www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - How to debug and watch globals in windows debugger?

reply ryuukk_ <ryuukk.dev gmail.com> writes:
I tried to look around, what ever i try, nothing works

There is a mention of "fully qualified" here: 
https://forum.dlang.org/post/peb1jj$tr$1 digitalmars.com

But what does that mean?

I tried this:

![screenshot](https://i.imgur.com/JZooIic.pngg)

But it doesn't work
Mar 26 2023
next sibling parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
Anyone got an idea?

Executable built with : ``dmd -g -debug`` on windows

On Visual Studio with the visuald addon i can debug the 
executable and inspect the globals

On VSCode i can debug the executable and inspect locals, but not 
globals

What does visuald does under the hood to be able to inspect 
globals?

No, i don't want to use the slow and bloated multi gigabytes 
Visual Studio just to inspect a variable

I'm starting to loose hope
Mar 27 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
If anyone know what is the problem exactly, then please speak up, 
this problem needs to be reported to the right place fixed so it 
can be fixed, all languages i tested don't have this issue (c, 
rust, zig, odin)

D is not a toy language, let's take this issue seriously, shall 
we? or is it?..
Mar 27 2023
parent reply "Richard (Rikki) Andrew Cattermole" <richard cattermole.co.nz> writes:
The person you need is WebFreak and they are not online right now.
Mar 27 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
On Monday, 27 March 2023 at 19:31:24 UTC, Richard (Rikki) Andrew 
Cattermole wrote:

 The person you need is WebFreak and they are not online right 
 now.



On IRC earlier today

```
[6:43:29 pm] <WebFreak[m]> nope, I haven't gotten globals to work 
in debugger before
[6:43:33 pm] <WebFreak[m]> I don't think they are emitted
[6:44:02 pm] <WebFreak[m]> on linux I think it might work with 
CodeLLDB, depending on what compiler you used to compile the 
executable
```

I asked about mode details (who didn't emit what?) and where to 
report the issue, so far question left unanswered


My list of "gripes and wishes" grows larger and larger everyday
Mar 27 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
To clarify, i tested both dmd/ldc, they both doesn't work on 
windows

To clarify even more:

I am a mere user who wants to debug its program on windows using 
vscode's debugger wich uses msvc

I can debug a c/rust/zig executable without issue and see globals

The problem only happens with dmd/ldc

Is it broken by design for decade and nothing should and could be 
done about it?

Is the only solution to inspect a global variable is to use 
visual studio and install visuald plugin?

If i tell you what i'm trying to debug i bet you'll laugh 
(loading dll compiled using dmd), but that's a story for another 
day, 1 frustration a day is plenty
Mar 27 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
Two questions for the D team:

Does globals have a special syntax? as you can see above i also 
tried the mangled name inside the PDB (using hex viewer 
https://github.com/WerWolv/ImHex)

Walter, how do you debug your programs written in D?
Mar 27 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
Fore more clarity, all syntax i have tried:

```
type_to_tiledef
kshared.defs.type_to_tiledef
_D7kshared4defs15type_to_tiledef
kshared:defs:type_to_tiledef
kshared::defs::type_to_tiledef
kshared->defs->type_to_tiledef
```
Mar 27 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
I've now waste an entire day trying to figure out what's wrong, 
perhaps trusted D for my projects was a bad idea, i now look like 
a fool
Mar 27 2023
parent WebFreak001 <d.forum webfreak.org> writes:
On Tuesday, 28 March 2023 at 01:04:19 UTC, ryuukk_ wrote:

 I've now waste an entire day trying to figure out what's wrong, 
 perhaps trusted D for my projects was a bad idea, i now look 
 like a fool


sorry to hear that, I haven't really been looking to much into 
the debugging problems here yet. My extension mostly just 
integrates the existing tools here (Visual Studio Debugger / GDB 
+ whatever is emitted in the pdb files)

What I was talking about is that it's possible dmd didn't emit 
the globals information in any way that the GDB / Visual Studio 
Debugger understand right now. I'm not a debugger expert myself, 
looking into this is probably something that would take a bit of 
research about the DWARF / PDB formats and checking how DMD / LDC 
emit it. I think you can probably get some great help here from 
Martin Kinkelin. (I think the windows and linux debug info 
emission works differently / are different modules, so there 
might be discrepancies here)

I can help you point towards where to check and who to ask here, 
but I don't think I can help you actually solve this issue right 
now. I haven't really been focusing too much on the debugger, 
especially not on the implementation of debug symbol emissions 
and the debugger itself. I have only been adding a few scripts 
here to get the best out the debugger that we currently have. 
More things require more investment into actually improving the 
project.

