[lldb] [llvm] Annotate disassembly with register‐resident variable locations (PR #147460)

[Adrian Prantl via llvm-commits]

adrian-prantl wrote:

> > adding an option to the dissemble CommandObject is the right solution. Personally I think I would prefer an option to the CommandObject. Do you think this is feasible, or are there more many points through which the disassembly is reachable, making an option infeasible?
> 
> After looking more closely into one of the failing test cases, TestFrameDisassemble.py, I noticed that its entry point for disassembly is through `SBFrame::Disassemble()`, which bypasses the CLI entirely and doesn’t go through `CommandObjectDisassemble`.

There are multiple options:
1. keep the original behavior and call the internal API from SBFrame::Disassemble with the flag to turn off the annotations
2. (optional) add an SBFrame::Disassemble(class DisassembleOptions) overload

> Since this means command-line options like `--rich` wouldn’t apply in this context, would it be okay if I start by introducing a new setting (e.g., `target.enable-rich-disassembly`) to gate the annotation output globally?

I think I would prefer implementing option (1) and worry about exposing the new behavior through the API later. I would like to avoid global or target-specific settings if possible, because they can introduce unwanted side effects. (Imagine an IDE that depends on the traditional disassemble format, and a user that turns on the setting in their lldbinit file).

> Once that’s in place and CI is passing, I can follow up with a separate change to add the `--rich` option to `CommandObjectDisassemble`, which would override the setting for CLI use cases and give users more fine-grained control.
> 
> Does that sound reasonable?



https://github.com/llvm/llvm-project/pull/147460