[lldb] [clang] Reland "[clang][DebugInfo] Emit global variable definitions for static data members with constant initializers" (PR #71780)

[via cfe-commits]

dyung wrote:

> > Hi @Michael137, we are seeing a failure in one of our internal tests that I bisected back to this change. Consider the following code:
> > ```c++
> > struct X
> > {
> >     static const int constant = 1;
> >     int x;
> > 
> >     X() { x = constant; }
> > };
> > const int X::constant;
> > 
> > int main()
> > {
> >     X x;
> >     x.x = X::constant;
> >     x.x = X::constant;
> >     x.x = X::constant;
> >     x.x = X::constant;
> >     x.x = X::constant;
> >     return 0;
> > }
> > ```
> > 
> > 
> >     
> >       
> >     
> > 
> >       
> >     
> > 
> >     
> >   
> > Prior to your change, the compiler would generate the following DWARF for the constant value:
> > ```
> > 0x0000003a:     DW_TAG_member
> >                   DW_AT_name    ("constant")
> >                   DW_AT_type    (0x00000057 "const int")
> >                   DW_AT_decl_file       ("/home/dyung/sandbox/test.cpp")
> >                   DW_AT_decl_line       (3)
> >                   DW_AT_external        (true)
> >                   DW_AT_declaration     (true)
> >                   DW_AT_const_value     (1)
> > ```
> > 
> > 
> >     
> >       
> >     
> > 
> >       
> >     
> > 
> >     
> >   
> > After your change, the DW_AT_const_value is gone from this DW_TAG_member group, but doesn't appear anywhere else in the DWARF output which seems to indicate that it was dropped completely which does not seem to be correct. Is this intended or am I missing something?
> 
> Right, we stopped putting the `DW_AT_const_value` on the declaration. Instead we put either the location or the constant on the definition, depending on what's available. In your example you define the variable out-of-class which would generate a `DW_TAG_variable` with a location, so we don't emit the constant for it on the definition:
> 
> ```
> 0x0000002e:   DW_TAG_variable
>                 DW_AT_specification     (0x0000004a "constant")
>                 DW_AT_location  (DW_OP_addr 0x90)
>                 DW_AT_linkage_name      ("_ZN1X8constantE")
> ```
> 
> What's the nature of the failure? Would you instead be able to read the value out of the definition?

Our test is expecting to find a DW_AT_const_value somewhere in the dwarf output which it no longer seems to generate. From what I gather from your commit message, the compiler should now be emitting a new DW_TAG_variable section that it previously did not with the DW_AT_const_value attached? Or is my understanding incorrect?

https://github.com/llvm/llvm-project/pull/71780