Inconsistencies with Loc in coqc and vscoq

[gares]

Here the problem: it seems that the ranges written in the terminal and the ranges sent via lsp are 1-based v.s. 0-based.

---

In this case I clicked on the error location as printed in the terminal, and code selects the exact text:

---

Here I peek the lsp traffic for the same error. The line number was decreased by 1, so I guess it is zero based.
The column numbers are the same, so (assuming they are also 0 based)< are coloured 1 char to the right.

[gares]

I don't know exactly why, but it seems the loc I generate are OK for coqc, but get somehow shifted by vscoq.

[SkySkimmer]

Is this only with elpi stuff or can you reproduce with something simple like Check doesntexist.?

[gares]

I think it is only elpi, but also only with code, since with coqc locs are ok

[rtetley]

So we compute the lsp range using the Loc.t from Coq via:
https://github.com/coq/vscoq/blob/96b677a32d43dfd9ce7aa2a9f1ac7480477b0f2e/language-server/dm/rawDocument.ml#L83C1-L87C4

Looking at the code we seem to use loc.bp and loc.ep from the type:

type t = {
  fname : source; (** filename or toplevel input *)
  line_nb : int; (** start line number *)
  bol_pos : int; (** position of the beginning of start line *)
  line_nb_last : int; (** end line number *)
  bol_pos_last : int; (** position of the beginning of end line *)
  bp : int; (** start position *)
  ep : int; (** end position *)
}

Other than that it does indeed seem to be 0 based. Maybe there is a difference in how you instantiate bp and ep but not other members of that record which lead to the inconsistencies ?