On 10/23/18 1:19 PM, Richard Biener wrote:
> On Tue, Oct 23, 2018 at 10:59 AM Martin Liška <mli...@suse.cz> wrote:
>>
>> Hi.
>>
>> I've returned to this long-lasting issue after quite some time. Thanks to
>> Honza I hope
>> I can now address the root cause which caused output of a string constant
>> when debug info
>> was emitted. The problematic situation happened with following back-trace:
>>
>> #0 mergeable_string_section (decl=<string_cst 0x7ffff67be210>, align=64,
>> flags=0) at /home/marxin/Programming/gcc/gcc/varasm.c:808
>> #1 0x0000000001779bf3 in default_elf_select_section (decl=<string_cst
>> 0x7ffff67be210>, reloc=0, align=64) at
>> /home/marxin/Programming/gcc/gcc/varasm.c:6739
>> #2 0x000000000176efb6 in get_constant_section (exp=<string_cst
>> 0x7ffff67be210>, align=64) at /home/marxin/Programming/gcc/gcc/varasm.c:3302
>> #3 0x000000000176f468 in build_constant_desc (exp=<string_cst
>> 0x7ffff67be210>) at /home/marxin/Programming/gcc/gcc/varasm.c:3371
>> #4 0x000000000176f81c in output_constant_def (exp=<string_cst
>> 0x7ffff67be210>, defer=1) at /home/marxin/Programming/gcc/gcc/varasm.c:3434
>> #5 0x000000000176d406 in decode_addr_const (exp=<addr_expr 0x7ffff682f8c0>,
>> value=0x7fffffffc540) at /home/marxin/Programming/gcc/gcc/varasm.c:2951
>> #6 0x000000000176d93f in const_hash_1 (exp=<addr_expr 0x7ffff682f8c0>) at
>> /home/marxin/Programming/gcc/gcc/varasm.c:3054
>> #7 0x000000000176fdc2 in lookup_constant_def (exp=<addr_expr
>> 0x7ffff682f8c0>) at /home/marxin/Programming/gcc/gcc/varasm.c:3557
>> #8 0x0000000000dd5778 in cst_pool_loc_descr (loc=<addr_expr
>> 0x7ffff682f8c0>) at /home/marxin/Programming/gcc/gcc/dwarf2out.c:17288
>>
>> That was in situation where we emit debug info of a function that has an
>> inlined __PRETTY_FUNCTION__ from
>> a different function. As seen, the constant is output due to const_hash_1
>> function call. Proper fix would
>> be to not emit these string constants for purpose of hash function.
>
> possibly sth like the following - that avoids all cases of calling
> output_constant_def. Probably worth testing
> separately.
Thanks for the hint. I can confirm it works for the pretty function patch.
I also tested the patch separately on current trunk. May I install the hunk
right now, or should I wait?
Thanks,
Martin
>
> diff --git a/gcc/varasm.c b/gcc/varasm.c
> index 91650eea9f7..9121dbd2c84 100644
> --- a/gcc/varasm.c
> +++ b/gcc/varasm.c
> @@ -3047,6 +3047,10 @@ const_hash_1 (const tree exp)
> }
>
> case ADDR_EXPR:
> + if (CONSTANT_CLASS_P (TREE_OPERAND (exp, 0)))
> + return const_hash_1 (TREE_OPERAND (exp, 0));
> +
> + /* Fallthru. */
> case FDESC_EXPR:
> {
> struct addr_const value;
>
>> However, I still see some minor ICEs, it's probably related to
>> decay_conversion in cp_fname_init:
>>
>> 1) ./xg++ -B.
>> /home/marxin/Programming/gcc/gcc/testsuite/g++.dg/cpp0x/lambda/lambda-__func__2.C
>>
>> /home/marxin/Programming/gcc/gcc/testsuite/g++.dg/cpp0x/lambda/lambda-__func__2.C:6:17:
>> internal compiler error: Segmentation fault
>> 6 | [] { return __func__; }();
>> | ^~~~~~~~
>> 0x1344568 crash_signal
>> /home/marxin/Programming/gcc/gcc/toplev.c:325
>> 0x7ffff6bc310f ???
