
Victor said:
Gianni said:
Minti wrote:

I tried the following code on Borland C++ complier [ 5.5.1 ] and
Microsoft VC7.0 both seem to give conflicting results


void foo(const int& x) { std::cout << "In const foo\n"; }
void foo(int& x) {std::cout << "In non-const foo\n"; }

int main(void)
{
foo(5);
int x = 23;
foo( (const int ) x );
return 0;
}

In case of Borland compiler the result is

In const foo
In const foo



gcc 3.4.0 and VC7.1 also says this.


Not the VC7.1 I have. It still says the same as VC7.0. What version of
the compiler do you have? Mine says

Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 13.10.3077
for 80x86
Copyright (C) Microsoft Corporation 1984-2002. All rights reserved.

Do I need to update it somehow?

Nope - It was my mistake ... Getting a bit punchy at 9:00AM !

Gianni, you were correct, the others suggested that "language extensions"
were turned on, and they were correct. I turned language extensions off
and got the behaviour you described. I wonder how long MSoft is going
to keep "extensions" _on_ by default...

Thanks to all for corrections.

V
 
J

Jake Montgomery

Leor said:
Gianni said:
Minti wrote:

I tried the following code on Borland C++ complier [ 5.5.1 ] and
Microsoft VC7.0 both seem to give conflicting results


void foo(const int& x) { std::cout << "In const foo\n"; }
void foo(int& x) {std::cout << "In non-const foo\n"; }

int main(void)
{
foo(5);
int x = 23;
foo( (const int ) x );
return 0;
}

In case of Borland compiler the result is

In const foo
In const foo
[Snip]


I suspect it is a compatibility "feature" of 7.0 (don't ask me what it is
supposed to be compatible /with/, since VC6 does it "right" by default...)

Using both MSVC 7.0 and 7.1 with /Za, I get both const. Without /Za, I get
the one non-const.

A revealing diagnostic comes out when you compile it with Comeau, though:

test.cpp(10): warning: type qualifier is meaningless on cast type
foo( (const int ) x );

This makes sense, as casts always produce a temporary that is not
assignable, I think. So clearly we have a case of the parameter needing to
be a reference-to-const so that it can bind to a temporary.

None of this explains why VC7.x does it the other way by default. MSVC
works in mysterious ways when it comes to its default compatibility mode...
-leor

I dont know if this sheds any light on why MS would have such a language
extension, but you can use a similar cast on an lvalue in VC7.1 with
language extensions on.

const int c = 4;
int n = 5;
volatile int v = 6;

(const int)n = 6; // Fine in VC 7.1 with extensions
(int)v = 3; // Fine in VC 7.1 with extensions

(int)c = 3; // illegal
(unsigned int)n = 7; // illegal


But still ... why?

- Jake Montgomery
 
L

Leor Zolman

I dont know if this sheds any light on why MS would have such a language
extension, but you can use a similar cast on an lvalue in VC7.1 with
language extensions on.

const int c = 4;
int n = 5;
volatile int v = 6;

(const int)n = 6; // Fine in VC 7.1 with extensions
(int)v = 3; // Fine in VC 7.1 with extensions

(int)c = 3; // illegal
(unsigned int)n = 7; // illegal

Yup, that's consistent with my "pet theory" in my reply to Minti above:
this language extension allows a cast to serve as a type modifier on the
actual lvalue operand, rather than necessarily resulting in an rvalue
temporary as the Standard dictates.
But still ... why?

Indeed. The kicker, to me, is that MSVC 6 does not seem to exhibit this
extension (at least not by default).
-leor
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Members online

No members online now.

Forum statistics

Threads
473,769
Messages
2,569,580
Members
45,054
Latest member
TrimKetoBoost

Latest Threads

Top