https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65858
--- Comment #2 from prathamesh3492 at gcc dot gnu.org --- Hi, Sorry for late response, I applied your patch and got the following ICE: -std=gnu++11 -flto --param lto-partitions=1 Full command line options: http://pastebin.com/6JSWH9YM ../../third_party/cld_2/src/internal/cld_generated_cjk_uni_prop_80.cc:7132:1: internal compiler error: in output_constructor, at lto-streamer-out.c:2152 } // End namespace CLD2 ^ 0xb25609 output_constructor /home/prathamesh.kulkarni/gnu-toolchain/src/gcc.git/gcc/lto-streamer-out.c:2152 0xb25609 lto_output() /home/prathamesh.kulkarni/gnu-toolchain/src/gcc.git/gcc/lto-streamer-out.c:2354 0xb7e6c1 write_lto /home/prathamesh.kulkarni/gnu-toolchain/src/gcc.git/gcc/passes.c:2395 0xb81c0e ipa_write_summaries_1 /home/prathamesh.kulkarni/gnu-toolchain/src/gcc.git/gcc/passes.c:2459 0xb81c0e ipa_write_summaries() /home/prathamesh.kulkarni/gnu-toolchain/src/gcc.git/gcc/passes.c:2519 0x877dda ipa_passes /home/prathamesh.kulkarni/gnu-toolchain/src/gcc.git/gcc/cgraphunit.c:2216 0x877dda symbol_table::compile() /home/prathamesh.kulkarni/gnu-toolchain/src/gcc.git/gcc/cgraphunit.c:2312 0x879d3c symbol_table::finalize_compilation_unit() /home/prathamesh.kulkarni/gnu-toolchain/src/gcc.git/gcc/cgraphunit.c:2461 0x645c40 cp_write_global_declarations() /home/prathamesh.kulkarni/gnu-toolchain/src/gcc.git/gcc/cp/decl2.c:4757 Reduced test-case: It appears that r222249, which fixed PR65801 introduced this issue. The following test-case also ICE's for x86. int x { 0.5 }; int main() { return 0; } prathamesh.kulkarni@ex40-01:~$ fsf-toolchain/bin/g++ -flto braced-init.cpp -c -Wno-narrowing -std=gnu++11 prathamesh.kulkarni@ex40-01:~$ fsf-toolchain/bin/g++ -flto braced-init.o lto1: internal compiler error: in get_constructor, at varpool.c:331 0xd22c43 varpool_node::get_constructor() ../../src/gcc/varpool.c:331 0xd23af8 varpool_node::assemble_decl() ../../src/gcc/varpool.c:602 0x6b8633 output_in_order ../../src/gcc/cgraphunit.c:2137 0x6b8b23 symbol_table::compile() ../../src/gcc/cgraphunit.c:2378 0x62b025 lto_main() ../../src/gcc/lto/lto.c:3496 I think it happens because, check_narrowing() returns false and it's caller cp/semantics.c:finish_compound_literal returns error_mark_node, which then gets streamed into object file. Conside this part of check_narrowing(): if (!ok) { if (cxx_dialect == cxx98) warning_at (EXPR_LOC_OR_LOC (init, input_location), OPT_Wnarrowing, "narrowing conversion of %qE from %qT to %qT inside { } " "is ill-formed in C++11", init, ftype, type); else if (!TREE_CONSTANT (init)) { if (complain & tf_warning_or_error) { pedwarn (EXPR_LOC_OR_LOC (init, input_location), OPT_Wnarrowing, "narrowing conversion of %qE from %qT to %qT inside { }", init, ftype, type); ok = true; } } else if (complain & tf_error) { global_dc->pedantic_errors = 1; pedwarn (EXPR_LOC_OR_LOC (init, input_location), OPT_Wnarrowing, "narrowing conversion of %qE from %qT to %qT inside { }", init, ftype, type); global_dc->pedantic_errors = flag_pedantic_errors; } } return cxx_dialect == cxx98 || ok; For the above test-case ok is false and we enter the nested else if (complain & tf_error) block. pedwarn() doesn't print anything here and returns 0. That's because the following condition becomes true in diagnostic.c:diagnostic_report_diagnostic(): /* This tests if the user provided the appropriate -Wfoo or -Wno-foo option. */ if (! context->option_enabled (diagnostic->option_index, context->option_state)) return false; So diagnostic_report_diagnostic() returns false to pedwarn() which then returns 0 to check_narrowing() and warning is not printed. return cxx_dialect == cxx98 || ok; Since we are compiling with gnu++11 mode, cxx_dialect is not cxx98 and ok is false, hence check_narrowing() returns false. cp/semantics.c:finish_compound_literal() calls check_narrowing(): if (SCALAR_TYPE_P (type) && !BRACE_ENCLOSED_INITIALIZER_P (compound_literal) && !check_narrowing (type, compound_literal, complain)) return error_mark_node; I suppose that check_narrowing() returns false here and the above if condition becomes true and finish_compound_literal() returns error_mark_node which is then streamed. Is that correct ? I tried with the following untested patch, which prevents the ICE, but not sure if it's the right approach: diff --git a/gcc/cp/typeck2.c b/gcc/cp/typeck2.c index 884957b..2a25c20 100644 --- a/gcc/cp/typeck2.c +++ b/gcc/cp/typeck2.c @@ -872,8 +872,7 @@ check_narrowing (tree type, tree init, tsubst_flags_t complain) bool ok = true; REAL_VALUE_TYPE d; - if (((!warn_narrowing || !(complain & tf_warning)) - && cxx_dialect == cxx98) + if ((!warn_narrowing || !(complain & tf_warning)) || !ARITHMETIC_TYPE_P (type)) return ok; Since we want -Wno-narrowing to also silence C++11 stricter narrowing rules (PR65801), I removed condition on cxx_dialect == cxx98. Thank you, Prathamesh
