----- Mail original ----- > De: "g esp" <[email protected]> > À: "Akim Demaille" <[email protected]> > Cc: [email protected] > Envoyé: Vendredi 8 Juin 2012 12:47:41 > Objet: Re: [PATCH] Fix tests optimization removal code with -Os > > > > ----- Mail original ----- > > De: "Akim Demaille" <[email protected]> > > À: "Gilles Espinasse" <[email protected]> > > Cc: [email protected] > > Envoyé: Vendredi 8 Juin 2012 08:49:29 > > Objet: Re: [PATCH] Fix tests optimization removal code with -Os > > > > Hi Gilles, > > > > Thanks for the feedback! > > > > Le 7 juin 2012 à 08:44, Gilles Espinasse a écrit : > > > > > Looking my test log and using -Os, I saw code does not what > > > comment > > > say in -Os case. > > > > I remove -O2 because I failed to find a means to silence > > GCC warnings in that case. So it is really a known > > *deficiency* to remove these flags, and the future > > is to keep them. So, unless it does change how the way > > the test suite behaves, I'd rather not do that. > > > > I understand. And I was a bit afraid -Wl,-O1 could match but that > should be only inside LDFLAGS, not CFLAGS > > Without -Os filtered, I have no warnings with my gcc-4.4.5 and could > have 0 failure on test suite. > > I say I /could/ because tests 173 and 301 fail if build following > actual LFS build order. > The reason is yet unclear, and I haven't yet look in details. > bison-2.5 test suite was fine with same package build order. >
Now I understand better. I need to apply my patch to fix tests 173 and 301 failures because both produced .c code emitting various 'may be used uninitialized in this function' warnings when -Os is not removed. Gilles
