Michael Haubenwallner wrote: > > This is an aliasing bug. Please see The C standard, Section 6.3.2.3, > > Pointers. > > Hmm, a bug in what product - the standard, gcc, or my application ? > > And - sorry for that question - where do one read the standard from ? > I've found C89, this seems to have only 5 sections... > And in the C99 draft, Section 6.3.2.3 has to do with aliasing, but I'm > completely lost here - I'll better let interpreting the standard up to > you :) You have to pay a small amount to get an official copy of the standard but there are several draft copies available for free. See <http://www.iso-9899.info/wiki/The_Standard>, <http://www.open-std.org/jtc1/sc22/wg14/>, <http://gcc.gnu.org/readings.html>, etc. For details on aliasing, there is some discussion of it here <http://gcc.gnu.org/bugs.html#nonbugs_c> which includes a link to <http://mail-index.netbsd.org/tech-kern/2003/08/11/0001.html>. Your code violates the aliasing rules of C99 by accessing type (char *) by dereferencing a (void **) cast to (char **). The standard says you cannot do this. Previous versions of gcc were not as strict at enforcing these rules, so you could get away with it then, but not any more. The standard was specifically written this way to make it possible for the compiler to perform a certain set of optimizations which would not otherwise be possible, and gcc is utilizing this freedom. It is thus a bug in your code because your code is not valid C. You might be tempted to consider it a bug in the ISO/IEC committees for allowing this in the standard, but that's not really a practical standpoint as the decision has been made (close to a decade ago now) and there's no good in arguing it now. You might be tempted to consider this a bug in gcc for breaking code that used to work, but again, gcc is breaking code that's invalid, which the standard specifically allows it to do. And besides, they gave you an out with "-fno-strict-aliasing" so that you can get the old behavior if you really want it. Also note that this discussion ("it's broken code" / "no, it's code that used to work") has already played out at length on the gcc development list, to the point of being somewhat of a sore spot now, and pretty much everyone's mind has been made up. Brian
