On Wed, Nov 02, 2005 at 05:53:49PM -0800, Joe Buck wrote: > On Wed, Nov 02, 2005 at 08:44:51PM -0500, Daniel Jacobowitz wrote: > > On Wed, Nov 02, 2005 at 05:32:51PM -0800, Joe Buck wrote: > > > On Thu, Nov 03, 2005 at 02:13:05AM +0100, Gabriel Dos Reis wrote: > > > > | Have -Wuninitialized be a very simple detector, which is either in > > > > the front-ends > > > > | or in the middle-end so it could be shared (just like -Wunused). > > > > | Have -Wuninitialized=2, be the current -Wuninitialized. > > > > > > > > That is backward. Have -Wuninitialized means whatever it means today. > > > > > > Agreed. We don't want it to change much; people who use -Wall -Werror > > > will be particularly pissed off if gcc produces new, but bogus, warnings > > > for uninitialized variables (please feel free to produce new, but *valid*, > > > warnings). > > > > People who use -Wall -Werror are _already_ pissed off about > > -Wuninitialized. It virtually guarantees that your build will fail on > > a new release of GCC. > > I don't have that experience, but that's mainly because I use more than > one compiler version and turn warnings on in all. Anything marginal > is probably gone already.
GDB and binutils have relatively limited, but increasing, exposure to -Werror problems since they've enabled it. So far my experience holds: every newly tried release of GDB triggers a couple new -Wuninitialized warnings. (It's lost in the noise for GDB, though, which died a terrible death relating to char * vs unsigned char * in prototypes that we're still trying to sort out.) -- Daniel Jacobowitz CodeSourcery, LLC
