fix: handle more cases of editing subtrees that depend on column values

[rooney]

Reopen #3218

[maxbrunsfeld]

Thanks for the fix @rooney. I left some minor suggestions.

[maxbrunsfeld]

This new condition makes good sense. Did you check if you can actually remove the previous condition? I would think that you'd only need to check depends_on_column in one spot. Do any tests fail if you remove the second part of this 3-part && expression?

[rooney]

Did you check if you can actually remove the previous condition?

Yea I did, and I can't: it would make the Haskell-like indentation test fail -- I originally thought that it's the test that were wrong.

I would think that you'd only need to check depends_on_column in one spot.

That would be the case if trees that depend-on-column make their supertree depend-on-column too. But that's not the case because currently the condition for setting the flag is:

// subtree.c line 394
    if (
      self.ptr->size.extent.row == 0 &&
      ts_subtree_depends_on_column(child)
    ) {
      self.ptr->depends_on_column = true;
    }

So it's possible for a tree to NOT depend on column, but it contains subtree that DOES. (I find that counterintuitive, but maybe there's a reason for that?).
That's why we need to check depends_on_column on two spots: one when iterating the bisecting-stack, and another when delving into the children -- the condition needs to be re-checked for each subtree.

Now, I did try changing the flag-setting code to:

// subtree.c line 394
    if (
      ts_subtree_depends_on_column(child)
    ) {
      self.ptr->depends_on_column = true;
    }

• All the tests still pass after this change.
• It's more intuitive: if a tree depends-on-column, then its supertree depends-on-column too.
• On subtree-edit, we just need to check the flag at one spot only.

But is it OK? If so, I'll make the change.

Do any tests fail if you remove the second part of this 3-part && expression?

No, all tests still pass without the second condition -- it's an optimization for when the column positions does not change e.g. if the length of inserted_text == deleted_length. Do you think it's incorrect/premature? Or is it ok but I should add a test for that (in the form of a recorder reading fewer bytes)?

[maxbrunsfeld]

No, then almost the entire tree would depend on its column.

A tree only depends on its start column of any of its children that start on its first line depend on their column. For all subsequent children, the tree's start column is decoupled from their column.