Ddwarf

[MartinNowak]

• emit debug infos for dynamic arrays as
  struct _Array_T { size_t length; T* ptr; }
• emit debug infos for delegates as
  struct _Delegate{ void* ctxptr; Function* funcptr; }
• emit debug infos for aarrays as
  struct _AArray_KeyT_ValueT { void* ptr; }
• remove D DWARF extensions
• remove dead code in toctype.c which changed the ctype to
  structs, the backend could not handle them
• emit same debug infos for -gc as the info is common C

Specifically the _Array_T type will hook to gdb's D support.
That is with gdb >= 2010-04-29 this will add full dynamic
array printing support.

The function pointer in _Delegate is correctly typed and
will be resolved to the actual symbol name during debugging.

For AAs we can't really add debug infos as their implementation
is hidden in druntime.

[MartinNowak]

Fixes Bug 4180
http://d.puremagic.com/issues/show_bug.cgi?id=4180

[WalterBright]

I'd rather #ifdef the D Dwarf extensions out, rather than remove them.

I'd also rather leave in the struct generation stuff as #ifdef, as it is useful for Codeview symbolic debug info generation, though that needs to be looked at in more detail.

[MartinNowak]

Sounds reasonable.

[MartinNowak]

I've rewritten this in a more conservative way.
It now only alters the -g behavior for DWARF.
The old behavior can be enabled with defining
USE_DWARF_D_EXTENSIONS in dwarf.h.
Hopefully it will not rot as the struct code in toctype.c.

It would be great if we'd mangle the types.
Then the typenames could be
_Array_ElementMangle
_AArray_KeyMangle_ValueMangle
_Delegate_TypeFunctionMangle

GDB has a D demangler and can be made to display them accordingly.
I will see how to add this later, there is no D mangler around the backend, is it?

[Marenz]

Do want!

[CyberShadow]

By the way, would it be possible to add a switch to emit DWARF debug info on Windows? Some debuggers/profilers (e.g. Very Sleepy) understand DWARF but not CodeView.

[llucax]

It almost fixes http://d.puremagic.com/issues/show_bug.cgi?id=3391 also

[MartinNowak]

Yeah, but AA support won't work.
I think we're better of to add python scripts for them, but they require a specific binary to be associated with. So this will only get usable with a shared phobos2.

[llucax]

Martin Nowak, el 19 de diciembre a las 04:02 me escribiste:

Yeah, but AA support won't work.
I think we're better of to add python scripts for them, but they
require a specific binary to be associated with. So this will only get
usable with a shared phobos2.

Yeah, STL support is really great using python extensions... Would be
great having something like that for D.

Leandro Lucarella (AKA luca) http://llucax.com.ar/