Junio C Hamano <gits...@pobox.com> wrote: > Eric Wong <e...@80x24.org> writes: > > > That seems too tedious. I'm learning towards just initializing > > var = NULL in the start of the for-loop: > > > > @@ -449,7 +449,8 @@ static inline struct hashmap_entry > > *hashmap_iter_first(struct hashmap *map, > > * containing a @member which is a "struct hashmap_entry" > > */ > > #define hashmap_for_each_entry(map, iter, var, member) \ > > - for (var = hashmap_iter_first_entry_offset(map, iter, \ > > + for (var = NULL /* squelch uninitialized warnings for OFFSETOF_VAR */, \ > > + var = hashmap_iter_first_entry_offset(map, iter, \ > > OFFSETOF_VAR(var, member)); \ > > That looks a bit too ugly for my taste. I've added an updated > version (see below) and then rebased the whole thing on top of it.
I prefer to minimize the amount of stuff we do to work around buggy compilers (in case they get fixed and old versions are obsoleted). If it's just clang with this problem, we know clang sets __GNUC__, so we can use __typeof__ directly (bypassing extra parentheses in our TYPEOF macro) to get around the warning: #if defined(__GNUC__) /* clang sets this, too */ #define OFFSETOF_VAR(ptr, member) offsetof(__typeof__(*ptr), member) #else /* !__GNUC__ */ ... That said, there could be other compilers which don't set __GNUC__ and have the same problem as clang. But maybe those compilers are too buggy and we already ignore invalid warnings on those compilers.
