Badly formed locations for Pexp_fun after 5.2 leads to Merlin confusion

[jchavarri]

Hi! Thanks for your work on ppxlib ❤️

We have stumbled upon this problem. Consider this file input.ml:

let make ~foo ~bar = foo ^ bar

Before 5.2, the locations for this file were the following (I cleaned up the ocaml.ppx.context part as it's not relevant for the issue and formatted them):

locations_5_1.txt

In 5.2, the locations are as follow:

locations_5_2.txt

Here's the diff between both files for clarity:

174,175c174,175
<               ((pos_fname input.ml) (pos_lnum 1) (pos_bol 0) (pos_cnum 30)))
<              (loc_ghost true)))
---
>               ((pos_fname input.ml) (pos_lnum 1) (pos_bol 0) (pos_cnum 18)))
>              (loc_ghost false)))

This location corresponds to the Pexp_fun ast node that corresponds to the ~bar argument. The change of location is problematic, because this AST node locations ends up being 14-18, while it has children that exceed that range, like a Pexp_apply which is immediate child with locations 21-30. Badly formed locations like this leads to issues with Merlin, in our experience.

Is this something possible due to the 5.1 -> 5.2 migration? is there anything we could do on our side to work around this problem? See related issue: reasonml/reason-react#840