Can you give us a single, self-contained file (no #include's) that has the 8 or so lines of code that triggers the problem? Maybe Marc doesn't need it, but I do ;-).

Brad

;;; Here it is ;)

;;; Built like this on Linux: ;;; > /usr/local/Gambit-C/bin/gsc problem.scm ;;; > gcc -I/usr/local/Gambit-C/include -c problem.c problem_.c ;;; Which gives me the following warnings on gcc 4: -- gcc version 4.0.2 20050808 (prerelease) (Ubuntu 4.0.1-4ubuntu9) ;;; problem.c: In function '___H__20_problem_23_0': ;;; problem.c:133: warning: pointer targets in passing argument 1 of 'glColor3bv' differ in signedness

;; C type, directly lifted from gl.h (c-declare "typedef signed char GLbyte; /* 1-byte signed */")

;; OpenGL function prototype, placed here directly so as to ;; not required inclusion of gl.h ;; The original prototype, as it appears in gl.h, has some ;; macros attached to it, to specify "extern", or "__stdcall", etc. ;; I've removed it for the sake of keeping this file simple (and ;; uncluttered with C macros. (c-declare "void glColor3bv( const GLbyte *v );")

;; Scheme types. (c-define-type GLbyte signed-char) (c-define-type GLbyte* (pointer GLbyte))

;; Problem. (define glColor3bv (c-lambda (GLbyte*) void "glColor3bv"))

On 12/27/05, Bradley Lucier lucier@math.purdue.edu wrote:

The problem seems to be in these lines in gambit.h:

#if CHAR_MAX == SCHAR_MAX #define ___SCHAR char #else #define ___SCHAR signed char #endif

I don't know why Marc thinks he needs these lines; I replaced them with

#define ___SCHAR signed char

Since defining ___SCHAR as "signed char" fixes things, that suggests that it was defined as "char" before. In turn that means that CHAR_MAX was not equal to SCHAR_MAX. While possible, that's a very odd setup. Can you create a file which contains the following:

#include <limit.h>

static const a = CHAR_MAX; static const b = SCHAR_MAX;

and run 'gcc -E <filename>.c'? The last two lines should give the values of CHAR_MAX and SCHAR_MAX for your system. Generally I would expect them both to be 127.

AGL

-- Adam Langley agl@imperialviolet.org http://www.imperialviolet.org 650-283-9641

I'm not a C language lawyer, but I understand the issue as follows.

"char" and "signed char" are distinct types even if they have the same ranges (as they do on tjay and my systems). "char" must behave either like "signed char" or "unsigned char" on any given system, but char, "signed char", and "unsigned char" are three distinct types.

On my system SCHAR_MAX and CHAR_MAX are both 127

Brad

On Dec 27, 2005, at 4:01 PM, Adam Langley wrote:

On 12/27/05, Bradley Lucier lucier@math.purdue.edu wrote:

...

The problem seems to be in these lines in gambit.h:

#if CHAR_MAX == SCHAR_MAX #define ___SCHAR char #else #define ___SCHAR signed char #endif

I don't know why Marc thinks he needs these lines; I replaced them with

#define ___SCHAR signed char

Since defining ___SCHAR as "signed char" fixes things, that suggests that it was defined as "char" before. In turn that means that CHAR_MAX was not equal to SCHAR_MAX. While possible, that's a very odd setup. Can you create a file which contains the following:

#include <limit.h>

static const a = CHAR_MAX; static const b = SCHAR_MAX;

and run 'gcc -E <filename>.c'? The last two lines should give the values of CHAR_MAX and SCHAR_MAX for your system. Generally I would expect them both to be 127.

AGL

-- Adam Langley agl@imperialviolet.org http://www.imperialviolet.org 650-283-9641

On 27-Dec-05, at 3:46 PM, Bradley Lucier wrote:

The problem seems to be in these lines in gambit.h:

#if CHAR_MAX == SCHAR_MAX #define ___SCHAR char #else #define ___SCHAR signed char #endif

I don't know why Marc thinks he needs these lines; I replaced them with

#define ___SCHAR signed char

and all was well ...

The above fragment of gambit.h is needed to allow some old C compilers which do not support "signed char" to compile Gambit. As I recall Sun' cc compiler on m68k had this problem.

I've fixed the problem by replacing that fragment with

#ifndef ___SCHAR #define ___SCHAR signed char #endif

This should work on modern C compilers, and for old ones the option "- D___SCHAR=char" can be added to the C compiler's command line to compile Gambit.

On a related subject, I have added the "wchar_t" type to the FFI (and wchar_t-string, nonnull-wchar_t-string, ...). This is useful for portability (on some C compilers wchar_t is 16 bits and on others it is 32 bits, this is now completely transparent to Scheme code).

Marc