On 2021/09/20 10:29, Stuart Henderson wrote: > Some of these are pretty hairy and it's a moving codebase. Since they > are applying compiler "printf-like" attribute to functions which are > using their own functions rather than libc printf, and we are purely > making this change to disable the spurious warning because there is > no abort in their %n handling, can't we just disable the annotations > instead?
actually my patch comment is too short, maybe this instead Don't do printf-like checks on Exim's internal printf-like functions on OpenBSD. libc printf functions abort() on %n use, and the compiler has been modified to warn about this; however this does not apply to Exim's functions which do still support %n. > Index: patches/patch-src_mytypes_h > =================================================================== > RCS file: patches/patch-src_mytypes_h > diff -N patches/patch-src_mytypes_h > --- /dev/null 1 Jan 1970 00:00:00 -0000 > +++ patches/patch-src_mytypes_h 20 Sep 2021 09:28:21 -0000 > @@ -0,0 +1,20 @@ > +$OpenBSD$ > + > +Don't do printf-like checks on OpenBSD, which warn about %n use. > + > +Index: src/mytypes.h > +--- src/mytypes.h.orig > ++++ src/mytypes.h > +@@ -33,7 +33,11 @@ local_scan.h includes it and exim.h includes them both > + the arguments of printf-like functions. This is done by a macro. */ > + > + #if defined(__GNUC__) || defined(__clang__) > +-# define PRINTF_FUNCTION(A,B) __attribute__((format(printf,A,B))) > ++# if defined(__OpenBSD__) > ++# define PRINTF_FUNCTION(A,B) /**/ > ++# else > ++# define PRINTF_FUNCTION(A,B) __attribute__((format(printf,A,B))) > ++# endif > + # define ARG_UNUSED __attribute__((__unused__)) > + # define WARN_UNUSED_RESULT __attribute__((__warn_unused_result__)) > + # define ALLOC __attribute__((malloc)) > > >
