On 12/10/18 10:30 AM, Kees Cook wrote:
On Sun, Dec 9, 2018 at 3:00 PM Dmitry V. Levin <ldv@xxxxxxxxxxxx> wrote:I've stumbled over the current macro-expand behaviour of the test harness: $ gcc -Wall -xc - <<'__EOF__' TEST(macro) { int status = 0; ASSERT_TRUE(WIFSIGNALED(status)); } TEST_HARNESS_MAIN __EOF__ $ ./a.out [==========] Running 1 tests from 1 test cases. [ RUN ] global.macro <stdin>:4:global.macro:Expected 0 (0) != (((signed char) (((status) & 0x7f) + 1) >> 1) > 0) (0) global.macro: Test terminated by assertion [ FAIL ] global.macro [==========] 0 / 1 tests passed. [ FAILED ] With this change the output of the same test looks much more comprehensible: [==========] Running 1 tests from 1 test cases. [ RUN ] global.macro <stdin>:4:global.macro:Expected 0 (0) != WIFSIGNALED(status) (0) global.macro: Test terminated by assertion [ FAIL ] global.macro [==========] 0 / 1 tests passed. [ FAILED ] The issue is very similar to the bug fixed in glibc assert(3) three years ago: https://sourceware.org/bugzilla/show_bug.cgi?id=18604 Cc: Shuah Khan <shuah@xxxxxxxxxx> Cc: Kees Cook <keescook@xxxxxxxxxxxx> Cc: Andy Lutomirski <luto@xxxxxxxxxxxxxx> Cc: Will Drewry <wad@xxxxxxxxxxxx> Cc: linux-kselftest@xxxxxxxxxxxxxxx Signed-off-by: Dmitry V. Levin <ldv@xxxxxxxxxxxx>Yeah, good idea. Acked-by: Kees Cook <keescook@xxxxxxxxxxxx>
Thanks. Applied to linux-kselftest next. -- Shuah
