Ah so this is related to my comment in #78. This was my initial idea for fixing the issues, but the ideal scenario would be to set the code up so that replace is simply deletion followed immediately by insertion. This would remove any possibility for #78 to occur.

Ah so this is related to my comment in #78. This was my initial idea for fixing the issues, but the ideal scenario would be to set the code up so that replace is simply deletion followed immediately by insertion. This would remove any possibility for #78 to occur.

Yes, this actually harder than I thought.

Yes, this actually harder than I thought.

It's strange that sometimes problems look easy when you start solving them, then you start writing code and it suddenly hits you that the problem is actually really hard 😆. I spent a good while chatting with @shtanton over what the best way of fixing this was - because it's not impossible to patch up the existing code with even more complexity, but we both agreed that what would be ideal is if the replace code could just be a series of delete_child/insert_children (which would fix this problem, because the insertion would implicitly create a new field). But that's currently insanely hard to do (because Json::Field stores its children in a way that you can't delete from), so I'm currently refactoring the ast module to make this possible.