scx (1.1.4-1) unstable; urgency=medium

  * bump version to 1.1.4

 -- LFRon <ronforever@outlook.com>  Sun, 18 Oct 2026 11:01:00 +0800

scx (1.1.3-2) unstable; urgency=medium

  * Rebuild against the updated Better-Deepin scx source (fork of
    sched-ext/scx v1.1.3), which now carries two scx_lavd commits so the
    scheduler loads and runs on sched_ext backports that predate the
    upstream v6.17 cgroup bandwidth ABI (e.g. the Anolis-aligned deepin
    6.6 backport), without changing behavior on kernels that have the
    ABI:
    - Load on kernels without the cgroup bandwidth ABI: bail out of the
      bandwidth init path before touching struct scx_cgroup_init_args
      bw_* fields and force-disable --enable-cpu-bw when the ABI is
      missing, instead of failing the whole BPF object load;
    - Attach only the futex tracepoints that exist on the running kernel,
      so futex boost works on x86_64, which lacks the dedicated
      futex_wait/futex_wake syscall tracepoints that made the previous
      all-or-nothing helper bail out.
  * No Debian patches are carried for these changes; kernel sources
    untouched.

 -- LFRon <ronforever@outlook.com>  Sat, 11 Oct 2026 20:05:00 +0800

scx (1.1.3-1) unstable; urgency=medium

  * Initial Debian packaging of the sched-ext scx scheduler collection.
  * Built as a Cargo workspace using the "manual wrapper approach"
    documented in the Debian Rust Team book
    (see https://rust-team.pages.debian.net/book/process-workspace.html):
    the /usr/share/cargo/bin/cargo wrapper, an architecture.mk-driven
    target triple and dh-cargo-built-using for Built-Using.
  * Ships the sched_ext schedulers (scx_bpfland, scx_rusty, scx_lavd,
    scx_layered, scx_p2dq, scx_rustland, ...) plus the systemd
    on-load service (scx.service) and its /etc/default/scx conffile in
    scx-scheds, and the scxtop observability TUI in its own package.
  * The BPF/C portion (skeletons, vmlinux.h bindings) is compiled by
    clang in the per-crate build.rs scripts (scx_cargo / libbpf-cargo),
    so no separate upstream C build system is used.
  * d/rules tries the offline Debian system crate registry first and
    only if it cannot satisfy every dependency of Cargo.lock it falls
    back to downloading the missing crates directly from crates.io
    over the network (a deliberate downstream convenience switch for
    distros that do not package every Rust dependency; then
    Built-Using / Static-Built-Using are not recorded at all).
  * xtask, vmlinux_docify, scx_forge_agent and scx_mitosis are excluded
    from the build (development helpers / optional agent tooling whose
    transitive network stack pulls cmake, openssl and aws-lc; see
    debian/README.source on how to re-enable them). Tests are skipped
    because the upstream test-suites require a running sched_ext
    kernel; a light autopkgtest smoke check is shipped instead.

 -- LFRon <ronforever@outlook.com>  Thu, 17 Sep 2026 18:39:00 +0800
