Debugger_gdbmi plugin

[Pecan]

Installation Guide: Debugger GDB/MI Plugin
A CodeBlocks plugin using the machine interface of the GDB debugger.

Source is available at github Source
Preparation

• Clone or download a zip
• Cloning creates a folder named debugger_gdbmi
• A zip file needs to be unzipped in a 'debugger_gdbmi' folder.

.cbp files included:

• debugger_gdbmi_wx32_64WK - compiles to the currently active src\devel_32_64 folder.
  
• debugger_gdbmi_wx32_64      - compiles to ..\..\..\devel32_64
          (use only when debugger_gdbmi resides within the contrib workspace.
• debugger_gdbmi_wx32_64Nly - compiles to $(#cb)\devel32_64
          (CodeBlocks will ask you to define the global 'cb' macro to point to CodeBlocks source folder.
• Note: The other .cbp(s) are defined as if debugger_gdbmi resides within the contrib workspace.
           Move debugger_gdbmi into that workspace or modify the .cbp(s)

Installation

• Load and compile debugger_gdbmi with the desired .cbp
• Close CodeBlocks and run src/update32_64
• You should find debugger_gdbmi.dll in the src\output32_64\share\CodeBlocks\plugins folder
• You should find debugger.gdbmi.zip in the src\output32_64\share\CodeBlocks folder

Configuration

     After installation, restart CodeBlocks and go to Settings->Debugger->GDB/MI default

• Set the executable Path to your gdb.exe.
• Verify the other options meet your requirements.
• Click on the Ok button.

Usage
    Go to MainMenu->Debug->Active Debuggers and select GDB/MI default.

Report issues in the Help or Plugins Development forum thread with your Build log and OS.

[killerbot]

wouldn't we add it to our normal source tree, as a contrib plug-in ?

[Pecan]

wouldn't we add it to our normal source tree, as a contrib plug-in ?

Yes, but I'm trying to determine if there is any interest in it.
I've found that adding a plugin to the contribs and especially updating autoMake is very painful and error prone.
I'd rather avoid the pain if there's no interest.

[MaxGaspa]

Yes, but I'm trying to determine if there is any interest in it.

As far as I'm concerned I'm very interested to try a modern GDB/MI instead of using the old annotation mechanism. CB seems to be the last IDE still using the annotation mechanism. CLion, CodeLite. KDevelop, Eclipse, QTcreator...are all using GDB/MI.

[SharkCZ]

I like that, thanks for your efforts in this area.

[ollydbg]

Code

[debug]ActionsMap::Run -> starting action: 0000000011aa5f50 id: 21
[debug]cmd==>210000000000-var-update 1 *
[debug]output==>210000000000^done,changelist=[{name="var5",value="std::vector of length 5, capacity 5",in_scope="true",type_changed="false",new_num_children="5",displayhint="array",dynamic="1",has_more="0"}]
[debug]cmd==>210000000001-var-list-children 2 "var5"
[debug]unparsable_output==>(gdb)
[debug]output==>210000000001^done,numchild="5",displayhint="array",children=[child={name="var5.[0]",exp="[0]",numchild="0",value="1",type="int"},child={name="var5.[1]",exp="[1]",numchild="0",value="2",type="int"},child={name="var5.[2]",exp="[2]",numchild="0",value="3",type="int"},child={name="var5.[3]",exp="[3]",numchild="0",value="4",type="int"},child={name="var5.[4]",exp="[4]",numchild="0",value="5",type="int"}],has_more="0"
[debug]unparsable_output==>(gdb)

question:

Is the "unparsable_output" correct? See the above log.

[Pecan]

Code

[debug]ActionsMap::Run -> starting action: 0000000011aa5f50 id: 21
[debug]cmd==>210000000000-var-update 1 *
[debug]output==>210000000000^done,changelist=[{name="var5",value="std::vector of length 5, capacity 5",in_scope="true",type_changed="false",new_num_children="5",displayhint="array",dynamic="1",has_more="0"}]
[debug]cmd==>210000000001-var-list-children 2 "var5"
[debug]unparsable_output==>(gdb)
[debug]output==>210000000001^done,numchild="5",displayhint="array",children=[child={name="var5.[0]",exp="[0]",numchild="0",value="1",type="int"},child={name="var5.[1]",exp="[1]",numchild="0",value="2",type="int"},child={name="var5.[2]",exp="[2]",numchild="0",value="3",type="int"},child={name="var5.[3]",exp="[3]",numchild="0",value="4",type="int"},child={name="var5.[4]",exp="[4]",numchild="0",value="5",type="int"}],has_more="0"
[debug]unparsable_output==>(gdb)

question:

Is the "unparsable_output" correct? See the above log.

Yes. It's ok. It's left over from code inherited from the original CodeBlocks debugger plugin.
I seem to remember that I tried to eliminate it, but removing it just caused a whole bunch of problems.

The original debugger code set that line as the prompt to be sent back from from gdb.
On some commands to gdb, there is original plugin code that waits for multiple responses to be returned to CodeBlocks; debugger plugin goes idle before sending any more commands. The idle condition is keyed to that prompt to indicate the end of any more responses and clears the idle condition.

Since the original debugger code and prompt mechanism worked with debugger_gdbmi (and caused problems without it), I got lazy and just left it for now.

I'll put this in the "revisit latter" ToDo queue. I'd also like to eliminate it if I can.

[gd_on]

A few remarks on debugger_gdmi plugin.
I have created a debugger_gdmi folder Inside plugins\contrib and put inside a copy of your git repos.
I have added 3 lines in my CodeBlocks_Windows.workspace as it is for all other plugins.

Code

	<Project filename="plugins/contrib/debugger_gdmi/debugger_gdbmi_Windows.cbp">
		<Depends filename="CodeBlocks_Windows.cbp" />
	</Project>

Building C::B as usual, produces as previously my C::B version, but containing this time debugger_gdmi.zip and debugger_gdmi.dll at the correct place.

My first debug tries did not work as expected.
I find in your distribution a gdbinit file. It contains a path to print.py, but it does not match a path on my system. I tried to replace this path by the one I have (for example C:\wxWidgets-3.3.2\misc\gdb), but it does not seem to change something. Is this gdbinit file useful ?

Finally, I made it work by avoiding space and/or accentuated characters within the path to my executable. The standard debugger.dll (which uses the same gdb.exe) seems to be more tolerant with that.

Other points to notice :
- with standard debugger, when I open the watches window, all my local variables are displayed. With gdbmi, I have to add one by one an explicit Watch to my local variables. I can then follow the content of my variables step by step when executing.
- at the end of my program, if I try to restart it, while the watches window is still opened, I see my variables, but their content is not updated: 2 solutions here, open new watches on the same variables and I see that only the new ones are updated, or simply rename them but without really changing the name (just clic on return at the end of the variable name).

Last question : can we have both plugins (standard and gdmi) activated ? Or is it better to have only one activated (as for codecompletion and clangd_client) ?

[ollydbg]

Last question : can we have both plugins (standard and gdmi) activated ? Or is it better to have only one activated (as for codecompletion and clangd_client) ?

Yes, you can have both of the 2 debugger plugins enabled. There is a menu item: Debug->Active debuggers->You can select one.

- with standard debugger, when I open the watches window, all my local variables are displayed. With gdbmi, I have to add one by one an explicit Watch to my local variables. I can then follow the content of my variables step by step when executing.

In the standard debugger plugin, this is the option: "Watch local variables". In the gdbmi plugin, I also see this option is ON, but I see the same issue here, which means local variables were not listed in the watches panel.