[Bug tree-optimization/36631] [4.3/4.4 Regression] attribute always_inline -> sorry, unimplemented: recursive inlining

["rguenth at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #5 from rguenth at gcc dot gnu dot org  2008-09-15 14:26 -------
This looks like a frontend bug to me.  For B::g we record in the DECL_ARGUMENTS

(gdb) call debug_tree (p)
 <parm_decl 0x7fae5daa2f30 c
    type <record_type 0x7fae5d15da80 C addressable needs-constructing type_1
type_5 BLK

with DECL_ARG_TYPE (p) being

 <record_type 0x7fae5d15da80 C addressable needs-constructing type_1 type_5 BLK
    size <integer_cst 0x7fae5da9c7b0 type <integer_type 0x7fae5daaf0c0
bit_size_type> constant invariant 8>

but we pass it a reference in trigger():

(gdb) call debug_tree (*$1)
 <addr_expr 0x7fae5d15f580
    type <reference_type 0x7fae5d163540
        type <record_type 0x7fae5d15da80 C addressable needs-constructing
type_1 type_5 BLK

like so:

(gdb) call debug_generic_expr (*expr_p)
g (&b, &TARGET_EXPR <D.2065, <<< Unknown tree: aggr_init_expr
  5
  __comp_ctor 
  D.2065
  0B
  (const struct C &) (const struct C *) (struct C *) c >>>
>)


which leads to (rightfully) CALL_CANNOT_INLINE_P set.  The TYPE_ARG_TYPES
also require a record for the arguments.


So either the function call is wrong or DECL_ARG_TYPE and TYPE_ARG_TYPES
are not correct.  The offending frontend code doing this is
convert_for_arg_passing which has

  /* Pass classes with copy ctors by invisible reference.  */
  else if (TREE_ADDRESSABLE (type))
    val = build1 (ADDR_EXPR, build_reference_type (type), val);

but it doesn't communicate this to the middle-end.

Jason, can you shed some light on when/if/why not C++ is setting
DECL_ARG_TYPE to what?  Thanks.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jason at gcc dot gnu dot org
         AssignedTo|rguenth at gcc dot gnu dot  |unassigned at gcc dot gnu
                   |org                         |dot org
             Status|ASSIGNED                    |NEW


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36631