Re: [Bug-gnulib] gl_GETOPT broken for Solaris?

[Paul Eggert]

Albert Chin <address@hidden> writes:

>   AC_DEFINE([optind], [rpl_optind],
>     [Define to rpl_optind if the replacement variable should be used.])
>
> Why? If the system has neither <getopt.h> nor the getopt_long_only
> symbol, why replace anything?

I think it's mostly just to simplify the m4 code.  I suspect that the
idea was that it shouldn't hurt.

> Removing the AC_DEFINE's in
> gl_GETOPT_SUBSTITUTE makes the coreutils test suite pass. Else, odd
> things happen, like 'src/basename' returning 'basename' *always*.

I suspect that's because of the "#undef getopt" in system.h.  Please
try this patch:

--- system.h.~1.94.~    2004-08-07 20:04:00 -0700
+++ system.h    2004-11-03 00:24:26 -0800
@@ -129,10 +129,7 @@ void *memrchr (const void *, int, size_t
 #endif
 
 #include <stdbool.h>
-
-#define getopt system_getopt
 #include <stdlib.h>
-#undef getopt
 
 /* The following test is to work around the gross typo in
    systems like Sony NEWS-OS Release 4.0C, whereby EXIT_FAILURE



> Also, why not just blindly use the getopt provided by the program?
> What would it harm?

I think the goal is to use the glibc getopt if available, under the
theory that this keeps the executables smaller and makes it easier to
fix things if getopt needs patching.