| Message ID | 20161014173321.GM7282@tucnak.redhat.com |
|---|---|
| State | New |
| Headers | show |
On 10/14/2016 01:33 PM, Jakub Jelinek wrote: > This also uses the infrastructure of the langhook patch I've sent earlier. > It emits (if not strict dwarf) DW_AT_inline on explicit or implicit inline > variables, and also tweaks dwarf2out so that for inline static data members > we consider in-class declarations as definitions (emit DW_AT_linkage_name > and no DW_AT_declaration) for them. Hmm, so there's no DW_TAG_variable for the inline static data member? That seems problematic to me. The DWARF 3 convention that static data members use DW_TAG_member seems rather misguided, since in fact they are variables. Why did that change? Jason
On Mon, Oct 31, 2016 at 09:52:28AM -0400, Jason Merrill wrote: > On 10/14/2016 01:33 PM, Jakub Jelinek wrote: > >This also uses the infrastructure of the langhook patch I've sent earlier. > >It emits (if not strict dwarf) DW_AT_inline on explicit or implicit inline > >variables, and also tweaks dwarf2out so that for inline static data members > >we consider in-class declarations as definitions (emit DW_AT_linkage_name > >and no DW_AT_declaration) for them. > > Hmm, so there's no DW_TAG_variable for the inline static data member? That > seems problematic to me. The DWARF 3 convention that static data members > use DW_TAG_member seems rather misguided, since in fact they are variables. > Why did that change? I had to bisect it, apparently it has been changed by myself in r145770, rationale in thread starting at https://gcc.gnu.org/ml/gcc-patches/2009-04/msg00039.html The current DWARF 5 wording is: "If the variable entry represents the defining declaration for a C++ static data member of a structure, class or union, the entry has a DW_AT_specification attribute, whose value is a reference to the debugging information entry representing the declaration of this data member. The referenced entry has the tag DW_TAG_member and will be a child of some class, structure or union type entry." on page 98 in DWARF5_Public_Review.pdf. And page 117 has in 5.7.6: "A data member (as opposed to a member function) is represented by a debugging information entry with the tag DW_TAG_member." So, using DW_TAG_variable as DW_TAG_{structure,class}_type children for the non-inline static data members would violate these two. C++17 is outside of the scope of DWARF5 I guess, so the question is to find out what do we want for those and agree on that with consumers and other producers. Incidentally, I've filed today a DWARF issue that Appendix A doesn't list for DW_TAG_member lots of attributes that are allowed for DW_TAG_variable and are useful for static data members. Some of them we want anyway, at least unless we agree on the non-inline static data members no longer be DW_TAG_member in the structure/class type and change the standard. In particular DW_AT_const_value and DW_AT_const_expr are desirable even for the static data members in C++98..14 (or non-inline ones in C++17). The rest, like DW_AT_linkage_name, DW_AT_location, etc. are generally more useful mainly for the inline ones. Jakub
On Mon, Oct 31, 2016 at 10:25 AM, Jakub Jelinek <jakub@redhat.com> wrote: > On Mon, Oct 31, 2016 at 09:52:28AM -0400, Jason Merrill wrote: >> On 10/14/2016 01:33 PM, Jakub Jelinek wrote: >> >This also uses the infrastructure of the langhook patch I've sent earlier. >> >It emits (if not strict dwarf) DW_AT_inline on explicit or implicit inline >> >variables, and also tweaks dwarf2out so that for inline static data members >> >we consider in-class declarations as definitions (emit DW_AT_linkage_name >> >and no DW_AT_declaration) for them. >> >> Hmm, so there's no DW_TAG_variable for the inline static data member? That >> seems problematic to me. The DWARF 3 convention that static data members >> use DW_TAG_member seems rather misguided, since in fact they are variables. >> Why did that change? > > I had to bisect it, apparently it has been changed by myself in > r145770, rationale in thread starting at > https://gcc.gnu.org/ml/gcc-patches/2009-04/msg00039.html > > The current DWARF 5 wording is: > "If the variable entry represents the defining declaration for a C++ static data > member of a structure, class or union, the