-
Notifications
You must be signed in to change notification settings - Fork 217
Architecture of the MIEngine
Let's dig into how the MIEngine is put together. To start, here is an architectural diagram for the various parts of the MIEngine:
The 'Visual Studio Core Debugger', 'Visual Studio Code' and 'GDB/LLDB' are the parts of the diagram that are external to this project.
The Visual Studio Core Debugger box represents the Visual Studio UI, and the rest of the core Visual Studio debugging logic. This will call into the MIDebugEngine through the AD7 interfaces (IDebugEngine2, IDebugProgram2, etc). The MIDebugEngine will also call back into AD7 interfaces to send events (ex: a breakpoint was hit) or in some cases to obtain the objects needed to fulfill the contracts of the AD7 interfaces (ex: to obtain an IDebugProcess2 from the SDM for the newly launched process).
The Visual Studio Code box represents the Visual Studio Code UI. This will interact with this project over the Debug Adapter Protocol (DAP). OpenDebugAD7 understands the DAP, and will covert between DAP and AD7 interfaces to drive the MIDebugEngine.
MIDebugEngine/AD7.Impl is the top layer of the MIEngine. This layer contains the code required to implement the AD7 interfaces.
MIDebugEngine/Engine.Impl is where most of the smarts of the MIEngine lives. It presents an object model to the AD7 layer to make it easy for the AD7 layer to do its job.
MICore contains the base functionality used by the Engine.Impl layer, though that definition is a bit fuzzy. There are essentially five things in MICore:
- The Debugger class which is the central processor of the text that we get back from GDB/LLDB
- The MICommandFactory abstract class and its derived classes. This is our preferred mechanism for dealing with differences between GDB, LLDB, and other MI based debuggers that we support.
- The result parser (ResultValue, etc) which handles parsing MI results
- The transport classes, which handle setting up the standard input/output connection with GDB
- The launch options code, which handles processing the launch option XML, and loading up custom launchers
The only enforced boundary between MICore and Engine.Impl is that the various launchers only depend on MICore. So the types of functionality used by the launchers must be defined there.
The Android Launch and iOS Launcher handle the details of getting everything prepped so that we can debug on those platforms.
Lastly, all of these dlls depend on DebugEngineHost. More information on how this work can be found here.