On 6/21/17 4:11 PM, Waldemar Brodkorb wrote: > Hi, > Thomas Petazzoni wrote, > >> Hello, >> >> On Wed, 31 May 2017 23:41:26 +0000, Anthony G. Basile wrote: >>> From: "Anthony G. Basile" <bluen...@gentoo.org> >>> >>> man feature_test_macros(7) specifies that _BSD_SOURCE and _SVID_SOURCE have >>> been deprecated in favor of _DEFAULT_SOURCE since libc 2.20. Specifying >>> either of the former is now equivalent to specifying just the latter. We >>> add >>> this macro to conform to this standard, but do not add the compiler warning >>> to maintain full backwards compatibility with earlier version of glibc and >>> uclibc. >>> >>> Signed-off-by: Anthony G. Basile <bluen...@gentoo.org> >> This patch breaks the build of the "knock" package in Buildroot: >> >> >> http://autobuild.buildroot.net/results/908/90863b5adb769a346acd3dc4bbe8d5fa497a0581/build-end.log >> >> src/knockd.c:1296:37: error: 'struct tcphdr' has no member named 'th_flags' >> >> The tcphdr structure is defined in uClibc in two different ways, >> depending on whether _FAVOR_BSD is defined or not. If _FAVOR_BSD is >> defined, then the structure does have this th_flags member, otherwise >> not. >> >> _FAVOR_BSD is defined as follows: >> >> #if defined _BSD_SOURCE && \ >> !(defined _POSIX_SOURCE || defined _POSIX_C_SOURCE || \ >> defined _XOPEN_SOURCE || defined _XOPEN_SOURCE_EXTENDED || \ >> defined _GNU_SOURCE || defined _SVID_SOURCE) >> # define __FAVOR_BSD 1 >> #endif >> >> When being built, the program passes both -D_BSD_SOURCE and >> -D_DEFAULT_SOURCE. However, this patch has the effect that defining >> _DEFAULT_SOURCE defines _SVID_SOURCE, which has the side effect of no >> longer defining _FAVOR_BSD. >> >> The definition of _DEFAULT_SOURCE is that it is equivalent to defining >> _BSD_SOURCE and _SVID_SOURCE. However, it turns out that in the context >> of uClibc, defining _SVID_SOURCE prevents __FAVOR_BSD from being >> defined. >> >> Where is the bug? In what the knock build system is doing, or in uClibc? > As it seems to compile fine for glibc based toolchain, it is a > regression in uClibc-ng. > > best regards > Waldemar
I just did a grep of glibc master/HEAD and it looks like __FAVOR_BSD has been removed. If we remove __FAVOR_BSD from uclibc, will buildroot stuff build fine? You can test by just removing those lines from features.h and forcing undef __FAVOR_BSD. If it works, I can produce a patch. -- Anthony G. Basile, Ph.D. Gentoo Linux Developer [Hardened] E-Mail : bas...@freeharbor.net GnuPG FP : 1FED FAD9 D82C 52A5 3BAB DC79 9384 FA6E F52D 4BBA GnuPG ID : F52D4BBA _______________________________________________ devel mailing list devel@uclibc-ng.org https://mailman.uclibc-ng.org/cgi-bin/mailman/listinfo/devel
