Fix roundtrip check to support 5.1 <-> 5.2 migrations

[NathanReb]

Printing to source code and parsing with the compiler before running the upward migration is expected to fail because
fun x y -> z and fun x -> fun y -> z parse to the same AST on 5.01- but have different representations on OCaml 5.02+.

We therefore need to first migrate it to the compiler version before printing it so that it is properly annotated with attributes by the downward migration, allowing the upward one to preserve the original AST.

[NathanReb]

CC @patricoferris !

[patricoferris]

Definitely seems to fix the diffing in the 5.2 AST branch (https://github.com/ocaml-ppx/ppxlib/pull/514/checks?check_run_id=29662622891 -- the errors now are because of something I need to potentially promote it seems). The Astlib.Compiler_pprintast.structure_item is unbound on some platforms it seems/compiler versions

[NathanReb]

Yeah I was kind of expecting Pprintast.structure_item not to exist on older compilers, I'll fix that!

[NathanReb]

the errors now are because of something I need to potentially promote it seems

I got those errors when testing on top of the 5.2 bump as well. My guess is that the deriving_inline bits in ast.ml were not updated because of the roundtrip error but I might be misunderstanding what's going on here.

[NathanReb]

What puzzles me is that there are two sides to the deriving_inline diff in ast.ml, there are fun x y -> z being turned to fun x -> fun y -> z which I get. I might again be wrong but I'd attribute those to ppx_traverse generating nested Pexp_function in 5.2 instead of generating flat Pexp_function with maximum arity, which in turn gets printed differently.

The other side are those weird formatting changes, such as removed spaces before = which I'm not quite sure where they are coming from.

[patricoferris]

The other side are those weird formatting changes, such as removed spaces before = which I'm not quite sure where they are coming from.

I thought this too, but this is actually always been the case! I tested on older versions of the codebase -- the = and extra bracketing is removed by a subsequent dune build @fmt --auto and any dune build @lint won't be retriggered afterwards :))

[patricoferris]

Oh and the fun x -> fun y -> is converted back to fun x y -> by OCamlformat!

[patricoferris]

Ah that's the problem, sorry for all the messages, that's what is broken here. The traversal and ocamlformat are fighting each other, so a lint followed by a format undoes the work of the lint.

[NathanReb]

We do still have a roundtripping issue though don't we? Can we go ahead and merge this PR?

[patricoferris]

I think we should leave some comment about the logic in this section, but otherwise this is good to go :))

[NathanReb]

The trunk build is failing for unrelated reason, I'll try and fix it in a separate PR. Merging!