I'm quite glad about your contributions you have been doing so 
far and I think the issues you are opening are good to find stuff 
to work on. (the DCD PRs you have open right now are very nice 
and appreciated, but need a bit of work still done to them before 
they can be merged. I have reviewed them, but didn't have time to 
actually make the changes and test them yet - if you want to get 
these changes in faster you can also implement or reply to my 
review comments)

Further people you could ask about here:
- Iain Buclaw (GDC developer), can probably tell you about what 
gcc does here and possibly has insights into the D backend code 
that generates debug info
- Martin Kinkelin (LDC developer), ditto
- Luís Ferreira (ljmf00 on discord) - has been working on LLDB 
for SAoC and implementing a bunch of things - if things are 
missing on the debugger side, he can probably help you with LLDB 
implementations (you probably only need to go this far if we want 
to add support for new custom data structure things that D has - 
most things can already be realized with the debug symbols)
Mar 28 2023
prev sibling parent reply "Richard (Rikki) Andrew Cattermole" <richard cattermole.co.nz> writes:
Have you tried installing mago?

https://github.com/rainers/mago

There are instructions for vs-code in README.
Mar 27 2023
next sibling parent reply "Richard (Rikki) Andrew Cattermole" <richard cattermole.co.nz> writes:
On 28/03/2023 2:06 PM, Richard (Rikki) Andrew Cattermole wrote:

 Have you tried installing mago?
 
 https://github.com/rainers/mago
 
 There are instructions for vs-code in README.


Binaries are available in the installer of VisualD, they can be 
extracted by 7zip (I've checked everything is in there).
Mar 27 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
On Tuesday, 28 March 2023 at 01:21:02 UTC, Richard (Rikki) Andrew 
Cattermole wrote:

 On 28/03/2023 2:06 PM, Richard (Rikki) Andrew Cattermole wrote:

 Have you tried installing mago?
 
 https://github.com/rainers/mago
 
 There are instructions for vs-code in README.


 Binaries are available in the installer of VisualD, they can be 
 extracted by 7zip (I've checked everything is in there).


what is: "Manual Installation of the Concord extension in VS 
Code", i can't find it anywhere

what are: "MagoNatCC.dll, MagoNatCC.vsdconfig and 
.vsdbg-config.json"? i can only find: "MagoNatDE.dll"

So far no bueno
Mar 27 2023
parent "Richard (Rikki) Andrew Cattermole" <richard cattermole.co.nz> writes:
On 28/03/2023 2:33 PM, ryuukk_ wrote:

 what is: "Manual Installation of the Concord extension in VS Code", i 
 can't find it anywhere


Extension is mago, but don't forget: cpptools-1.4.0 or later needed.


 what are: "MagoNatCC.dll, MagoNatCC.vsdconfig and .vsdbg-config.json"? i 
 can only find: "MagoNatDE.dll"


Look in $1..\Packages\Debugger

https://github.com/rainers/mago/blob/0b2ce787114a71b4088ebf14eb922c5b3c46f5be/EED/MagoNatCC/.vsdbg-config.json#L4
Mar 27 2023
prev sibling parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
On Tuesday, 28 March 2023 at 01:06:50 UTC, Richard (Rikki) Andrew 
Cattermole wrote:

 Have you tried installing mago?

 https://github.com/rainers/mago

 There are instructions for vs-code in README.


I did not try mago, but it shouldn't be needed as pdb is 
universally understood by tools

It also doesn't work with remedybg for example:

https://i.imgur.com/P1Ph3Jh.png

While i will try with mago, it shouldn't satisfy us, workarounds 
are temporary solutions

This issue being fixed should be the goal, i opened a ticket for 
it: https://issues.dlang.org/show_bug.cgi?id=23811
Mar 27 2023
parent reply "Richard (Rikki) Andrew Cattermole" <richard cattermole.co.nz> writes:
On 28/03/2023 2:25 PM, ryuukk_ wrote:

 On Tuesday, 28 March 2023 at 01:06:50 UTC, Richard (Rikki) Andrew 
 Cattermole wrote:

 Have you tried installing mago?

 https://github.com/rainers/mago

 There are instructions for vs-code in README.