entry has a DW_AT_specification > attribute, whose value is a reference to the debugging information entry > representing the declaration of this data member. The referenced entry has > the tag DW_TAG_member and will be a child of some class, structure or > union type entry." on page 98 in DWARF5_Public_Review.pdf. Yes, this changed in DWARF 3; DWARF 2 didn't specify the tag. I think this was a mistake. > Incidentally, I've filed today a DWARF issue that Appendix A doesn't list > for DW_TAG_member lots of attributes that are allowed for DW_TAG_variable > and are useful for static data members. Using DW_TAG_variable would address that, too. Jason
--- gcc/dwarf2out.c.jj 2016-10-14 15:22:57.000000000 +0200 +++ gcc/dwarf2out.c 2016-10-14 16:39:12.226917016 +0200 @@ -18896,7 +18896,10 @@ add_linkage_name (dw_die_ref die, tree d && VAR_OR_FUNCTION_DECL_P (decl) && TREE_PUBLIC (decl) && !(VAR_P (decl) && DECL_REGISTER (decl)) - && die->die_tag != DW_TAG_member) + && (die->die_tag != DW_TAG_member + || (VAR_P (decl) + && (lang_hooks.decls.decl_dwarf_attribute (decl, DW_AT_inline) + != -1)))) add_linkage_name_raw (die, decl); } @@ -21382,6 +21385,20 @@ gen_variable_die (tree decl, tree origin return; } + /* For static data members, the declaration (or definition for inline + variables) in the class is supposed to have DW_TAG_member tag; + the specification if any should still be DW_TAG_variable referencing + the DW_TAG_member DIE. */ + enum dwarf_tag tag = DW_TAG_variable; + if (declaration && class_scope_p (context_die)) + { + tag = DW_TAG_member; + /* Inline static data members are defined inside of the class. */ + if (lang_hooks.decls.decl_dwarf_attribute (decl_or_origin, DW_AT_inline) + != -1) + declaration = false; + } + if (old_die) { if (declaration) @@ -21414,14 +21431,7 @@ gen_variable_die (tree decl, tree origin goto gen_variable_die_location; } } - - /* For static data members, the declaration in the class is supposed - to have DW_TAG_member tag; the specification should still be - DW_TAG_variable referencing the DW_TAG_member DIE. */ - if (declaration && class_scope_p (context_die)) - var_die = new_die (DW_TAG_member, context_die, decl); - else - var_die = new_die (DW_TAG_variable, context_die, decl); + var_die = new_die (tag, context_die, decl); if (origin != NULL) origin_die = add_abstract_origin_attribute (var_die, origin); @@ -21521,6 +21531,17 @@ gen_variable_die (tree decl, tree origin && (origin_die == NULL || get_AT (origin_die, DW_AT_const_expr) == NULL) && !specialization_p) add_AT_flag (var_die, DW_AT_const_expr, 1); + + if (!dwarf_strict) + { + int inl = lang_hooks.decls.decl_dwarf_attribute (decl_or_origin, + DW_AT_inline); + if (inl != -1 + && !get_AT (var_die, DW_AT_inline) + && (origin_die == NULL || get_AT (origin_die, DW_AT_inline) == NULL) + && !specialization_p) + add_AT_unsigned (var_die, DW_AT_inline, inl); + } } /* Generate a DIE to represent a named constant. */ --- gcc/cp/cp-objcp-common.c.jj 2016-10-14 15:01:27.000000000 +0200 +++ gcc/cp/cp-objcp-common.c 2016-10-14 15:46:14.650303379 +0200 @@ -173,6 +173,16 @@ cp_decl_dwarf_attribute (const_tree decl return 1; break; + case DW_AT_inline: + if (VAR_P (decl) && DECL_INLINE_VAR_P (decl)) + { + if (DECL_VAR_DECLARED_INLINE_P (decl)) + return DW_INL_declared_inlined; + else + return DW_INL_inlined; + } + break; + default: break; } --- gcc/testsuite/g++.dg/debug/dwarf2/inline-var-1.C.jj 2016-10-14 16:55:30.345512927 +0200 +++ gcc/testsuite/g++.dg/debug/dwarf2/inline-var-1.C 2016-10-14 16:56:45.704558635 +0200 @@ -0,0 +1,26 @@ +// { dg-do compile } +// { dg-options "-O -std=c++1z -g -dA -gno-strict-dwarf" } +// { dg-require-weak "" } +// { dg-final { scan-assembler-times "0x3\[^\n\r]* DW_AT_inline" 6 } } +// { dg-final { scan-assembler-times "0x1\[^\n\r]* DW_AT_inline" 2 } } +// { dg-final { scan-assembler-not " DW_AT_declaration" } } +// { dg-final { scan-assembler-times " DW_AT_\[^\n\r]*linkage_name" 7 } } + +inline int a; +struct S +{ + static inline double b = 4.0; + static constexpr int c = 2; + static constexpr inline char d = 3; +} s; +template <int N> +inline int e = N; +int &f = e<2>; +template <int N> +struct T +{ + static inline double g = 4.0; + static constexpr int h = 2; + static inline constexpr char i = 3; +}; +T<5> t;