Stuart Henderson writes:
>> +make this function compile correctly under clang
>> +Index: lib/utils/mu-str.c
>> +--- lib/utils/mu-str.c.orig
>> ++++ lib/utils/mu-str.c
>> +@@ -49,7 +49,7 @@ mu_str_size_s (size_t s)
>> + char*
>> + mu_str_size (size_t s)
>> + {
>> +- return g_strdup (mu_str_size_s(s));
>> ++ return g_format_size_for_display ((goffset)s);
>> + }
>
> I'm no expert on glib2 but this seems a bit odd for "make this function
> compile correctly under clang", have you talked to upstream about it
> at all?
>
That patch is one I submitted, so I can elaborate.
The two functions involved look like this (unpatched):
const char*
mu_str_size_s (size_t s)
{
static char buf[32];
char *tmp;
tmp = g_format_size_for_display ((goffset)s);
strncpy (buf, tmp, sizeof(buf));
buf[sizeof(buf) -1] = '\0'; /* just in case */
g_free (tmp);
return buf;
}
char*
mu_str_size (size_t s)
{
return g_strdup (mu_str_size_s(s));
}
One of the tests was failing, and I found this was because the compiler
was optimizing the function calls out of mu_str_size(), so that
it directly returned the address of mu_str_size_s()'s static buffer.
This happens with base-clang or ports-clang at -O1 or higher.
I just tried ports-gcc and it compiles mu_str_size() correctly, but the
build fails in the link stage. So maybe ports-gcc should be removed from
COMPILER unless that can be fixed.
