Re: Size of basic types

Re: Size of basic types

[Paul Schlie]

> Zack Weinberg writes:
>>Jean-Eric Cuendet <jec@rptec.ch> writes:
>> Hi,
>> I work under 64 bits Linux on AMD64 and GCC 3.4
>> I have:
>> int => 32
>> long => 64
>> long long => 64
>> void* => 64
>>
>> Is it possible to have:
>> int => 32
>> long => 32
>> long long => 64
>> void* => 64
>
> No.  This will never be supported

Do you mean GCC can't support the desired type size specification for some
reason; or that it's not consistent Linux's 64-bit kernel/library API type
size requirements/assumptions?

Re: Size of basic types

[Zack Weinberg]

Paul Schlie wrote:
> > Zack Weinberg writes:
> >>Jean-Eric Cuendet <jec@rptec.ch> writes:
> >> Hi,
> >> I work under 64 bits Linux on AMD64 and GCC 3.4
> >> I have:
> >> int => 32
> >> long => 64
> >> long long => 64
> >> void* => 64
> >>
> >> Is it possible to have:
> >> int => 32
> >> long => 32
> >> long long => 64
> >> void* => 64
> >
> > No.  This will never be supported
> 
> Do you mean GCC can't support the desired type size specification
> for some reason; or that it's not consistent Linux's 64-bit
> kernel/library API type size requirements/assumptions?

I mean primarily that it's not consistent with the amd64-linux
platform ABI and therefore would require a hell of a lot of work to
implement (you'd need e.g. a new C library port) which I do not think
is worth doing.

Secondarily, I consider it an important property of C-as-she-is-spoke
that sizeof(void*) <= sizeof(long), and I do not think GCC should ever
facilitate breaking that property.

zw

Re: Size of basic types

[Paul Schlie]

> From: Zack Weinberg <zack@codesourcery.com>
>> Paul Schlie wrote:
>>> Zack Weinberg writes:
>>>> Jean-Eric Cuendet <jec@rptec.ch> writes:
>>>> Hi,
>>>> I work under 64 bits Linux on AMD64 and GCC 3.4
>>>> I have:
>>>> int => 32
>>>> long => 64
>>>> long long => 64
>>>> void* => 64
>>>> 
>>>> Is it possible to have:
>>>> int => 32
>>>> long => 32
>>>> long long => 64
>>>> void* => 64
>>> 
>>> No.  This will never be supported
>> 
>> Do you mean GCC can't support the desired type size specification
>> for some reason; or that it's not consistent Linux's 64-bit
>> kernel/library API type size requirements/assumptions?
> 
> I mean primarily that it's not consistent with the amd64-linux
> platform ABI and therefore would require a hell of a lot of work to
> implement (you'd need e.g. a new C library port) which I do not think
> is worth doing.
> 
> Secondarily, I consider it an important property of C-as-she-is-spoke
> that sizeof(void*) <= sizeof(long), and I do not think GCC should ever
> facilitate breaking that property.
> 
> zw

Thanks, I didn't appreciate that there was a dependency between the size of
a pointer and the size of a long; I had mistakenly presumed that as long as
there was a supported int datatype which was at least as large as a pointer
(as in the above example long long == pointer size) GCC's infrastructure
would support it.

Re: Size of basic types

[Zack Weinberg]

Paul Schlie wrote:
> > From: Zack Weinberg <zack@codesourcery.com>
...
> > Secondarily, I consider it an important property of C-as-she-is-spoke
> > that sizeof(void*) <= sizeof(long), and I do not think GCC should ever
> > facilitate breaking that property.
> 
> Thanks, I didn't appreciate that there was a dependency between the
> size of a pointer and the size of a long; I had mistakenly presumed
> that as long as there was a supported int datatype which was at
> least as large as a pointer (as in the above example long long ==
> pointer size) GCC's infrastructure would support it.

I think GCC could be made to support it as a target; however, it is a
pervasive assumption of C programs that 'long' is the machine pointer
size.  (Formerly and with somewhat less justification, 'int'.)  I
suspect GCC would have trouble being *hosted* on such a platform, and
I think it is bad to break this assumption.

zw

Re: Size of basic types

[Robert Dewar]

Paul Schlie wrote:

> Thanks, I didn't appreciate that there was a dependency between the size of
> a pointer and the size of a long; I had mistakenly presumed that as long as
> there was a supported int datatype which was at least as large as a pointer
> (as in the above example long long == pointer size) GCC's infrastructure
> would support it.

Zack did not say that GCC could not support it, just that it would be
essentially a new port and a huge amount of work, and that it *might*
be the case that breaking the pointer/long assumption would cause
further difficulties.

That seems a fair analysis to me ...

Re: Size of basic types

[Paul Schlie]

> From: Robert Dewar <dewar@adacore.com>
> 
> Paul Schlie wrote:
> 
>> Thanks, I didn't appreciate that there was a dependency between the size of
>> a pointer and the size of a long; I had mistakenly presumed that as long as
>> there was a supported int datatype which was at least as large as a pointer
>> (as in the above example long long == pointer size) GCC's infrastructure
>> would support it.
> 
> Zack did not say that GCC could not support it, just that it would be
> essentially a new port and a huge amount of work, and that it *might*
> be the case that breaking the pointer/long assumption would cause
> further difficulties.
> 
> That seems a fair analysis to me ...

