On Fri, 1 Oct 2004, Stas Bekman wrote:
> Randy Kobes wrote:
> > Hi,
> > On Win32, there's a problem in compiling modperl_cmd.c
> > under src/modules/perl/ with the current cvs about trying to
> > use a function with the wrong number of arguments (lines 311
> > and 317). I think this is due to how modperl_var_modify_t is
> > defined within mod_perl.h, specifically with the presence of
> > _stdcall within the definition of MP_FUNC_T in
> > modperl_common_util.h for Win32. This patch:
> > ================================================================
> > Index: src/modules/perl/mod_perl.h
> > ===================================================================
> > RCS file: /home/cvs/modperl-2.0/src/modules/perl/mod_perl.h,v
> > retrieving revision 1.70
> > diff -u -r1.70 mod_perl.h
> > --- src/modules/perl/mod_perl.h 20 Sep 2004 18:14:47 -0000 1.70
> > +++ src/modules/perl/mod_perl.h 1 Oct 2004 03:14:50 -0000
> > @@ -130,10 +130,10 @@
> >
> > #define MgTypeExt(mg) (mg->mg_type == '~')
> >
> > -typedef void MP_FUNC_T(modperl_var_modify_t) (apr_table_t *,
> > - apr_table_t *,
> > - const char *,
> > - const char *);
> > +typedef void (*modperl_var_modify_t) (apr_table_t *,
> > + apr_table_t *,
> > + const char *,
> > + const char *);
>
>
>
> > /* we need to hook a few internal things before APR_HOOK_REALLY_FIRST */
> > #define MODPERL_HOOK_REALLY_REALLY_FIRST (-20)
> > ====================================================================
> > allows things to build and the tests run (I'm getting a
> > couple of failures, but I'll look at those separately).
> >
>
> Is it possible that the right place to fix that is in modperl_common_util.h:
>
> #ifdef WIN32
> # define MP_FUNC_T(name) (_stdcall *name)
> /* XXX: not all functions get inlined
> * so its unclear what to and not to include in the .def files
> */
> # undef MP_INLINE
> # define MP_INLINE
> #else
> # define MP_FUNC_T(name) (*name)
> #endif
>
> as you can see there already a special case for WIN32. And it was added
> there for some reason. Do you know why? May be it's because of this nature
> of that specific function that you've a problem with? Should it stay
> intact for other uses?
You're right that the
#ifdef WIN32
# define MP_FUNC_T(name) (_stdcall *name)
/* XXX: not all functions get inlined
* so its unclear what to and not to include in the .def files
*/
# undef MP_INLINE
# define MP_INLINE
is needed elsewhere, specifically in modperl_filter.h:
typedef ap_filter_t * MP_FUNC_T(modperl_filter_add_t) ...
where the _stcall is necessary. I think the situation
is analagous to what's described in ap_config.h:
===================================================================
#if !defined(WIN32)
/**
* Apache Core dso functions are declared with AP_DECLARE(), so they may
* use the most appropriate calling convention. Hook functions and other
* Core functions with variable arguments must use AP_DECLARE_NONSTD().
* @code
* AP_DECLARE(rettype) ap_func(args)
* @endcode
*/
#define AP_DECLARE(type) type
/**
* Apache Core dso variable argument and hook functions are declared with
* AP_DECLARE_NONSTD(), as they must use the C language calling convention.
* @see AP_DECLARE
* @code
* AP_DECLARE_NONSTD(rettype) ap_func(args [...])
* @endcode
*/
#define AP_DECLARE_NONSTD(type) type
/**
* Apache Core dso variables are declared with AP_MODULE_DECLARE_DATA.
* This assures the appropriate indirection is invoked at compile time.
*
* @note AP_DECLARE_DATA extern type apr_variable; syntax is required for
* declarations within headers to properly import the variable.
* @code
* AP_DECLARE_DATA type apr_variable
* @endcode
*/
#define AP_DECLARE_DATA
#elif defined(AP_DECLARE_STATIC)
#define AP_DECLARE(type) type __stdcall
#define AP_DECLARE_NONSTD(type) type
#define AP_DECLARE_DATA
#elif defined(AP_DECLARE_EXPORT)
#define AP_DECLARE(type) __declspec(dllexport) type __stdcall
#define AP_DECLARE_NONSTD(type) __declspec(dllexport) type
#define AP_DECLARE_DATA __declspec(dllexport)
#else
#define AP_DECLARE(type) __declspec(dllimport) type __stdcall
#define AP_DECLARE_NONSTD(type) __declspec(dllimport) type
#define AP_DECLARE_DATA __declspec(dllimport)
#endif
====================================================================
So perhaps, in analogy, we could have a
#ifdef WIN32
# define MP_DECLARE(name) *name __stdcall
# define MP_DECLARE_NONSTD(name) *name
/* XXX: not all functions get inlined
* so its unclear what to and not to include in the .def files
*/
# undef MP_INLINE
# define MP_INLINE
#else
# define MP_DECLARE(name) *name
# define MP_DECLARE_NONSTD *name
#endif
and then use MP_DECLARE/MP_DECLARE_NONSTD?
--
best regards,
randy
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
