String gaps are collapsed even when semantically meaningful

[amesgen]

Describe the bug

The string literal

"\65\ \0"

is formatted to

"\650"

which are different strings for all GHC versions but 9.12.1 (on 9.12.1, they are the same, but that is a bug that is already fixed on GHC master: https://gitlab.haskell.org/ghc/ghc/-/issues/25784).

Concretely, the first string is equal to "A0" (two characters), while the second one is equal to "ʊ" (single character).

This has been broken ever since #376.

Expected behavior

The input should be formatted to a string literal representing the same string.

I see multiple options:

1. Never fully collapse string gaps, i.e. all gaps (when in single-line layout or in a multi-line literal) are replaced with \ \.

   The input above would be a fixed point in this case.

   I think this is the simplest option.

2. Replace string gaps (in single-line layout or in a multi-line literal) with the special escape character \&.

   I think this is potentially confusing as \& is rarely used (I had never heard of it before.)

3. Like 1./2., but only when necessary to be semantics-preserving.

   Not sure if this is worth the (small) additional complexity.

Environment

Ormolu 0.8.0.0 or current master (4369d7c)

[amesgen]

I opened #1161 implementing option 1.

---

cc @brandonchinn178 in case you have thoughts as you fixed https://gitlab.haskell.org/ghc/ghc/-/issues/25784

[brandonchinn178]

Oh! I didn't know ormolu/fourmolu automatically collapsed string gaps. Yeah, that's a bit surprising for a formatter to do, IMO. I think leaving the string gaps as is makes sense

EDIT: oh I didn't read carefully. IMO we should leave string gaps alone instead of minimizing them. But option 1 is fine too.