qt: make the WebSocket backend and server clang-tidy clean (#514)

[Yaraslaut]

Closes #514.

Scope

The issue names three spots. I fixed those and the twenty others in the same two files that fail the same way — the stated concern is "a consumer that also runs clang-tidy with this check as an error gets a build failure on unmodified vendored code", and 3 of 23 doesn't deliver that. The Qt WebSocket backend and server, headers included, now produce zero clang-tidy findings.

The three from the issue

1. bugprone-empty-catch. The handler now assigns the "disconnected" fallback itself rather than relying on the initializer above it:

std::string message;
try { std::rethrow_exception(exc); }
catch (const std::exception& concrete) { message = concrete.what(); }
catch (...) { message = "disconnected"; }

Every caller passes a make_exception_ptr, so rethrow_exception always throws and one handler always assigns. Behaviour identical, and the handler now does the work its comment describes instead of being lexically empty — which the check rejects however well commented.

2. Narrowing in the backoff. Cast up to double explicitly so the multiplication is openly floating-point, leaving the one deliberate narrowing on the outside where the existing cast documents it.

Provably the same arithmetic — rep * double already promotes the integral operand, so this only names the promotion. Verified rather than asserted:

static_assert(std::is_same_v<decltype(rep{7} * 1.7), decltype(static_cast<double>(rep{7}) * 1.7)>);
constexpr bool same(rep n, double m) { return static_cast<rep>(n * m) == static_cast<rep>(static_cast<double>(n) * m); }
static_assert(same(1, 1.5) && same(250, 1.7) && same(999983, 2.3) && same(0, 3.0) && same(7, 0.5));

3. cppcoreguidelines-owning-memory on the QObject-parented new. I did not take the project-wide suppression the issue suggested, and want to flag the disagreement rather than bury it.

I measured first: new QTimer(this) is the only such site in all of src/. (There are three more repo-wide — examples/common/testkit, examples/common/wasm_spike, examples/kanban/gui_lib — but they're in three different directories, so no single directory config would have covered them either, and examples aren't what a consumer vendors.) Disabling the check for src/qt/ would turn off a genuinely useful check across the directory to silence one line, and would silently accept a future new that really is unowned. So: a per-site NOLINT carrying the Qt-ownership rationale. If QObject-parented new becomes common in src/, tests/.clang-tidy is the precedent for doing it directory-wide properly, and I'd switch then.

The other twenty

Mechanical, no behaviour change: designated initializers (4), const correctness (5), explicit != nullptr instead of implicit pointer-to-bool (7), consumeToken made const, auto on a cast initializer, structured-binding names above the identifier-length floor (3), a redundant member init pair, and deleted copy/move on two classes that were never copyable in practice (both hold Qt signal/slot connections bound to their own address).

One worth a second look was readability-math-missing-parentheses on state.tokens + elapsedSeconds * capacity — that is correct token-bucket refill math, and the check only wants the precedence stated. Parenthesised, not changed.

Verification

• clang-tidy: 0 findings across src/qt/qt_websocket_backend.cpp, src/qt/qt_websocket_server.cpp, include/morph/qt/qt_websocket_backend.hpp, include/morph/qt/qt_websocket_server.hpp (was 23 + 5 in the headers)
• clang and gcc, both with MORPH_BUILD_QT=ON: clean
• 1689/1689 ctest tests pass, including all 67 Qt tests and both WebSocket interop cases
• tree-wide clang-format --dry-run -Werror, spec-citation lint: clean

🤖 Generated with Claude Code

https://claude.ai/code/session_01SxkvtHvan2fWKDDaBpqkgv

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!