Enforcement of absence of field does not work at the toplevel of the dune selector

[mbarbin]

Bug report from Claude Opus 4.6 below:

Bug: (dune (not (has_field lint))) fails to auto-fix, while (dune (library (not (has_field lint)))) works

Description

When using (not (has_field <field>)) as a shorthand at the dune predicate level, dunolint reports an enforce failure with no automatic fix available, even though the documentation states that negation of has_field should suggest removing the field.

Steps to reproduce

Given a library stanza with a lint field:

  (library
   (name catch_the_bunny_stdlib)
   (public_name catch-the-bunny.stdlib)
   (flags :standard -w +a-4-40-41-42-44-45-48-66 -warn-error +a)
   (libraries dyn ordering pp)
   (instrumentation
    (backend ppx_windtrap))
   (lint (pps))
   (preprocess no_preprocessing))

Using the rule (dune (not (has_field lint))) produces:

Error: Enforce Failure.
The following condition does not hold: (not (has_field lint))
Dunolint is able to suggest automatic modifications to satisfy linting rules
when a strategy is implemented, however in this case there is none available.
Hint: You need to attend and fix manually.

Workaround

Wrapping the condition in library (...) works correctly:

(dune (library (not (has_field lint))))

This correctly suggests removing the lint field.

Root cause and implementation notes

In library.ml (and executable.ml), the outer Linter.enforce function handles all Not cases with a blanket Not _ -> Eval fallthrough, which bypasses the inner Top.enforce that correctly handles Not (has_field _) by removing the field.

The T (positive) branch correctly delegates shorthand predicates like T (has_field _)toTop.enforce, but the corresponding Notbranch does not. This affects allhas_field variants (lint, instrumentation, preprocess, name, public_name) when used as shorthands at the dune` predicate level.

[mbarbin]

This report was generated by Claude Opus 4.6 in a session that also led to #179 .