Compiler-independence for 64-bit integral constants

[Mike Nordell]

Currently there are quite a few places in ROS src where 64-bit integral
constants are either hard-coded to be GNU C only (?), or almost worse
littered with conditional compilation.

Neither is satisfactory. 1 is making it impossible to compile using e.g. an
MS compiler, and 2 makes the code so ugly that... Well, you understand.

So, in the interest of fixing this once and for all, the only option is to
name a macro to handle these compiler specifics in one place, and then use
that macro when defining 64-bit integral constants. The question is then,
what to name it? The macro will be defined something like (omitting unsigned
here):

#if defined __GNUC__
#define ZZZ(X) X##LL
#elif defined _MSC_VER
#define ZZZ(X) X##i64
#else
#error Unknown compiler for 64-bit integral constant suffix
#endif

Serious suggestions that have surfaced so far are:

DEFINE_INT64
MAKE_INT64

If suggesting something else, keep in mind that
- while short might be beautiful (and fast to type), using short macros is
likely to sooner or later create conflicts.
- leading underscore followed by capital character or another underscore is
not legal for C++.
- It's to make it clear that it's a 64-bit int defined, why suggestions
mentioning LL or i64, as in compiler-specific extensions/implementation
details, are likely not to help or be accepted since that's exactly what we
want to get away from.

Please comment.

/Mike

[Martin Fuchs]

Hi,
> So, in the interest of fixing this once and for all, the only option is to
> name a macro to handle these compiler specifics in one place, and then use
> that macro when defining 64-bit integral constants. The question is then,
> what to name it? The macro will be defined something like (omitting unsigned
> here):

I already used the following defines:  (in winefile and explorer)

#ifdef _MSC_VER
#define	LONGLONGARG _T("I64")
#else
#define	LONGLONGARG _T("L")
#endif
> #if defined __GNUC__
> #define ZZZ(X) X##LL
> #elif defined _MSC_VER
> #define ZZZ(X) X##i64
> #else
> #error Unknown compiler for 64-bit integral constant suffix
> #endif
> Serious suggestions that have surfaced so far are:
> DEFINE_INT64
> MAKE_INT64

May be we can combine this ideas to a new proposal:

#ifdef __GNUC__
#define	FMT_INT64 _T("L")
#elif defined(_MSC_VER)
#define	FMT_INT64 _T("I64")
#else
#error Unknown compiler for 64-bit integral constant suffix
#endif

"FMT" stands for "format string", I think this is a bit clearer than "DEFINE" or "MAKE".

This are some usage examples:

printf("%" FMT_INT64 "d", index);  /* format 64 bit decimal integer */

sprintf(buffer, "%" FMT_INT64 "x", index);  /* format 64 bit hex number, lower case */

sprintf(buffer, "%" FMT_INT64 "X", index);  /* format 64 bit hex number, upper case */

_stprintf(buffer, TEXT("%") FMT_INT64 TEXT("X"), index); /* compatible both to UNICODE and ANSI builds */

Regards,

Martin

[Casper Hornstrup]

> -----Original Message-----
> From: [e-mail] 
> [mailto:[e-mail]] On Behalf Of Martin Fuchs
> Sent: 9. oktober 2004 17:32
> To: ReactOS Development List
> Subject: [ros-dev] Re: Compiler-independence for 64-bit 
> integral constants
>
> Hi,
>
> "FMT" stands for "format string", I think this is a bit 
> clearer than "DEFINE" or "MAKE".

There is really no good reason not to spell it out (eg. name it
FORMAT_INT64 instead).

Casper

[Mike Nordell]

Martin Fuchs wrote:
> I already used the following defines:  (in winefile and explorer)
>
> #ifdef _MSC_VER
> #define LONGLONGARG _T("I64")
> #else
> #define LONGLONGARG _T("L")
> #endif

I think you misunderstood me. I wasn't talking about printf-ish format
string specifiers, I was talking about integral literals. Where some code
currently is littered with:

#if defined __GNUC__
  a = 1LL;
#elif defined _MSC_VER
  a = 1i64;
#else
#error unknown compiler for 64-bit integral literal
#endif

to have it somewhat compiler independent, I was aiming for a single:

a = ZZZ(1);

The sprinkled conditional compilation approach is not only butt-ugly but
also a maintenance nightmare, especially if there comes along a compiler
using neither i64 nor LL suffix for 64-bit integral literals.

Your example does however nicely display an example of what I didn't want to
see - exposing compiler specific extentions/implementation details.

Staying in the realm of C89, which we must do if any MS compilers are to be
able to translate the code (since the word from MS is "we don't care about
C99, and will not bother with it"), "long long" and variations thereof is a
compiler specific extension to C to store a 64-bit integer as much as
__int64 is.

What I would prefer is to find a macro name that isn't so long that people
wildly objects to it, while at the same time make the casual
reader/maintainer of the code immediately understand it's a 64-bit integer
being defined without having to know specific implementation details between
2..n compilers.

Another thing to perhaps consider/look into is - what will happen when a
single "long" is 64 bits (on a 64-bit CPU)? Will then "long long" be 128
bits? If so, any names containing "long long" or any variations thereof will
then obviously be wrong if what's intended in the code is specifically a
64-bit value.

Your example did however bring up the next logical question, even if
somewhat superceeded by the fact msvcrt.dll printf format specification by
design and necessity handles the "%i64d"-style format specifications.

For code intended to run on not-ReactOS too, or with a runtime library not
compatible with MS, macros such as yours can be useful. In the realm of
ReactOS I however don't see this as a problem, and even if it is I think its
time would be better placed as a follow-up discussion once a macro name to
define 64-bit integral literals has been decided upon.

