yesno fails with optimization and gcc>=4.7 on x86_64

[Allan McRae]

Hi,

Compiling with -O2 generate codes which mangles the return value from
the yesno function, causing the overwrite prompt to ignore 'n'.

To reproduce:

$ echo "original" >out
$ gzip out
$ echo "replaced" >out
$ gzip out
gzip: out.gz already exists; do you wish to overwrite (y or n)? n
$ gzip -cd out.gz
replaced

This can be seen in Arch Linux and Fedora 18 x86_64 builds (and likely
any other distribution using gcc-4.7)

A very helpful person in the gcc bug tracker [1] pointed out it was
probably yesno being declared with two different function declarations.
 And sure enough, the gnulib one declares with bool and gzip.h uses int
as the return value.

[1] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56211

The following fixes this for me:


2013-02-05  Allan McRae  <address@hidden>

        * gzip.h: Return bool from yesno to be consistent with gnulib.



diff -Naur gzip-1.5-orig/gzip.h gzip-1.5/gzip.h
--- gzip-1.5-orig/gzip.h        2012-01-01 18:53:58.000000000 +1000
+++ gzip-1.5/gzip.h     2013-02-05 22:08:44.498810902 +1000
@@ -42,6 +42,7 @@
 #include <sys/types.h> /* for off_t */
 #include <time.h>
 #include <string.h>
+#include <stdbool.h>
 #define memzero(s, n) memset ((voidp)(s), 0, (n))

 #ifndef RETSIGTYPE
@@ -323,4 +324,4 @@
 extern int inflate (void);

         /* in yesno.c */
-extern int yesno (void);
+extern bool yesno (void);