Yes, thanks, understood.

Re: Size of basic types

[Kai Henningsen]

dewar@adacore.com (Robert Dewar)  wrote on 03.12.04 in <41B14380.50307@adacore.com>:

> Paul Schlie wrote:
>
> > Thanks, I didn't appreciate that there was a dependency between the size
> > of a pointer and the size of a long; I had mistakenly presumed that as
> > long as there was a supported int datatype which was at least as large as
> > a pointer (as in the above example long long == pointer size) GCC's
> > infrastructure would support it.
>
> Zack did not say that GCC could not support it, just that it would be
> essentially a new port and a huge amount of work, and that it *might*
> be the case that breaking the pointer/long assumption would cause
> further difficulties.
>
> That seems a fair analysis to me ...

Well, it *might* break gcc; it *will* break a lot of programs who relied  
on the promises of C90.

We've had that discussion lots of time before in various different places.

For example, you can't cast a pointer, ptrdiff_t, size_t, or ssize_t to  
long without losing information in that model. Lots of source code does  
exactly that, for a number of different reasons, because before C99 that  
was guaranteed to work. (Don't forget that C90 didn't have long long.)

MfG Kai

Re: Size of basic types

[jec]

>> Any idea how to resolve that?
>
> OurBool-> unsigned?

That s exactly what I did to solve my problen!

> Or a union.

What do you mean? I don t know what are union...
Thanks to explain.

Re: Size of basic types

[Gabriel Paubert]

On Thu, Dec 02, 2004 at 10:04:34PM +0100, Jean-Eric Cuendet wrote:
> 
> >>Is it possible to have:
> >>int => 32
> >>long => 32
> >>long long => 64
> >>void* => 64
> >
> >
> >No.  This will never be supported.
> 
> OK, thanks for your quick answer.
> But is it possible in GCC to define my custom types.
> Saying:
>   newtype MyInt [32bits]
> or something like that?
> I can not do a typedef.
> 
> My main problem is that:
> On 32bits platform:
>   OurBool => int => 32bits
>   OurInt  => long => 32bits
>   OurLong => long long => 64bits
> But in 64 bits:
>   OurBool => int => 32bits
>   OurInt  => long => 64bits      <=== The problem
>   OurLong => long long => 64bits
> 
> So I changed OurInt to int instead of long. But then, I have problems 
> because we have overloaded methods with OurInt and OurBool, which were 
> different in 32bits mode (int and long) but are the same now in 64bits 
> (the 2 are int)...
> Any idea how to resolve that?

OurBool-> unsigned?

Or a union.

	Gabriel

RE: Size of basic types

[Dave Korn]

> -----Original Message-----
> From: gcc-owner On Behalf Of Allan Sandfeld Jensen
> Sent: 03 December 2004 08:38

> On Thursday 02 December 2004 22:04, Jean-Eric Cuendet wrote:
> > So I changed OurInt to int instead of long. But then, I 
> have problems
> > because we have overloaded methods with OurInt and OurBool, 
> which were
> > different in 32bits mode (int and long) but are the same 
> now in 64bits
> > (the 2 are int)...
> > Any idea how to resolve that?
> >
> You have use long where you should have used int, so:
> #define long int
> 
> `Allan

<g>  And what #define do you suggest for long long?

    cheers, 
      DaveK
-- 
Can't think of a witty .sigline today....
 

Re: Size of basic types

[Matthias B.]

On Thu, 02 Dec 2004 22:04:34 +0100 Jean-Eric Cuendet <jec@rptec.ch> wrote:

> So I changed OurInt to int instead of long. But then, I have problems 
> because we have overloaded methods with OurInt and OurBool, which were 
> different in 32bits mode (int and long) but are the same now in 64bits 
> (the 2 are int)...
> Any idea how to resolve that?

Make at least one of the types a true class with one data member (an int)
and a lot of operator functions. It's a lot of typing but otherwise
simple. And overloading on a true class works fine.

MSB

-- 
He who SHOUTS is always wrong.

Re: Size of basic types

[Allan Sandfeld Jensen]

On Thursday 02 December 2004 22:04, Jean-Eric Cuendet wrote:
> So I changed OurInt to int instead of long. But then, I have problems
> because we have overloaded methods with OurInt and OurBool, which were
> different in 32bits mode (int and long) but are the same now in 64bits
> (the 2 are int)...
> Any idea how to resolve that?
>
You have use long where you should have used int, so:
#define long int

`Allan

Re: Size of basic types

[Luca Benini]

Jean-Eric Cuendet wrote:

> But it's just a typedef on int
> So useless in my case...
> 
Sorry, but i don't understant sizeof(int32_t) is 4 byte...
If do you want sizeof(long) == 4 you can use int32_t...
i miss something....


