function editor

[cben]

The UX I had in mind is extremely similar to this thing I just found:
https://archive.org/details/akkartik-2min-2020-12-06

• language: support function definitions (no definition syntax, just given as name->string dictionary)
• support multiple editors, one per function
• when cursor stands on a function, embed an editor of the refenced function underneath it
• allow live editing of the embedded editor right there — this supports "example-driven" development, where to get useful feedback on function execution, you need to set up some params before calling it
• language: return a full execution trace
• when cursor travels inside an embedded editor, show stack at that point inside
• keybindings and/or buttons to single-step execution including nested functions

[cben]

language: return a full execution trace

Do have evalSmallStep() semantics helper now, but results are presently stored in CodeMirror flat mode state, associated with source tokens.
That's not gonna suffice when same function can be called from several places, and the editors descending into the function code are not always materialized?

• So one obvious approach is a standalone trace data structure.

• Evaluation is deterministic, so might be OK to keep state in CodeMirror, with "medium step" semantics where function calls are atomic (like "n" in debuggers), and sub-editors materialize sub-traces (re-evaluating from that point) when opened?
  This feels less "modular" — editors know "mechanics" of how calls work — but the whole UI of opening nested editors for calls has to know that.

  • Open Q: is mode state per-editor or per-doc? Multiple editors for same function should stay in sync, so I assume they need to a share a doc object? But see https://codemirror.net/5/doc/manual.html#linkedDoc ?

[cben]

I'm likely to switch CM5 -> CM6 so entangling too much with CM modes might be unwise...