www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Modules and debugging in DWARF

reply "Iain Buclaw" <ibuclaw gdcproject.org> writes:
Hi all,

I've been making some changes in GDC that will eventually be made 
updated to be specially handled in GDB when it comes to non-local 
symbol lookup, completion, etc.

Here's the current implementation in a brief nutshell.  It's a 
bit rough, but hopefully someone with knowledge of DMD's 
internals may understand it well enough to update the DMD backend 
to produce the correct debug code.


The the debug generation of 'module foo' emits a DW_TAG_module at 
the file name and location of the module foo.  If there is no 
module declaration in the file, then the location defaults to 
line 1 in the source file.

/* module foo; */
  <1><2d>: Abbrev Number: 2 (DW_TAG_module)
     <2e>   DW_AT_name        : foo
     <32>   DW_AT_decl_file   : 1
     <33>   DW_AT_decl_line   : 2


Packages are handled by emitting the parent tags first, then the 
sub-modules in order.  Children of parent modules are referenced 
using DW_AT_sibling.

/* module std.foo; */
  <1><2d>: Abbrev Number: 2 (DW_TAG_module)
     <2e>   DW_AT_name        : std
     <32>   DW_AT_decl_file   : 1
     <33>   DW_AT_decl_line   : 2
  <2><38>: Abbrev Number: 3 (DW_TAG_module)
     <39>   DW_AT_name        : (indirect string, offset: 0xe0): 
foo
     <3d>   DW_AT_decl_file   : 1
     <3e>   DW_AT_decl_line   : 2
...
/* module std.bar; */
  <1><4d>: Abbrev Number: 2 (DW_TAG_module)
     <4e>   DW_AT_name        : std
     <52>   DW_AT_decl_file   : 2
     <53>   DW_AT_decl_line   : 3
  <2><58>: Abbrev Number: 3 (DW_TAG_module)
     <59>   DW_AT_name        : (indirect string, offset: 0x4b): 
bar
     <5d>   DW_AT_decl_file   : 2
     <5e>   DW_AT_decl_line   : 3

The DW_TAG_module being emitted twice isn't intentional here.  
And more a side effect of the front-end sending two different 
instantiations of the same 'std' package to the backend.  If it 
is genuinely needed, then expect this to change when I come round 
to it.


Imported modules are tagged with DW_TAG_imported_module, and the 
DW_AT_import points to the external module declaration.

/* import object.d.  */
  <3><3f>: Abbrev Number: 4 (DW_TAG_imported_module)
     <40>   DW_AT_decl_file   : 1
     <41>   DW_AT_decl_line   : 1
     <42>   DW_AT_import      : <0x53>   [Abbrev Number: 6 
(DW_TAG_module)]
...
  <1><53>: Abbrev Number: 6 (DW_TAG_module)
     <54>   DW_AT_name        : (indirect string, offset: 0x0): 
object
     <58>   DW_AT_declaration : 1


Renamed imports are identical, but with the addition of 
DW_AT_name to reference the alias.

/* import simd = core.simd;  */
  <3><46>: Abbrev Number: 5 (DW_TAG_imported_module)
     <47>   DW_AT_decl_file   : 1
     <48>   DW_AT_decl_line   : 4
     <49>   DW_AT_name        : (indirect string, offset: 0x81): 
simd
     <4d>   DW_AT_import      : <0x63>   [Abbrev Number: 6 
(DW_TAG_module)]
...
  <1><58>: Abbrev Number: 7 (DW_TAG_module)
     <59>   DW_AT_name        : (indirect string, offset: 0x8d): 
core
     <5d>   DW_AT_decl_file   : 2
     <5e>   DW_AT_decl_line   : 13
     <5f>   DW_AT_sibling     : <0x69>
  <2><63>: Abbrev Number: 6 (DW_TAG_module)
     <64>   DW_AT_name        : (indirect string, offset: 0x81): 
simd
     <68>   DW_AT_declaration : 1


And currently I'm testing selective imports to be represented as 
DW_TAG_imported_declaration.  Don't have any examples for yet, 
but it looks like it would be the imported declaration tag 
pointing to the real declaration.


Regards
Iain.
Feb 21 2014
parent reply "Sarath Kodali" <sarath dummy.com> writes:
On Saturday, 22 February 2014 at 00:30:18 UTC, Iain Buclaw wrote:
 Hi all,
[skip]
 And currently I'm testing selective imports to be represented 
 as DW_TAG_imported_declaration.  Don't have any examples for 
 yet, but it looks like it would be the imported declaration tag 
 pointing to the real declaration.


 Regards
 Iain.
Below is dwarf info generated by g++ for selective imports in C++ code. < 1><0x00000045> DW_TAG_typedef DW_AT_name va_list DW_AT_decl_file 0x00000002 /usr/lib/gcc/i586-suse-linux/4.7/include/stdarg.h DW_AT_decl_line 0x00000066 DW_AT_type <0x00000029> < 1><0x00000050> DW_TAG_namespace DW_AT_name std DW_AT_decl_file 0x00000004 /home/sarath/development/dbg/tests/c/<built-in> DW_AT_decl_line 0x00000000 DW_AT_sibling <0x00000079> < 2><0x0000005b> DW_TAG_imported_declaration DW_AT_decl_file 0x00000005 /usr/include/c++/4.7/cstdarg DW_AT_decl_line 0x00000038 DW_AT_import <0x00000045> - Sarath
Feb 22 2014
parent Iain Buclaw <ibuclaw gdcproject.org> writes:
On 22 February 2014 18:15, Sarath Kodali <sarath dummy.com> wrote:
 On Saturday, 22 February 2014 at 00:30:18 UTC, Iain Buclaw wrote:
 Hi all,
[skip]
 And currently I'm testing selective imports to be represented as
 DW_TAG_imported_declaration.  Don't have any examples for yet, but it looks
 like it would be the imported declaration tag pointing to the real
 declaration.


 Regards
 Iain.
Below is dwarf info generated by g++ for selective imports in C++ code. < 1><0x00000045> DW_TAG_typedef DW_AT_name va_list DW_AT_decl_file 0x00000002 /usr/lib/gcc/i586-suse-linux/4.7/include/stdarg.h DW_AT_decl_line 0x00000066 DW_AT_type <0x00000029> < 1><0x00000050> DW_TAG_namespace DW_AT_name std DW_AT_decl_file 0x00000004 /home/sarath/development/dbg/tests/c/<built-in> DW_AT_decl_line 0x00000000 DW_AT_sibling <0x00000079> < 2><0x0000005b> DW_TAG_imported_declaration DW_AT_decl_file 0x00000005 /usr/include/c++/4.7/cstdarg DW_AT_decl_line 0x00000038 DW_AT_import <0x00000045> - Sarath
Yeah, I don't need to know that per say (although I will have to bear in mind how GDB represents it in its runtime environment). GCC provides a handy debug hooks called 'imported_module_or_decl'. I just need to make sure that the trees passed to it are generated correctly. :-) I just posted this information because this is what gdc does today, and at least DMD developers have the misfortune for having to have to worry about this. :o) Regards Iain.
Feb 22 2014