Repository navigation
Tags
Tags
fix(hpx): gate force_tunnel setter behind ws-yawc `RequestOptions::force_tunnel_mut` is only called by the WebSocket backend, but it was compiled unconditionally. With `hpx/json` and the default feature set the `ws-yawc` module is absent, so the crate denied the method as dead code and the CodSpeed benchmark build failed. Also regenerate Cargo.lock: the v2.6.0 bump left the workspace members at 2.5.35 and `flate2`/`tokio-boring` behind their manifest requirements, so `cargo check --locked` in the release publish gate could not update it.
fix(hpx-dl): drop unused native deps from metalink fuzz target The metalink fuzz target only exercises parse_metalink, which is pure Rust (quick-xml) and never touches TLS or SQLite. Building it with hpx-dl's default features pulled in boring (OpenSSL) and libsqlite3-sys, whose statically-linked C globals trip AddressSanitizer on Linux and abort the fuzzer before any input is tested. Build the target with default-features = false so it links only what parse_metalink needs.
fix(hpx-dl): use paused tokio clock in speed limiter timing tests The Coverage & Doctests CI job failed because `test_unlimited_returns_immediately` (and its three sibling "returns immediately" assertions) measured wall-clock elapsed time against a 50ms upper bound. Under the coverage-instrumented `cargo llvm-cov nextest --workspace --all-features` run the runner is fully loaded, so the no-op `wait_for` path itself took >50ms of wall time and the test failed all three nextest retry attempts (https://github.com/longcipher/hpx/actions/runs/38684351735/job/116255702525). Switch these four tests to `start_paused = true` and measure with `tokio::time::Instant`. If `wait_for` ever regresses into sleeping, the paused clock auto-advances to the sleep deadline and `elapsed` reflects it, so the assertion still catches regressions while staying immune to wall-clock jitter on loaded runners.
PreviousNext