LOCAL_NATIVE_COVERAGE should imply -O0.
Optimization can lead to incorrect coverage results.
Change-Id: I8ade9ab246e292d26ac18821baf9970e29dc1a1f
diff --git a/core/binary.mk b/core/binary.mk
index 7b77fb2..efb0db0 100644
--- a/core/binary.mk
+++ b/core/binary.mk
@@ -259,8 +259,10 @@
my_target_global_ldflags := $($(LOCAL_2ND_ARCH_VAR_PREFIX)TARGET_GLOBAL_LDFLAGS)
endif # my_clang
-# If the global flag NATIVE_COVERAGE is set, my_native_coverage will be true
-# unless the module explicitly sets my_native_coverage := false.
+# To enable coverage for a given module, set LOCAL_NATIVE_COVERAGE=true and
+# build with NATIVE_COVERAGE=true in your enviornment. Note that the build
+# system is not sensitive to changes to NATIVE_COVERAGE, so you should do a
+# clean build of your module after toggling it.
ifeq ($(NATIVE_COVERAGE),true)
ifeq ($(my_native_coverage),true)
# We can't currently generate coverage for clang binaries for two
@@ -276,8 +278,8 @@
# -no-integrated-as. Since most of the assembly in our tree is
# incompatible with clang's assembler, we can't turn off this flag.
ifneq ($(my_clang),true)
- my_target_global_cflags += --coverage
- my_target_global_ldflags += --coverage
+ my_cflags += --coverage -O0
+ my_ldflags += --coverage
endif
endif
else