Luca

-- 
But how can I implement unimaginable chaos in my compiler?
				-Geerhardt Goos-

Re: Size of basic types

[Daniel Jacobowitz]

On Thu, Dec 02, 2004 at 10:04:34PM +0100, Jean-Eric Cuendet wrote:
> 
> >>Is it possible to have:
> >>int => 32
> >>long => 32
> >>long long => 64
> >>void* => 64
> >
> >
> >No.  This will never be supported.
> 
> OK, thanks for your quick answer.
> But is it possible in GCC to define my custom types.
> Saying:
>   newtype MyInt [32bits]
> or something like that?
> I can not do a typedef.
> 
> My main problem is that:
> On 32bits platform:
>   OurBool => int => 32bits
>   OurInt  => long => 32bits
>   OurLong => long long => 64bits
> But in 64 bits:
>   OurBool => int => 32bits
>   OurInt  => long => 64bits      <=== The problem
>   OurLong => long long => 64bits
> 
> So I changed OurInt to int instead of long. But then, I have problems 
> because we have overloaded methods with OurInt and OurBool, which were 
> different in 32bits mode (int and long) but are the same now in 64bits 
> (the 2 are int)...
> Any idea how to resolve that?

I'm not at all sure this will work, but you could try:

typedef long OurInt __attribute__((mode(SI)));

That will be a very strange type - basically "long" for the purposes of
the type system but only 32 bits.  I have no idea what overload
resolution will do with it.

-- 
Daniel Jacobowitz

Re: Size of basic types

[Jean-Eric Cuendet]

> Can you use:
> 
>  int32_t
> 
> from <stdint.h>?

Its *is* 32 bits! :-)
But it's just a typedef on int
So useless in my case...

Re: Size of basic types

[Jean-Eric Cuendet]

> Not in C or any of the related languages.  I believe Ada has something
> along these lines.

OK. Bad...

> Change OurBool to some other type (the real bool comes to mind)?

The problem is that it is persisted as 32 bits already. So changing to 
something other than 32 bits is not an option...
And there is only int that is 32 bits in 64 bits platform.
-jec

Re: Size of basic types

[Zack Weinberg]

Jean-Eric Cuendet <jec@rptec.ch> writes:

>>>Is it possible to have:
>>>int => 32
>>>long => 32
>>>long long => 64
>>>void* => 64
>> No.  This will never be supported.
>
> OK, thanks for your quick answer.
> But is it possible in GCC to define my custom types.
> Saying:
>   newtype MyInt [32bits]
> or something like that?
> I can not do a typedef.

Not in C or any of the related languages.  I believe Ada has something
along these lines.

> My main problem is that:
> On 32bits platform:
>   OurBool => int => 32bits
>   OurInt  => long => 32bits
>   OurLong => long long => 64bits
> But in 64 bits:
>   OurBool => int => 32bits
>   OurInt  => long => 64bits      <=== The problem
>   OurLong => long long => 64bits
>
> So I changed OurInt to int instead of long. But then, I have problems 
> because we have overloaded methods with OurInt and OurBool, which were 
> different in 32bits mode (int and long) but are the same now in 64bits 
> (the 2 are int)...
> Any idea how to resolve that?

Change OurBool to some other type (the real bool comes to mind)?

zw

Re: Size of basic types

[Jean-Eric Cuendet]

>>Is it possible to have:
>>int => 32
>>long => 32
>>long long => 64
>>void* => 64
> 
> 
> No.  This will never be supported.

OK, thanks for your quick answer.
But is it possible in GCC to define my custom types.
Saying:
   newtype MyInt [32bits]
or something like that?
I can not do a typedef.

My main problem is that:
On 32bits platform:
   OurBool => int => 32bits
   OurInt  => long => 32bits
   OurLong => long long => 64bits
But in 64 bits:
   OurBool => int => 32bits
   OurInt  => long => 64bits      <=== The problem
   OurLong => long long => 64bits

So I changed OurInt to int instead of long. But then, I have problems 
because we have overloaded methods with OurInt and OurBool, which were 
different in 32bits mode (int and long) but are the same now in 64bits 
(the 2 are int)...
Any idea how to resolve that?

Thanks a lot.
-jec

Re: Size of basic types

[Zack Weinberg]

Jean-Eric Cuendet <jec@rptec.ch> writes:

> Hi,
> I work under 64 bits Linux on AMD64 and GCC 3.4
> I have:
> int => 32
> long => 64
> long long => 64
> void* => 64
>
> Is it possible to have:
> int => 32
> long => 32
> long long => 64
> void* => 64

No.  This will never be supported.

zw

Size of basic types

[Jean-Eric Cuendet]

Hi,
I work under 64 bits Linux on AMD64 and GCC 3.4
I have:
int => 32
long => 64
long long => 64
void* => 64

Is it possible to have:
int => 32
long => 32
long long => 64
void* => 64
?

Thanks.
-jec

-- 
Jean-Eric Cuendet
Riskpro Technologies SA
Av du 14 avril 1b, 1020 Renens Switzerland
Principal: +41 21 637 0110  Fax: +41 21 637 01 11
Direct: +41 21 637 0123
E-mail: jean-eric.cuendet at rptec.ch
http://www.rptec.ch
--------------------------------------------------------