diff options
| author | Stephan Bergmann <sbergman@redhat.com> | 2014-03-14 12:01:42 +0100 |
|---|---|---|
| committer | Stephan Bergmann <sbergman@redhat.com> | 2014-03-14 12:20:42 +0100 |
| commit | ddfe9eec96ffe322b4952c25e2c3209da3e44a39 (patch) | |
| tree | 964a730aa755295e52323e343d6cdcacac6a3069 | |
| parent | 3ceaeff1cfbd606c4839832d8558617c0513be1d (diff) | |
Don't Valgrind into java children in unit tests
...similarly to desktop/scripts/soffice.sh preventing that for LO itself.
One problem is that, even if it does not find any errors, Valgrind writes onto
a traced child's stderr, and when jvmfwk searches for a JRE to use, it takes
output to stderr into account, so would start to behave differently when run
under Valgrind.
--trace-children-skip appears to be new since Valgrind 3.6, and older versions
would probably fail early when they see this unknown-to-them option. If this
turns out to be a problem in practice (current version is 3.9), we probably need
to make this conditional on a configure.ac check.
Change-Id: Ia6a72bdcd666d68ed0539170f3fc476292e82b96
| -rw-r--r-- | solenv/gbuild/CppunitTest.mk | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/solenv/gbuild/CppunitTest.mk b/solenv/gbuild/CppunitTest.mk index ba3ddac11570..a227540bcfc8 100644 --- a/solenv/gbuild/CppunitTest.mk +++ b/solenv/gbuild/CppunitTest.mk @@ -27,7 +27,7 @@ gb_CppunitTest__interactive := $(true) endif ifneq ($(strip $(VALGRIND)),) -gb_CppunitTest_VALGRINDTOOL := valgrind --tool=$(VALGRIND) --num-callers=50 --error-exitcode=1 --trace-children=yes --leak-check=no +gb_CppunitTest_VALGRINDTOOL := valgrind --tool=$(VALGRIND) --num-callers=50 --error-exitcode=1 --trace-children=yes --trace-children-skip='*/java,*/gij' --leak-check=no ifeq ($(strip $(VALGRIND)),memcheck) G_SLICE := always-malloc GLIBCXX_FORCE_NEW := 1 |