WIP - fixing reconstruction

[StachuDotNet]

trafficstars

given the code

("hello",12345)

if I select between the two |s:

("he|llo",12|345)

Upon hitting 'delete' I expect the remaining state to be "he|"

The tuple will be killed as the deletions happen, so I expect the 12345 to disappear. Then, I should have the "hel" remaining, and be on the right side of the e.

Instead, I get "hel|" - one of the "l"s remains.

---

This seems to be because how deleteCaretRange works:

https://github.com/darklang/dark/blob/d6d113b9dc99bf8c2401fdd5ce01f4bf40ec2e72/client/src/fluid/Fluid.res#L4930-L4953

• put the cursor at the RHS of the selection (in this case, 10)
• record the 'col' of the LHS of the selection (in this case, 4)
• hit 'backspace' until either:
  • the last backspace changed neither AST nor cursor position
  • the cursor is now at the recorded LHS

because the ( of the tuple is deleted, however, the initial LHS evaluation is sort of invalidated. The code thinks to go to position 4, but position 4 is eventualy between the 2 "l"s, rater than between "e" and "l".

---

My first attempt here was to rewrite the algorithm pretty dramatically. It works but breaks many other tests - will review and try again tomorrow.

[StachuDotNet]

Note: I needed undo to work locally, so branched this off of my branch that fixes such.

[pbiggar]

Haven't looked at the algorithm, but the way I would solve this is to use ASTRefs instead of positions

[StachuDotNet]

this has been superseded by #4531