[DTrace-devel] [PATCH 2/4] test: Skip trace() of a 1-byte struct

[eugene.loh at oracle.com]

From: Eugene Loh <eugene.loh at oracle.com>

With commit 3a551bfd ("trace: fix char-array handling"), this test
started to FAIL.  Meanwhile, the behavior of trace() on a 1-byte
struct is poorly defined.  Users wishing clear semantics should use
print() or other actions.

Skip the test.

Signed-off-by: Eugene Loh <eugene.loh at oracle.com>
---
 .../actions/trace/tst.struct-1-byte.d         | 24 +++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/test/unittest/actions/trace/tst.struct-1-byte.d b/test/unittest/actions/trace/tst.struct-1-byte.d
index 36de485f8..8bfcbdd53 100644
--- a/test/unittest/actions/trace/tst.struct-1-byte.d
+++ b/test/unittest/actions/trace/tst.struct-1-byte.d
@@ -5,6 +5,30 @@
  * http://oss.oracle.com/licenses/upl.
  */
 
+/*
+ * With Solaris or legacy DTrace on Linux, the script gives
+ *       52
+ * That is, contents are dumped as an integer because the trace()
+ * argument is 1, 2, 4, or 8 bytes -- specifically, it is 1 byte.
+ *
+ * Before commit 3a551bfd ("trace: fix char-array handling"), we got
+ *           0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f  0123456789abcdef
+ *       0: 34                                               4
+ * That is, contents were dumped as raw bytes.
+ *
+ * With that patch, we get:
+ *       4
+ * That is, contents are dumped as the string "4" since the contents
+ * are a printable char followed optionally by NUL bytes.
+ *
+ * The truth is that the semantics of trace() are poorly defined.
+ * So we are hard-pressed to say what is correct.  The test has
+ * little value.  Skip it.
+ *
+ * Users who want type-aware printing can use the print() action.
+ */
+/* @@skip: poorly defined semantics */
+
 /*
  * ASSERTION: The trace() action prints a struct { int8_t } correctly.
  *
-- 
2.43.5