[clr-interp] Implement interpreter debugging support in CoreCLR

[kotlarmilos]

Description

This issue outlines work for debugging support on the CoreCLR interpreter.

Entry points

Tasks:

• Add a new optimization tier OptimizationTierInterpreted
• Report this tier through DAC
• After interpreting a method, call JITComplete to register it with the debugger using the bytecode address instead of native code

Interpreter

Tasks:

• Implement the INTOP_BREAKPOINT opcode in the interpreter
• Call the debugger interface and raise a STATUS_BREAKPOINT exception with bytecode IP, frame, stack parameters
• In the SEH filter, detect STATUS_BREAKPOINT and call FirstChanceNativeException so the debugger is notified
• Resume normal execution after the debugger handles the breakpoint
• Implement InterpreterCodeManager::GetFunctionSize() to return the byte length of the interpreted code

Patches

Tasks:

• Add a new patch kind: PATCH_KIND_NATIVE_INTERPRETER for interpreter breakpoints
• In ApplyPatch, detect interpreter addresses and and insert an INTOP_BREAKPOINT instruction into the bytecode
• In UnapplyPatch, restore the original bytecode

IL mapping

Tasks:

• Update AddBindAndActivateILReplicaPatch to bind breakpoints directly to IL offsets for interpreted methods
• In MatchPatch and ScanForTriggers, skip comparing bytecode addresses to native IPs
• In DispatchPatchOrSingleStep, handle interpreter breakpoints separately to avoid false IP-change detection
• Initialize CodeRegionInfo for interpreted methods using the IL bytecode range (no hot/cold split)
• Skip native bound initialization in DebuggerJitInfo for interpreted methods

DoD: Hit breakpoint in a sample desktop app via Debugger.Break() and in VSCode. Run debugger tests in the CI.

[dotnet-policy-service]

Tagging subscribers to this area: @BrzVlad, @janvorli, @kg
See info in area-owners.md if you want to be subscribed.

[jkotas]

Implement InterpreterCodeManager::GetFunctionSize() to return the IL size

This does not look right. Why should InterpreterCodeManager be in the business of returning the IL size?

[kotlarmilos]

Yes, that is probably incorrect. Is byte length of the interpreted code stream (not IL) more accurate?

[jkotas]

Yes, bytecode is the common term we use to describe the output of the interpreter compiler.