 
 I did not try mago, but it shouldn't be needed as pdb is universally 
 understood by tools


Agreed, I would expect globals (including TLS) to work without a 
debugger extension.


 While i will try with mago, it shouldn't satisfy us, workarounds are 
 temporary solutions


Its not a workaround, mago (in this case) is an extension to MS's 
debugger framework which provides D specific features, you'd want it 
regardless if you work with the debugger with D a lot. Debuggers 
typically have language specific features to make them easier to work with.

Jan should be shipping this I think.
Mar 27 2023
next sibling parent WebFreak001 <d.forum webfreak.org> writes:
On Tuesday, 28 March 2023 at 04:22:24 UTC, Richard (Rikki) Andrew 
Cattermole wrote:

 On 28/03/2023 2:25 PM, ryuukk_ wrote:

 On Tuesday, 28 March 2023 at 01:06:50 UTC, Richard (Rikki) 
 Andrew Cattermole wrote:

 Have you tried installing mago?

 https://github.com/rainers/mago

 There are instructions for vs-code in README.


 
 I did not try mago, but it shouldn't be needed as pdb is 
 universally understood by tools


 Agreed, I would expect globals (including TLS) to work without 
 a debugger extension.


 While i will try with mago, it shouldn't satisfy us, 
 workarounds are temporary solutions


 Its not a workaround, mago (in this case) is an extension to 
 MS's debugger framework which provides D specific features, 
 you'd want it regardless if you work with the debugger with D a 
 lot. Debuggers typically have language specific features to 
 make them easier to work with.

 Jan should be shipping this I think.


I have looked into trying to get mago to work so far, but didn't 
have any working state yet, which is why that is currently stuck. 
(+ I don't personally use Windows for development, so I don't 
have much usage experience for code-d / debugging integration, 
outside when I actively develop Windows fixes on it)

If you get mago to work with vscode feel free to tell me how you 
did it so we can add a debug configuration preset for code-d 
users / other users who want to manually use it.
Mar 28 2023
prev sibling parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
On Tuesday, 28 March 2023 at 04:22:24 UTC, Richard (Rikki) Andrew 
Cattermole wrote:

 On 28/03/2023 2:25 PM, ryuukk_ wrote:

 On Tuesday, 28 March 2023 at 01:06:50 UTC, Richard (Rikki) 
 Andrew Cattermole wrote:

 Have you tried installing mago?

 https://github.com/rainers/mago

 There are instructions for vs-code in README.


 
 I did not try mago, but it shouldn't be needed as pdb is 
 universally understood by tools


 Agreed, I would expect globals (including TLS) to work without 
 a debugger extension.


 While i will try with mago, it shouldn't satisfy us, 
 workarounds are temporary solutions


 Its not a workaround, mago (in this case) is an extension to 
 MS's debugger framework which provides D specific features, 
 you'd want it regardless if you work with the debugger with D a 
 lot. Debuggers typically have language specific features to 
 make them easier to work with.

 Jan should be shipping this I think.


It is a workaround, it should work already out of the box with 
existing tools

It works for other languages, why not for D? problem

Check here:

https://godbolt.org/z/nYGfMTqYY


int global; on C++ has !dbg

int global; on D doesn't have !dbg

This is a bug in D, therefore it should be fixed
Mar 28 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
On Tuesday, 28 March 2023 at 11:07:02 UTC, ryuukk_ wrote:

 On Tuesday, 28 March 2023 at 04:22:24 UTC, Richard (Rikki) 
 Andrew Cattermole wrote:

 On 28/03/2023 2:25 PM, ryuukk_ wrote:

 On Tuesday, 28 March 2023 at 01:06:50 UTC, Richard (Rikki) 
 Andrew Cattermole wrote:

 Have you tried installing mago?

 https://github.com/rainers/mago

 There are instructions for vs-code in README.


 
 I did not try mago, but it shouldn't be needed as pdb is 
 universally understood by tools


 Agreed, I would expect globals (including TLS) to work without 
 a debugger extension.


