The problem: In the Ruby interpreter C code there's a couple of sections that call the __syscall function on macOS and BSD. This is very bad behaviour in macOS land because it is a private (and volatile) API.
The __syscall usage is only included conditionally based on HAVE_ defines, so I would like to find out if I can compile Ruby without it. Here is an example from io.c:
#if defined(HAVE___SYSCALL) && (defined(__APPLE__) || defined(__OpenBSD__))
/* Mac OS X and OpenBSD have __syscall but don't define it in headers */
off_t __syscall(quad_t number, ...);
#endif
...some time later...
static VALUE
rb_f_syscall(int argc, VALUE *argv)
{
VALUE arg[8];
...a bunch of platform checks that usually end up doing this...
# define SYSCALL __syscall
...some time later...
switch (argc) {
case 1:
retval = SYSCALL(num);
break;
case 2:
retval = SYSCALL(num, arg[0]);
break;
case 3:
retval = SYSCALL(num, arg[0],arg[1]);
break;
... and so on up to case 8...
}
... function returns and then...
#undef SYSCALL
}
Sharp-eyed readers will also notice that even the Ruby devs don't like using __syscall - they want to replace it with the DL (Fiddle) library instead.
The constraint: I do not want to fork the Ruby interpreter source to do this as that will lead to continuous cumbersome maintenance of the fork. Instead I would like to pass an argument to the build tool when it compiles Ruby that does this.
The question: Can I force one of the related HAVE__ macros to be undefined, or perhaps disable the AC_CHECK_HEADERS in configure.in, and so prevent the use of __syscall here, by passing an argument to the build tool?
You should be able to tweak one of the ac_cv_*
variables at ./configure
time.
Particularly, for ruby-2.1.9 this should work:
./configure ac_cv_func___syscall=no
(For reference, you can see the ac_cv_*
variables set in config.log
.)