On 3/21/19 9:43 AM, Richard Biener wrote:
On Thu, 21 Mar 2019, Richard Biener wrote:
This also avoids the ICE in PR89779 but IMHO is not a real fix.
Still it restores a previously active check against released SSA names
which now have error_mark_node type rather than NULL. The new way
opens up consolidation so I've adjusted tree_nop_conversion plus
operand_equal_p which got a defensive check at the same time
(and checks for error_mark_node right before that check).
Bootstrap & regtest running on x86_64-unknown-linux-gnu.
The testcase comes with the other patch (but still included below).
Ugh, looks like my defense installed for released SSA names is now
relied upon by the C++ FE via
#2 0x0000000001864a96 in tree_strip_nop_conversions (
exp=<convert_expr 0x7ffff69bc920>)
at /space/rguenther/src/svn/trunk2/gcc/tree.c:12848
#3 0x0000000000ab2d2f in iterative_hash_template_arg (
arg=<convert_expr 0x7ffff69bc920>, val=3363282904)
at /space/rguenther/src/svn/trunk2/gcc/cp/pt.c:1751
for
<convert_expr 0x7ffff69bc920
type <integer_type 0x7ffff68985e8 int public type_6 SI
size <integer_cst 0x7ffff689c078 constant 32>
unit-size <integer_cst 0x7ffff689c090 constant 4>
align:32 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff68985e8 precision:32 min <integer_cst 0x7ffff689c030 -2147483648>
max <integer_cst 0x7ffff689c048 2147483647>
pointer_to_this <pointer_type 0x7ffff68a19d8>>
arg:0 <scope_ref 0x7ffff69d0b40 tree_0
arg:0 <template_type_parm 0x7ffff69d17e0 i tree_0 type_0 type_6
VOID
align:8 warn_if_not_align:0 symtab:0 alias-set -1
canonical-type 0x7ffff69d1150
index 0 level 1 orig_level 1
chain <type_decl 0x7ffff69d2098 i>>
arg:1 <identifier_node 0x7ffff69c39c0 c
normal local bindings <(nil)>>>>
where both the SCOPE_REF and TEMPLATE_TYPE_PARM have NULL
TREE_TYPE.
So I'll restore the NULL check in tree_nop_conversion.
But since the C++ FE overloads almost every tree field possible
I wonder whether that STRIP_NOPS in iterative_hash_template_arg
is a good idea?
I think it still makes sense; TREE_TYPE of an expression is either a
type, error_mark_node, or NULL_TREE (for type-dependent expressions in a
template).
Also I hope the FE doesn't call into operand_equal_p with such
typed trees.
It shouldn't, we have cp_tree_equal to use instead.
Jason