Re: Martin, Matt, all: Don't compare signed against unsigned, please

Stephen> Dunno; does this tell you anything? SJT> Sorry, it was late (it's later now...). I can read `info gcc', really SJT> I can.... SJT> -Wsign-compare SJT> Martin: the data destruction _was_ your bug (identified as yours by SJT> lack of a ChangeLog, and being in the same CVS diff as a Martin SJT> mega-patch that was in the ChangeLog, 1998-03-22, eliminate MAC_* SJT> macros). That is the only one in lstream.c, though. SJT> Matt: regex.c is chock full of warnings. font-lock.c, search.c, and SJT> symbol.c have a few too. syntax.c has none, though. I don't know if SJT> any of those functions use a "negative value is end of loop" SJT> convention, but that would explain inflooping on selected platforms if SJT> they're there. SJT> **** ObPatch (not on -patches because I don't know if this is the SJT> **** right way to munge configure.in): I don't want -Wsign-compare in the default CFLAGS. Too many warnings, and too annoying to fix. I don't really want to get warnings for this code: void foo (unsigned int s) { int i; for (i = 0; i <= s; i++) ... } Of course, it's valuable to use -Wsign-compare to audit existing problems such as ssize_t, etc. I myself have fixed many many -Wsign-compare warnings. If you really really want to add this warning flag, you better fix all the warnings first; else folks will start to ignore other warnings. SJT> Index: configure.in SJT> =================================================================== SJT> RCS file: /usr/CVSroot/XEmacs/xemacs/configure.in,v SJT> retrieving revision 1.111.2.147 SJT> diff -u -r1.111.2.147 configure.in SJT> --- configure.in 2001/02/13 06:34:39 1.111.2.147 SJT> +++ configure.in 2001/02/15 15:15:11 SJT> ＠＠ -1879,6 +1879,10 ＠＠ SJT> dnl Should we take debugging options into consideration? SJT> if test "$GCC" = "yes"; then SJT> CFLAGS="-g -O3 -Wall -Wno-switch -Winline -Wmissing-prototypes -Wshadow" SJT> + dnl gcc 2.95.3 seems to have changed size_t to unsigned, screwing up SJT> + dnl comparisons between buffer offsets and bytes written values (which SJT> + dnl are often negative to indicate failure) SJT> + CFLAGS="$CFLAGS -Wsign-compare" SJT> dnl glibc is intentionally not `-Wpointer-arith'-clean. SJT> dnl Ulrich Drepper has rejected patches to fix the glibc header files. SJT> test "$have_glibc" != "yes" && CFLAGS="$CFLAGS -Wpointer-arith"