Qualified first type argument after `<` is parsed as `Lident` instead of `Ldot`

According to my hours-long talk to a LLM, I have just found a non-reported bug in Reason parser. I am NOT an expert in any of this, but I tried to double check as much as I could before reporting it. It looks reasonable to me.

The following is generated by an LLM and carefully edited by me.

Versions

• Reason 3.18.0
• OCaml 5.2.0
• Melange 7.0.1-52

Minimal reproduction

create files

Create qualified_type.ml:

type t = string

Create with_space.re:

type values = array< Qualified_type.t>;

Create without_space.re:

type values = array<Qualified_type.t>;

compile

NOTE: set switch with eval "$(opam env --root="$HOME/.opam" --switch=hippercampus_melange --set-switch)"

Compile the defining module and parse both Reason files:

melc -c qualified_type.ml -o _output/qualified_type.cmj
refmt --parse re --print binary with_space.re > _output/with_space.pp.ml
refmt --parse re --print binary without_space.re > _output/without_space.pp.ml

Type-check both parser outputs:

melc -I _output -i _output/with_space.pp.ml
melc -I _output -i _output/without_space.pp.ml

check the types created

From the studies directory, compile the AST inspector:

ocamlfind ocamlc -package compiler-libs.common -c  inspect_longident.ml -o _output/inspect_longident.cmo
ocamlfind ocamlc -package compiler-libs.common -linkpkg   _output/inspect_longident.cmo -o _output/inspect_longident.exe

compare the types:

_output/inspect_longident.exe \
    _output/with_space.pp.ml \
    _output/with_space.inspected.pp.ml

Lident("array")
Ldot(Lident("Qualified_type"), "t")

_output/inspect_longident.exe \
    _output/without_space.pp.ml \
    _output/without_space.inspected.pp.ml

Lident("array")
Lident("Qualified_type.t")

FAIL: Ldot(Lident("Qualified_type"), "t") vs Lident("Qualified_type.t")

Actual behavior

The spaced form succeeds. The unspaced form fails:

Error: Unbound type constructor Qualified_type.t

The spaced form represents the qualified type correctly:

Ldot(Lident("Qualified_type"), "t")

The unspaced form represents the same qualified type as one identifier containing a dot:

Lident("Qualified_type.t")

Expected behavior

Both forms should produce:

Ldot(Lident("Qualified_type"), "t")

Alternatively, if the unspaced form is not supported, the parser should reject
it as invalid syntax instead of creating a malformed qualified identifier.

Diagnosis

The lexer consumes <Qualified_type.t as one LESSIDENT token. The
first_less_than_type_ident grammar rule then constructs it unconditionally
as:

Ppxlib.Longident.Lident $1

This is valid for an unqualified name but incorrect when $1 contains a
module-qualified type.

The same rule is present in the current Reason master branch:

• Lexer
• Parser grammar

Workarounds

The spaced form works:

type values = array< Qualified_type.t>;

The canonical parenthesized Reason form also works:

type values = array(Qualified_type.t);

Start with src/reason-parser/reason_declarative_lexer.mll and src/reason-parser/reason_parser.mly, focusing on the LESSIDENT token and first_less_than_type_ident rule. Reproduce the spaced and unspaced examples with refmt and inspect their outputs. Done means the unspaced qualified type is represented as Ldot like the spaced form, or is rejected as invalid syntax.