blob: 9471df32e91a75368493ad798b1568021cf6cd70 [file] [log] [blame]
We were seeing reproducibility issues where one host would use the internal
logwtmp wrapper, another would use the one in libutil. The issue was that in
some cases the "\#include" was making it to CC, in others, "#include". The
issue seems to be related to shell escaping.
The root cause looks to be:
http://git.savannah.gnu.org/cgit/make.git/commit/?id=c6966b323811c37acedff05b576b907b06aea5f4
Instead of relying on shell quoting, use make to indirect the variable
and avoid the problem.
See https://github.com/paulusmack/ppp/issues/233
Upstream-Status: Backport [https://github.com/paulusmack/ppp/commit/b4430f7092ececdff2504d5f3393a4c6528c3686]
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Index: ppp-2.4.9/pppd/Makefile.linux
===================================================================
--- ppp-2.4.9.orig/pppd/Makefile.linux
+++ ppp-2.4.9/pppd/Makefile.linux
@@ -80,7 +80,8 @@ PLUGIN=y
#USE_SRP=y
# Use libutil; test if logwtmp is declared in <utmp.h> to detect
-ifeq ($(shell echo '\#include <utmp.h>' | $(CC) -E - 2>/dev/null | grep -q logwtmp && echo yes),yes)
+UTMPHEADER = "\#include <utmp.h>"
+ifeq ($(shell echo $(UTMPHEADER) | $(CC) -E - 2>/dev/null | grep -q logwtmp && echo yes),yes)
USE_LIBUTIL=y
endif
@@ -143,7 +144,8 @@ CFLAGS += -DHAS_SHADOW
#LIBS += -lshadow $(LIBS)
endif
-ifeq ($(shell echo '\#include <crypt.h>' | $(CC) -E - >/dev/null 2>&1 && echo yes),yes)
+CRYPTHEADER = "\#include <utmp.h>"
+ifeq ($(shell echo $(CRYPTHEADER) | $(CC) -E - >/dev/null 2>&1 && echo yes),yes)
CFLAGS += -DHAVE_CRYPT_H=1
LIBS += -lcrypt
endif