gas/config/tc-arm.c:7819: error: array subscript is above array bounds

[Hans-Peter Nilsson]

Compiling gas at 43dd762689eb8 for e.g. arm-eabi using an old gcc like
gcc-4.3 yields a warning that causes more than one concern:

/x/gas/config/tc-arm.c: In function 'parse_operands':
/x/gas/config/tc-arm.c:7819: error: array subscript is above array bounds

That line seems to have been introduced with commit 1b883319, says git
blame.  The author is CC:ed.

7819:	  po_reg_or_fail (REG_TYPE_ZR);

That po_reg_or_fail call is a macro, with the definition containing:

#define po_reg_or_fail(regtype)					\
  do								\
    {								\
      val = arm_typed_reg_parse (& str, regtype, & rtype,	\
				 & inst.operands[i].vectype);	\
      if (val == FAIL)						\
	{							\
	  first_error (_(reg_expected_msgs[regtype]));		\
	  goto failure;						\
	}							\
...

It's the reg_expected_msgs[REG_TYPE_ZR] that causes the warning.
Checking reg_expected_msgs[] shows:

const char * const reg_expected_msgs[] =
{
  [REG_TYPE_RN]	    = N_("ARM register expected"),
...
  [REG_TYPE_MQ]	    = N_("MVE vector register expected"),
  [REG_TYPE_RNB]    = N_("")
 };

That is, no entry for REG_TYPE_ZR, which would make the argument to
first_error NULL if reg_expected_msgs[REG_TYPE_ZR] was within bounds.
But:

enum arm_reg_type
{
  REG_TYPE_RN,
...
  REG_TYPE_RNB,
  REG_TYPE_ZR
};

To wit, the warning seems to be correct.  Now, that warning *can* be
trivially "fixed" with:

diff --git a/gas/config/tc-arm.c b/gas/config/tc-arm.c
index b70028f..1dab180 100644
--- a/gas/config/tc-arm.c
+++ b/gas/config/tc-arm.c
@@ -690,7 +690,8 @@ const char * const reg_expected_msgs[] =
   [REG_TYPE_MMXWCG] = N_("iWMMXt scalar register expected"),
   [REG_TYPE_XSCALE] = N_("XScale accumulator register expected"),
   [REG_TYPE_MQ]	    = N_("MVE vector register expected"),
-  [REG_TYPE_RNB]    = N_("")
+  [REG_TYPE_RNB]    = N_(""),
+  [REG_TYPE_ZR]	    = N_("ZR register expected"),
 };

 /* Some well known registers that we refer to directly elsewhere.  */

...but I'm worrying also about:

- I don't know if that error message would be valid and informative; I
don't know what a "ZR" register is; it's just a copy-paste to me.

- There's apparently no (error) test-case that would have exposed the
missing message.

- This is an *old* gcc, 4.3.  Building with a more recent gcc, say
gcc-6.3, does not yield this warning.  That seems like a regression,
assuming that the warning is as valid as it looks, at a glance.
Perhaps a warning that had too many false positives and was silenced?
Off-topic for this list though.

brgds, H-P