Now that -m64 + LLD + MINGW works, the question is obvious.

The background idea is of course to allow the use of GDB under Windows. At first glance the debug info API is not uniform. This is visible if you compare dmd.backend.dwarf and dmd.backend.cgcv and the naive idea which would be to replace all the calls starting with 'cv8_' with their "dwarf_" equivalent won't work (let's say to test in a first time, later a new compiler switch could be added to select the debug info format).

On Friday, 17 August 2018 at 19:28:47 UTC, Basile B. wrote:
> Now that -m64 + LLD + MINGW works, the question is obvious.
>
> The background idea is of course to allow the use of GDB under Windows. At first glance the debug info API is not uniform. This is visible if you compare dmd.backend.dwarf and dmd.backend.cgcv and the naive idea which would be to replace all the calls starting with 'cv8_' with their "dwarf_" equivalent won't work (let's say to test in a first time, later a new compiler switch could be added to select the debug info format).

Conctretly the idea is to do that :

------------------------- src/dmd/backend/mscoffobj.c -------------------------
index 45f1f4219..1b1d34376 100644
@@ -26,8 +26,13 @@
 #include        "mach.h"
 #include        "outbuf.h"
 #include        "filespec.h"
-#include        "cv4.h"
-#include        "cgcv.h"
+
+//#include        "cv4.h"
+//#include        "cgcv.h"
+
+#include        "dwarf2.h"
+#include        "dwarf.h"
+
 #include        "dt.h"

 #include        "aa.h"
@@ -430,7 +435,8 @@ MsCoffObj *MsCoffObj::init(Outbuffer *objbuf, const char *filename, const char *
     assert(SegData[UDATA]->SDseg == UDATA);

     if (config.fulltypes)
-        cv8_initfile(filename);
+        //cv8_initfile(filename);
+        dwarf_initfile(filename);
     assert(objbuf->size() == 0);
     return obj;
 }
@@ -464,7 +470,8 @@ void MsCoffObj::initfile(const char *filename, const char *csegname, const char
     }
 #endif
     if (config.fulltypes)
-        cv8_initmodule(filename, modname);
+        //cv8_initmodule(filename, modname);
+        dwarf_initmodule(filename, modname);
 }

 /************************************
@@ -670,7 +677,8 @@ void MsCoffObj::termfile()
     //dbg_printf("MsCoffObj::termfile\n");
     if (configv.addlinenumbers)
     {
-        cv8_termmodule();
+        //cv8_termmodule();
+        dwarf_termmodule();
     }
 }

@@ -692,7 +700,8 @@ void MsCoffObj::term(const char *objfilename)

     if (configv.addlinenumbers)
     {
-        cv8_termfile(objfilename);
+        //cv8_termfile(objfilename);
+        dwarf_termfile(objfilename);
     }

 #if SCPP
@@ -1032,7 +1041,7 @@ void MsCoffObj::linnum(Srcpos srcpos, int seg, targ_size_t offset)
     if (srcpos.Slinnum == 0 || !srcpos.Sfilename)
         return;

-    cv8_linnum(srcpos, offset);
+    //cv8_linnum(srcpos, offset);
 }


@@ -1850,7 +1859,8 @@ void MsCoffObj::func_start(Symbol *sfunc)
     sfunc->Soffset = Offset(cseg);

     if (config.fulltypes)
-        cv8_func_start(sfunc);
+        //cv8_func_start(sfunc);
+        dward_func_start(sfunc);
 }

 /*******************************
@@ -1863,7 +1873,8 @@ void MsCoffObj::func_term(Symbol *sfunc)
 //          sfunc->Sident, sfunc->Soffset,Offset(cseg),sfunc->Sxtrnnum);

     if (config.fulltypes)
-        cv8_func_term(sfunc);
+        //cv8_func_term(sfunc);
+        dwarf_func_term(sfunc);
 }

 /********************************


However this doesn't build for now and i don't get error messages for some reasons (dmc sterrr not redirected ?)

On Friday, 17 August 2018 at 19:35:40 UTC, Basile B. wrote:
> On Friday, 17 August 2018 at 19:28:47 UTC, Basile B. wrote:
>> Now that -m64 + LLD + MINGW works, the question is obvious.
>>
>> The background idea is of course to allow the use of GDB under Windows. At first glance the debug info API is not uniform. This is visible if you compare dmd.backend.dwarf and dmd.backend.cgcv and the naive idea which would be to replace all the calls starting with 'cv8_' with their "dwarf_" equivalent won't work (let's say to test in a first time, later a new compiler switch could be added to select the debug info format).

Just to make sure, do you know Mago-Mi? It is the Mago debug engine with a GDB compatible interface. That means at every place GBD is supported, also Mago-Mi can be used.
For example IntelliJ D Language plugin or Visual Studio Code or DlangIDE...

Unfortunately Mago-Mi has some issues, written down here
https://wiki.dlang.org/SAOC_2018_ideas#Multi_IDE_debugger_support_.28for_windows.29

Kind regards
André

On Friday, 17 August 2018 at 19:45:35 UTC, Andre Pany wrote:
> On Friday, 17 August 2018 at 19:35:40 UTC, Basile B. wrote:
>> On Friday, 17 August 2018 at 19:28:47 UTC, Basile B. wrote:
>>> Now that -m64 + LLD + MINGW works, the question is obvious.
>>>
>>> The background idea is of course to allow the use of GDB under Windows. [...]
>
> Just to make sure, do you know Mago-Mi?

Yes however this doesn't mean that it's not worth generating DWARF info if it's technically possible.

On Friday, 17 August 2018 at 19:28:47 UTC, Basile B. wrote:
> Now that -m64 + LLD + MINGW works, the question is obvious.
>
> The background idea is of course to allow the use of GDB under Windows. At first glance the debug info API is not uniform. This is visible if you compare dmd.backend.dwarf and dmd.backend.cgcv and the naive idea which would be to replace all the calls starting with 'cv8_' with their "dwarf_" equivalent won't work (let's say to test in a first time, later a new compiler switch could be added to select the debug info format).

Actually i expected an answer with someone with expertise in the object formats and debug infos, for example @Walter Bright or @Rainer Shutze, what do you think ?