After an evaluation, GNOME has moved from Bugzilla to GitLab. Learn more about GitLab.
No new issues can be reported in GNOME Bugzilla anymore.
To report an issue in a GNOME project, go to GNOME GitLab.
Do not go to GNOME Gitlab for: Bluefish, Doxygen, GnuCash, GStreamer, java-gnome, LDTP, NetworkManager, Tomboy.
Bug 615379 - g_new macros crash if sizeof(struct_type) == 0
g_new macros crash if sizeof(struct_type) == 0
Status: RESOLVED FIXED
Product: glib
Classification: Platform
Component: general
2.24.x
Other All
: Normal critical
: ---
Assigned To: gtkdev
gtkdev
Depends on:
Blocks:
 
 
Reported: 2010-04-10 18:18 UTC by Jim Evins
Modified: 2012-01-10 15:28 UTC
See Also:
GNOME target: ---
GNOME version: ---


Description Jim Evins 2010-04-10 18:18:59 UTC 
The new version of the g_new family of macros will crash with a floating point exception if sizeof(struct_type) == 0.  Previous versions would safely return a NULL.
Comment 1 Tor Lillqvist 2010-04-10 20:52:45 UTC 
Isn't a struct type with size zero an impossible non-standard abomination anyway? (And when I say non-standard, I mean non-C89.)
Comment 2 Jim Evins 2010-04-10 21:34:44 UTC 
I cannot cite whether or not it violates standard, but it is possible with GCC.  You may end up with an empty structure because all your fields happen to be #ifdef'ed out.  In my case, I followed a common pattern for my objects whether they had private data or not.  Neither may be a good idea, but the problem is that code that worked before now crashes.
Comment 3 Behdad Esfahbod 2010-04-22 23:48:13 UTC 
diff --git a/glib/gmem.h b/glib/gmem.h
index 2fef766..54f153b 100644
--- a/glib/gmem.h
+++ b/glib/gmem.h
@@ -88,7 +88,7 @@ gpointer g_try_realloc_n  (gpointer    mem,
          if (__s == 1)                 \
            __p = g_##func (__n);                               \
          else if (__builtin_constant_p (__n) &&                \
-                  __n <= G_MAXSIZE / __s)                      \
+                  (__s == 0 || __n <= G_MAXSIZE / __s))        \
            __p = g_##func (__n * __s);                         \
          else                                                  \
            __p = g_##func##_n (__n, __s);                      \
@@ -102,7 +102,7 @@ gpointer g_try_realloc_n  (gpointer  mem,
          if (__s == 1)                                         \
            __p = g_##func (__p, __n);                          \
          else if (__builtin_constant_p (__n) &&                \
-                  __n <= G_MAXSIZE / __s)                      \
+                  (__s == 0 || __n <= G_MAXSIZE / __s))        \
            __p = g_##func (__p, __n * __s);                    \
          else                                                  \
            __p = g_##func##_n (__p, __n, __s);                 \
Comment 4 Behdad Esfahbod 2010-04-22 23:49:15 UTC 
Committed to master.
Comment 5 Patrick Welche 2012-01-10 15:28:27 UTC 
Comment 1 seems to preempt Bug 641350