 While i will try with mago, it shouldn't satisfy us, 
 workarounds are temporary solutions


 Its not a workaround, mago (in this case) is an extension to 
 MS's debugger framework which provides D specific features, 
 you'd want it regardless if you work with the debugger with D 
 a lot. Debuggers typically have language specific features to 
 make them easier to work with.

 Jan should be shipping this I think.


 It is a workaround, it should work already out of the box with 
 existing tools

 It works for other languages, why not for D? problem

 Check here:

 https://godbolt.org/z/nYGfMTqYY


 int global; on C++ has !dbg

 int global; on D doesn't have !dbg

 This is a bug in D, therefore it should be fixed


Nevermind the godbolt link
Mar 28 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
I opened an issue on microsoft's github, let's see what they have 
to say: https://github.com/microsoft/vscode-cpptools/issues/10751
Mar 28 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
C program:

```C
int notice_me_global = -1;

void main()
{
     notice_me_global = -5;
}
```

``cl app_c.c  /DEBUG /Zi /EHsc /std:c11 /link /DEBUG 
/out:app_c.exe``

``llvm-pdbutil.exe dump --globals app_c.pdb > dump_c``

```
   688476 | S_GDATA32 [size = 32] `notice_me_global`
            type = 0x0074 (int), addr = 0003:0040
```






D program:

```D
__gshared int notice_me_global = -1;

extern(C) void main()
{
     notice_me_global = 5;
     int* a;
     *a = 1;
}
```

``dmd -betterC -m64 -g -debug app_d.d``

``llvm-pdbutil.exe dump --globals app_d.pdb > dump_d``

```
        0 | S_GDATA32 [size = 40] `app_d.notice_me_global`
            type = 0x0074 (int), addr = 0003:3504
```



Open question:


Why is it:

For D: ``0 |``
For C: ``688476 |``


Potential issue:

Debugger doesn't understand: the ``.`` in the name 
``app_d.notice_me_global``, mangled name also doesn't work

That's the info i provided in the microsoft tracker, maybe some 
people here can notice something else that's wrong


dump_c: 
https://gist.github.com/ryuukk/40c6d7d1cd3d6a010242d505380fe233
dump_d: 
https://gist.github.com/ryuukk/38f43383025907a8f44049503887206c
Mar 28 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
Ready to test folder: 
https://github.com/microsoft/vscode-cpptools/issues/10751#issuecomment-1486948783

Contains simple source to reproduce the issue + build script

Also contains binaries + dump in case you just want to see the 
data

Hopefully we can figure that out
Mar 28 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
On Tuesday, 28 March 2023 at 14:02:39 UTC, ryuukk_ wrote:

 Ready to test folder: 
 https://github.com/microsoft/vscode-cpptools/issues/10751#issuecomment-1486948783

 Contains simple source to reproduce the issue + build script

 Also contains binaries + dump in case you just want to see the 
 data

 Hopefully we can figure that out


I noticed in the pdb, there are no mention of the ``app_d`` module


I can only find:

```
       40 | S_PROCREF [size = 28] `app_d.main`
            module = 1, sum name = 0, offset = 40
```

and

```
        0 | S_GDATA32 [size = 40] `app_d.notice_me_global`
            type = 0x0074 (int), addr = 0003:3504
```

Is that why the debugger can't inspect global variables?

Why doesn't the pdb contain anything about ``app_d``?
Mar 28 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
On linux everything works properly out of the box, i tested with 
gdb and i can inspect globals

More info + linux dump on my comment: 
https://github.com/microsoft/vscode-cpptools/issues/10751#issuecomment-1487694435

So it's a problem either with msvc, or the way dmd/ldc write 
information for the linker?

Anyways..

linux: working out of the box

Windows: poop

Not a good outlook, then we wonder why people move away, 
debugging is essential

Am i the only want who want to improve things, is it a lost cause?

Should i move on?
Mar 28 2023
parent reply Steven Schveighoffer <schveiguy gmail.com> writes:
On 3/28/23 6:42 PM, ryuukk_ wrote:


 
 Am i the only want who want to improve things, is it a lost cause?


I'm not one to use a debugger often, but it is very helpful for many 
people. I lament that the Windows debugging situation out of the box is 
dreadful. The students I teach D to I don't show the debugger because 
I'm worried it would confuse them more than help.