>>
>> /usr/src/debug/glibc-2.27-6.1.x86_64/signal/../sysdeps/unix/sysv/linux/x86_64/sigaction.c:0
>> 0x9db134 is_capture_proxy(tree_node*)
>> /home/marxin/Programming/gcc/gcc/cp/lambda.c:261
>> 0xaeecb7 tsubst_expr(tree_node*, tree_node*, int, tree_node*, bool)
>> /home/marxin/Programming/gcc/gcc/cp/pt.c:16700
>> 0xaee5fd tsubst_expr(tree_node*, tree_node*, int, tree_node*, bool)
>> /home/marxin/Programming/gcc/gcc/cp/pt.c:16636
>> 0xaf0ffb tsubst_expr(tree_node*, tree_node*, int, tree_node*, bool)
>> /home/marxin/Programming/gcc/gcc/cp/pt.c:16942
>>
>> where
>> (gdb) p debug_tree(decl)
>> <var_decl 0x7ffff7fee480 __func__
>> type <pointer_type 0x7ffff69b5540
>> type <integer_type 0x7ffff69b5498 char readonly unsigned string-flag
>> type_6 QI
>> size <integer_cst 0x7ffff698df78 constant 8>
>> unit-size <integer_cst 0x7ffff698df90 constant 1>
>> align:8 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
>> 0x7ffff69b5498 precision:8 min <integer_cst 0x7ffff698dfd8 0> max
>> <integer_cst 0x7ffff698dfc0 255>
>> pointer_to_this <pointer_type 0x7ffff69b5540>>
>> unsigned DI
>> size <integer_cst 0x7ffff698de88 constant 64>
>> unit-size <integer_cst 0x7ffff698dea0 constant 8>
>> align:64 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
>> 0x7ffff69b5540>
>> readonly used tree_2 unsigned DI
>> /home/marxin/Programming/gcc/gcc/testsuite/g++.dg/cpp0x/lambda/lambda-__func__2.C:6:17
>> size <integer_cst 0x7ffff698de88 64> unit-size <integer_cst 0x7ffff698dea0
>> 8>
>> align:64 warn_if_not_align:0 context <function_decl 0x7ffff67abd00
>> operator()>
>> value-expr <nop_expr 0x7ffff67b7660 type <pointer_type 0x7ffff69b5540>
>> constant
>> arg:0 <addr_expr 0x7ffff67b7640 type <pointer_type 0x7ffff67bd3f0>
>> constant
>> arg:0 <string_cst 0x7ffff67b7620 type <array_type 0x7ffff67bd348>
>> constant "operator()\000">>>>
>>
>> and
>> #0 0x00000000009db134 in is_capture_proxy (decl=<tree 0x0>) at
>> /home/marxin/Programming/gcc/gcc/cp/lambda.c:261
>>
>> 2 ) ./xg++ -B.
>> /home/marxin/Programming/gcc/gcc/testsuite/g++.dg/ext/pretty2.C -c
>> /home/marxin/Programming/gcc/gcc/testsuite/g++.dg/ext/pretty2.C:16:24:
>> internal compiler error: Segmentation fault
>> 16 | __PRETTY_FUNCTION__), 0))
>> | ^
>> /home/marxin/Programming/gcc/gcc/testsuite/g++.dg/ext/pretty2.C:36:10: note:
>> in expansion of macro ‘assert’
>> 36 | int a = (assert (foo ()), 1);
>> | ^~~~~~
>> 0x1344568 crash_signal
>> /home/marxin/Programming/gcc/gcc/toplev.c:325
>> 0x7ffff6bc310f ???
>>
>> /usr/src/debug/glibc-2.27-6.1.x86_64/signal/../sysdeps/unix/sysv/linux/x86_64/sigaction.c:0
>> 0x9db270 is_capture_proxy(tree_node*)
>> /home/marxin/Programming/gcc/gcc/cp/lambda.c:265
>> 0x9dbad9 is_normal_capture_proxy(tree_node*)
>> /home/marxin/Programming/gcc/gcc/cp/lambda.c:274
>> 0x9c1fe6 mark_use(tree_node*, bool, bool, unsigned int, bool)
>> /home/marxin/Programming/gcc/gcc/cp/expr.c:114
>> 0x89d9ab convert_like_real
>> /home/marxin/Programming/gcc/gcc/cp/call.c:6905
>>
>> where:
>>
>> (gdb) p debug_tree(decl)
>> <var_decl 0x7ffff7fee870 __PRETTY_FUNCTION__
>> type <pointer_type 0x7ffff69b4540
>> type <integer_type 0x7ffff69b4498 char readonly unsigned string-flag
>> type_6 QI
>> size <integer_cst 0x7ffff698df78 constant 8>
>> unit-size <integer_cst 0x7ffff698df90 constant 1>
>> align:8 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
>> 0x7ffff69b4498 precision:8 min <integer_cst 0x7ffff698dfd8 0> max
>> <integer_cst 0x7ffff698dfc0 255>
>> pointer_to_this <pointer_type 0x7ffff69b4540>>
>> unsigned type_6 DI
>> size <integer_cst 0x7ffff698de88 constant 64>
>> unit-size <integer_cst 0x7ffff698dea0 constant 8>
>> align:64 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
>> 0x7ffff69b4540>
>> readonly used tree_1 tree_2 unsigned read decl_6 DI
>> /home/marxin/Programming/gcc/gcc/testsuite/g++.dg/ext/pretty2.C:36:10 size
>> <integer_cst 0x7ffff698de88 64> unit-size <integer_cst 0x7ffff698dea0 8>
>> align:64 warn_if_not_align:0
>> value-expr <nop_expr 0x7ffff67b69a0 type <pointer_type 0x7ffff69b4540>
>> constant
>> arg:0 <addr_expr 0x7ffff67b6980 type <pointer_type 0x7ffff67c3738>
>> constant
>> arg:0 <string_cst 0x7ffff67b6960 type <array_type 0x7ffff67c3690>
>> constant "top level\000">>> chain <var_decl 0x7ffff7fee7e0
>> a>>
>>
>> Can please a C++ maintainer help me with that? Having that I hope I can
>> finish the patch.
>>
>> Thank you,
>> Martin
