Re: [bug-gnulib] stat and lstat should define their replacements

[Derek Price]

Paul Eggert wrote:

>Derek Price <address@hidden> writes:
>
>  
>
>>+/* Get the system versions when something else was defined by config.h.  */
>>+#undef lstat
>>+#undef stat
>>    
>>
>
>A big worry here is hosts that use something like "#define stat
>stat64" when compiled in large-file mode.  On such hosts, you can't
>#undef 'stat' and 'lstat' as you'll then get the small-file functions.
>This may help to explain the infelicity in this area of gnulib.
>

Actually, you can, as specified by my patch, since #undef lstat and
#undef stat occur just after inclusion of <config.h> but before the
inclusion of <sys/stat.h>.  I think the problem is my #define stat and
#define lstat in config.h.  From a file other than the lib/stat.c file
which performs the #undefs you cited, then a host like Solaris which
later does a #define stat stat64 in <sys/stat.h> can cause the host to
issue redefinition warnings.

Of course, this should be harmless, since all the references in other
files should want the rpl_stat and rpl_lstat functions, and the #undefs
in lib/stat.c will allow it to get the correct versions there.  And, as
you mention, the GCC 3.4 fix will even silence the warnings.

Cheers,

Derek