/Mike

[Patrick Mauritz]

Am Samstag, 9. Oktober 2004 18:31 schrieb Mike Nordell:
> Another thing to perhaps consider/look into is - what will happen when a
> single "long" is 64 bits (on a 64-bit CPU)? Will then "long long" be 128
> bits? If so, any names containing "long long" or any variations thereof
> will then obviously be wrong if what's intended in the code is specifically
> a 64-bit value.

afaik the garantees are:
char <= short <= int <= long <= long long
and sizeof(char)==1 (thus the other types can only be n chars wide)

it might be better to actually use/define [u]intX_t somewhere and stick to 
those for code that depends on those ranges.

patrick mauritz

[Kohn Emil Dan]

How about avoiding the compiler suffix altogether and only cast the
constants to a (platform independent) typedef, something like
this:

#ifdef __GNUC__
typedef long long INT64;
#elif defined _MSC_VER
typedef __int64 INT64;
#else
#error Please define the INT64 typedef according to your platform
#endif

and then use something like

a = (INT64)0x1;

Or we could hide the cast behind a macro such as

#define ZZZ(x) ((INT64)(x))

(I have no idea what to put instead of ZZZ MAKE_INT64 or DEFINE_INT64).

I see several advantages to this scheme:

1) It can be used both for constants and other expressions

2) It does not care about literal suffixes (e.g. whether the ll suffix
means that the constant has a size of 64 bit or 128 bits). All we have to
do is to define the INT64 typedef appropriately. I doubt that the compiler
will generate a different code between

x = 0ll;

and

x = (long long)0;

What do you think?

Emil

[Royce Mitchell III]

Kohn Emil Dan wrote:
> How about avoiding the compiler suffix altogether and only cast the
> constants to a (platform independent) typedef, something like
> this:
>
> #ifdef __GNUC__
> typedef long long INT64;
> #elif defined _MSC_VER
> typedef __int64 INT64;
> #else
> #error Please define the INT64 typedef according to your platform
> #endif

I can forsee one possible problem with your suggestion... I don't think 
the following would give the desired results:

(INT64)0x100000000
-- 
------------------------
http://wecanstopspam.org

[Barubary]

> I can forsee one possible problem with your suggestion... I don't think 
> the following would give the desired results:
>
> (INT64)0x100000000

Compilers *do* support that, it's just not standard.  GCC issues a warning 
about it, but it does encode it as a 64 bit "long long" constant.  M$VC--  
issues no warning, yet it does emit an "__int64" constant.

Perhaps in this matter you should just add to the makefile to tell GCC to 
ignore that particular warning?  Then the only differences become typedefs 
and *printf (which isn't used by ReactOS anyway, being replaced with custom 
versions).  No special wrapping macros all over the place, except for casts, 
I suppose, in some instances.  These compilers assume (rightly so) that 
0x100000000 is a 64-bit constant.

Really, I think all constant integers should be internally represented as 
the largest possible integer type.  Then leave the compiler to bitch only 
when an implicit cast occurs on an out-of-range value.  That's how it is now 
when working with chars, because constants are implied ints.  The times when 
the difference matters are very small.  Other than the meaning of the 
expression "0x80000000 >> 1", I can't think of any cases where the 
difference matters.

I personally would say C should standardize char=1 short=2 int=4 long=8 and 
maybe long long=16 and intptr=void *.  Never going to happen though.

Melissa

$ gcc test.c -o test.exe
test.c: In function `main':
test.c:7: warning: integer constant is too large for "long" type
$ ./test
FACEC0DEDEADBEEF

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

test.c
Microsoft (R) Incremental Linker Version 7.10.3077
Copyright (C) Microsoft Corporation.  All rights reserved.

/out:test.exe
test.obj

C:\test>test
FACEC0DEDEADBEEF

[Mark Weaver]

> #ifdef __GNUC__
> #define	FMT_INT64 _T("L")
> #elif defined(_MSC_VER)
> #define	FMT_INT64 _T("I64")
> #else
> #error Unknown compiler for 64-bit integral constant suffix
> #endif
>
> "FMT" stands for "format string", I think this is a bit clearer than "DEFINE" or "MAKE".
>
> This are some usage examples:
>
> printf("%" FMT_INT64 "d", index);  /* format 64 bit decimal integer */
>
> sprintf(buffer, "%" FMT_INT64 "x", index);  /* format 64 bit hex number, lower case */
>
> sprintf(buffer, "%" FMT_INT64 "X", index);  /* format 64 bit hex number, upper case */

These don't work when compiling for a unicode build.  It's relatively 
common to deal with mixed strings (e.g. smtp/pop/imap/http server/client 
programs should typically deal with files using unicode for paths, but 
the protocol specifies ASCII strings).  I would plump for 
FMT_INT64A/FMT_INT64W/FMT_INT64 to take care of this.

[Steven Edwards]

Hi,

--- Mike Nordell [e-mail] wrote:
> Currently there are quite a few places in ROS src where 64-bit
> integral
> constants are either hard-coded to be GNU C only (?), or almost worse
> littered with conditional compilation.

Now that this thread has died down I just wanted to say that I dont
really  care how we do it as long as its not a pain to maintain. Right
now we are only about a hours worth of maintaince away from having
ntosknrl and friends building on MS_VC again(if this issue is fixed).

Thanks
Steven

__________________________________
Do you Yahoo!?
Y! Messenger - Communicate in real time. Download now. 
http://messenger.yahoo.com