[PATCH] [RFC] Fix ptype bug actually exercised in userdef.exp

[PATCH] [RFC] Fix ptype bug actually exercised in userdef.exp

[Pedro Alves]

I happened to notice a bug with ptype &Ref, and found out userdef.exp
actually exercises the bug.  With:

class Container
{
public:
  Member m;

  Member& operator* ();
};

Member& Container::operator* ()
{
  return this->m;
}

And 'c' is of type Container:

(gdb) p c
$1 = {m = {z = -9192}}
(gdb) p *c
$2 = (Member &) @0x7fffffffda20: {z = -9192}
(gdb) ptype *c
type = class Member {
  public:
    int z;
} &

(gdb) p &*c
$3 = (Member *) 0x7fffffffda20

(gdb) ptype &*c
type = class Member {
  public:
    int z;
} &*
(gdb)

Notice that last print (&*c) on says the type is a pointer - that's
how you get the address behind a reference.  But notice the last ptype
instead says the type of the same expression is a pointer _reference_.
This looks like a bug to me.

This patch fixes it.  The issue is that we're entering the VALUE_LVAL
(x) == lval_memory branch by mistake for references.  The fix is just
to swap the tests so references are checked first, like value_addr
also handles references first.

Comments?

Tested on x86_64 Fedora 17.

2013-02-13  Pedro Alves  <palves@redhat.com>

	* eval.c (evaluate_subexp_for_address) <default_case_after_eval,
	EVAL_AVOID_SIDE_EFFECTS>: Swap and handle TYPE_CODE_REF before
	lval_memory.

2013-02-13  Pedro Alves  <palves@redhat.com>

	* gdb.cp/userdef.exp (ptype &*c): Don't expect an &.
---
 gdb/eval.c                       |   10 ++++------
 gdb/testsuite/gdb.cp/userdef.exp |    2 +-
 2 files changed, 5 insertions(+), 7 deletions(-)

diff --git a/gdb/eval.c b/gdb/eval.c
index f4b39cb..96efa21 100644
--- a/gdb/eval.c
+++ b/gdb/eval.c
@@ -2929,12 +2929,10 @@ evaluate_subexp_for_address (struct expression *exp, int *pos,
 	{
 	  struct type *type = check_typedef (value_type (x));
 
-	  if (VALUE_LVAL (x) == lval_memory || value_must_coerce_to_target (x))
-	    return value_zero (lookup_pointer_type (value_type (x)),
-			       not_lval);
-	  else if (TYPE_CODE (type) == TYPE_CODE_REF)
-	    return value_zero (lookup_pointer_type (TYPE_TARGET_TYPE (type)),
-			       not_lval);
+	  if (TYPE_CODE (type) == TYPE_CODE_REF)
+	    return value_zero (lookup_pointer_type (TYPE_TARGET_TYPE (type)), not_lval);
+	  else if (VALUE_LVAL (x) == lval_memory || value_must_coerce_to_target (x))
+	    return value_zero (lookup_pointer_type (value_type (x)), not_lval);
 	  else
 	    error (_("Attempt to take address of "
 		     "value not located in memory."));
diff --git a/gdb/testsuite/gdb.cp/userdef.exp b/gdb/testsuite/gdb.cp/userdef.exp
index 7c20b33..51a1bb5 100644
--- a/gdb/testsuite/gdb.cp/userdef.exp
+++ b/gdb/testsuite/gdb.cp/userdef.exp
@@ -133,7 +133,7 @@ gdb_test "break A2::operator +" ".*Breakpoint $decimal at.*"
 gdb_test "print c" "\\\$\[0-9\]* = {m = {z = .*}}"
 gdb_test "print *c" "\\\$\[0-9\]* = \\(Member &\\) @$hex: {z = .*}"
 gdb_test "print &*c" "\\\$\[0-9\]* = \\(Member \\*\\) $hex"
-gdb_test "ptype &*c" "type = (struct|class) Member {(\[\r\n \]+public:)?\[\r\n \]+int z;\[\r\n\].*} &\\*"
+gdb_test "ptype &*c" "type = (struct|class) Member {(\[\r\n \]+public:)?\[\r\n \]+int z;\[\r\n\].*} \\*"
 
 gdb_test "print operator== (mem1, mem2)" " = false"
 gdb_test "print operator== (mem1, mem1)" " = true"

Re: [PATCH] [RFC] Fix ptype bug actually exercised in userdef.exp

[Pedro Alves]

On 02/13/2013 08:31 PM, Pedro Alves wrote:
> +	  if (TYPE_CODE (type) == TYPE_CODE_REF)
> +	    return value_zero (lookup_pointer_type (TYPE_TARGET_TYPE (type)), not_lval);
> +	  else if (VALUE_LVAL (x) == lval_memory || value_must_coerce_to_target (x))
> +	    return value_zero (lookup_pointer_type (value_type (x)), not_lval);
>  	  else

Whoops, too long lines.  Pretend there's a newline before each of those
"not_lval", like in the present code...

-- 
Pedro Alves

Re: [PATCH] [RFC] Fix ptype bug actually exercised in userdef.exp

[Tom Tromey]

>>>>> "Pedro" == Pedro Alves <palves@redhat.com> writes:

Pedro> This patch fixes it.  The issue is that we're entering the VALUE_LVAL
Pedro> (x) == lval_memory branch by mistake for references.  The fix is just
Pedro> to swap the tests so references are checked first, like value_addr
Pedro> also handles references first.

Pedro> Comments?

It looks reasonable to me.

Tom

Re: [PATCH] [RFC] Fix ptype bug actually exercised in userdef.exp

[Pedro Alves]

On 02/13/2013 08:56 PM, Tom Tromey wrote:
>>>>>> "Pedro" == Pedro Alves <palves@redhat.com> writes:
> 
> Pedro> This patch fixes it.  The issue is that we're entering the VALUE_LVAL
> Pedro> (x) == lval_memory branch by mistake for references.  The fix is just
> Pedro> to swap the tests so references are checked first, like value_addr
> Pedro> also handles references first.
> 
> Pedro> Comments?
> 
> It looks reasonable to me.

Thanks.  Checked in (v2).

-- 
Pedro Alves