fix(runtime/explicit): skip deleted constructors - #450

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands.}

Addressed in 14f3d58.

The constructor matcher now checks the declaration suffix after the closing ) for = delete;, allowing an optional noexcept / noexcept(...) suffix and continuing across following cleansed lines until the declaration terminates. This covers both Foo(int) noexcept = delete; and a newline before = delete; without changing the existing argument analysis.

Added regressions for both forms.

Validation:

python -m pytest cpplint_unittest.py -q --no-cov
206 passed

python -m pre_commit run --files cpplint.py cpplint_unittest.py
Passed

git diff --check
Passed

Addressed the nested-expression case in 962e80f.

The suffix reader now tracks parentheses, brackets, and braces, and only treats ; or { as a terminator at declaration depth. A { inside noexcept(...) no longer ends the scan, so Foo(int) noexcept(noexcept(T{})) = delete; is recognized as deleted.

Added that regression alongside the existing one-line and multiline cases.

Validation:

python -m pytest cpplint_unittest.py -q --no-cov
206 passed

python -m pre_commit run --files cpplint.py cpplint_unittest.py
Passed

git diff --check
Passed

_GetConstructorSuffix() is the right shape, and I confirmed #386's case plus the noexcept and split-line forms are clean on 962e80f (the leading requires-clause form works too).

One gap: the is_deleted regex (cpplint.py, ~line 3982) enumerates allowed suffixes, so these valid declarations still emit runtime/explicit:

template <class T>
Foo(T v) requires Integral<T> = delete;   // C++20, trailing requires-clause
Foo(int f) throw() = delete;              // valid through C++17
Foo(int f) = delete("use Bar instead");   // C++26, P2573

Each compiles under g++ -pedantic-errors -fsyntax-only at the relevant standard.

Since _GetConstructorSuffix() already stops at the top-level ; or {, the suffix either ends in = delete; or it isn't deleted, so checking the end covers all three without enumerating forms:

is_deleted = bool(
    re.search(r"=\s*delete\s*(?:\(.*\))?\s*;\s*$", constructor_suffix, re.DOTALL)
)

Tested locally on your branch: 206 tests pass, and no over-suppression — inline body, member-init body, declaration followed by a deleted overload, two constructors on one line, macro suffix, and a trailing // = delete comment each still warn exactly once. Worth adding a regression for the trailing requires-clause.

Addressed in bea44fe.

is_deleted now searches for a deleted marker at the end of the suffix collected by _GetConstructorSuffix(), rather than enumerating permitted suffix forms. This covers a trailing requires clause, throw(), and the diagnostic form of = delete(...) while preserving the existing declaration-level terminator handling.

Added regressions for all three forms.

Validation:

.venv\Scripts\python.exe -m pytest cpplint_unittest.py -q --no-cov
206 passed

.venv\Scripts\python.exe -m pre_commit run --files cpplint.py cpplint_unittest.py
Passed

git diff --check
Passed

Rebased onto the current develop branch and resolved the changelog conflict by retaining both unrelated entries.

Validation after the rebase:

python -m pytest --no-cov cpplint_unittest.py -k testExplicitSingleArgumentConstructors -q
1 passed

python -m pytest --no-cov -q
231 passed

git diff --check upstream/develop...HEAD
Passed

Addressed in b833ee8. The suffix scanner now distinguishes the initial requires-clause from subsequent requires-expressions, keeps scanning through their requirement bodies, and stops at the actual constructor body.

The regressions cover the reported form, optional requires-expression parameters, multiline requirements, compound requirements, and conjunctions. Matching non-deleted declarations still warn, and body-boundary checks prevent scanning into a later deleted constructor. The C++20 declarations pass clang++ -std=c++20 -pedantic-errors -fsyntax-only.

The full pytest suite passes (233 tests, 96.40% coverage), as do Pylint and the configured pre-commit hooks for the changed files.