I both don't use Windows and mostly do debugging via writeln, so for 
something that doesn't affect me much, I tend not to put effort in. I'm 
hoping someone who is motivated enough will do it.

If I recall correctly, visual studio debugging is better than the other 
solutions. But I don't know having never got debugging to work. I've 
heard that mago is supposed to work better, and was written for D, but 
is Visual Studio dependent: https://github.com/rainers/mago

I wonder if there isn't some way to make this work well for other IDEs? 
I couldn't get it to work on VSCode.


 Should i move on?


In any case, the debugging experience isn't perfect, but seems like a 
drastic measure to leave D altogether if you can't debug globals.

-Steve
Mar 29 2023
parent reply ryuukk_ <ryuukk.dev gmail.com> writes:
On Wednesday, 29 March 2023 at 15:26:21 UTC, Steven Schveighoffer 
wrote:

 On 3/28/23 6:42 PM, ryuukk_ wrote:


 
 Am i the only want who want to improve things, is it a lost 
 cause?


 I'm not one to use a debugger often, but it is very helpful for 
 many people. I lament that the Windows debugging situation out 
 of the box is dreadful. The students I teach D to I don't show 
 the debugger because I'm worried it would confuse them more 
 than help.


I only use windows because i'm making a game and i have to test 
on that platform


 I both don't use Windows and mostly do debugging via writeln, 
 so for something that doesn't affect me much, I tend not to put 
 effort in. I'm hoping someone who is motivated enough will do 
 it.


I am motivated, hopefully someone more knowledgeable can join me 
figure that out


 If I recall correctly, visual studio debugging is better than 
 the other solutions. But I don't know having never got 
 debugging to work. I've heard that mago is supposed to work 
 better, and was written for D, but is Visual Studio dependent: 
 https://github.com/rainers/mago

 I wonder if there isn't some way to make this work well for 
 other IDEs? I couldn't get it to work on VSCode.


That's unfortunate that the solution is bound to a proprietary 
editor that requires a license

Effort should be made in the open and be universal, so it's easy 
to port/adapt to other tools/platforms

Coordinated efforts is the way to go, Go/Rust/Zig people are 
demonstrating this very well and it helped their growth



 Should i move on?


 In any case, the debugging experience isn't perfect, but seems 
 like a drastic measure to leave D altogether if you can't debug 
 globals.

 -Steve


I exaggerate a little bit i must admit it, but I'm trying to 
raise awareness, when other languages have no problem with 
debugging out of the box on popular IDEs, why should it have 
problems only with D?

Currently, at least on windows, there are 2 problems with 
debugging:


- global variables
- extern(C) variables

I'm willing to find the cause for the 1st one, then i'll attempt 
the 2nd one, so expect more frustrated posts

I should emphasis again that the reason why i'm into this mess is 
due to a bug i'm trying to debug related to loading a shared DLL 
built using DMD on windows, wich apparently is known to be broken 
for YEARS.. great, more people should voice their complains, D is 
not a new language..

So what to expect other than frustration? specially when people 
keep recommending to use "workarounds", well, if the fix is a 
workaround, one shouldn't be surprised to see people getting 
frustrated every so often

Also it's more of an accumulation of problems that makes me very 
annoyed

https://issues.dlang.org/show_bug.cgi?id=20737

This for example shouldn't be a thing in 2023, period
Mar 29 2023
next sibling parent "Richard (Rikki) Andrew Cattermole" <richard cattermole.co.nz> writes:
On 30/03/2023 10:09 AM, ryuukk_ wrote:

 more people should voice their complains


Making noise about problems in D.learn is not the place to do it, I 
don't think compiler developers tend to monitor it much.

Show your analysis of the situation in General, that is monitored. 
Especially since the fix for this stuff is probably pretty simple for TLS.
Mar 29 2023
prev sibling parent Salih Dincer <salihdb hotmail.com> writes:
On Wednesday, 29 March 2023 at 21:09:57 UTC, ryuukk_ wrote:

 Also it's more of an accumulation of problems that makes me 
 very annoyed

 https://issues.dlang.org/show_bug.cgi?id=20737

 This for example shouldn't be a thing in 2023, period


I agree with you because it's such a simple problem and it's 
surprising if it exists.  I've never used -betterC though, but 
it's a pity.

SDB 79
Mar 29 2023