Struct CtprofMetricDef 

#[non_exhaustive]
pub struct CtprofMetricDef {
    pub name: &'static str,
    pub rule: AggRule,
    pub sched_class: Option<&'static str>,
    pub config_gates: &'static [&'static str],
    pub is_dead: bool,
    pub description: &'static str,
    pub section: Section,
}

One metric exposed by the comparison pipeline.

The auto-scale ladder for the rendered cell is derived from AggRule::ladder at render time — there is no separate unit tag on the metric def. A registry entry that pairs an AggRule variant with a category-mismatched ladder fails at compile time (the ladder mapping is a closed match on the variant, not a free-form string).

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

name: &'static str

rule: AggRule

sched_class: Option<&'static str>

Scheduler-class scope for the metric. None means class-agnostic — every task class accumulates the value (e.g. nr_migrations). Concrete spellings:

• "cfs-only" — incremented strictly inside CFS-class call paths (kernel/sched/fair.c), zero under SCHED_EXT / SCHED_FIFO / SCHED_RR / SCHED_DEADLINE / SCHED_IDLE. Examples: nr_wakeups_affine, nr_wakeups_affine_attempts, nr_failed_migrations_*, nr_forced_migrations, slice_max.
• "fair-policy" — emitted only when fair_policy(p->policy) returns true. Per kernel/sched/sched.h:194,203, that admits SCHED_NORMAL, SCHED_BATCH, AND SCHED_EXT (under CONFIG_SCHED_CLASS_EXT). Zero under SCHED_FIFO/RR/DL/IDLE. Example: fair_slice_ns.
• "non-ext" — written by the schedstats sleep/wait family wrappers __update_stats_enqueue_sleeper (kernel/sched/stats.c:48) and __update_stats_wait_end (kernel/sched/stats.c:21), called from fair.c, rt.c, deadline.c but NOT ext.c — i.e. CFS/RT/DL accumulate, sched_ext bypasses. Examples: wait_sum, wait_count, wait_max, voluntary_sleep_ns, sleep_max, block_sum, block_max, iowait_sum, iowait_count.

config_gates: &'static [&'static str]

Kernel CONFIG options that gate the metric. &[] means no gating (always populated when the source path runs). One element typically; multi-element when more than one gate is required (e.g. core_forceidle_sum requires CONFIG_SCHED_CORE AND CONFIG_SCHEDSTATS). Concrete spellings match the literal Kconfig symbol so an operator can grep CONFIG_X /boot/config-$(uname -r) to confirm. Verified gates:

• "CONFIG_SCHEDSTATS" — gates every __schedstat_* / schedstat_* macro call. Off → the macro is do { } while (0) per kernel/sched/stats.h:75-82.
• "CONFIG_SCHED_INFO" — gates the lighter-weight sched_info_* accounting (run_time_ns, wait_time_ns, timeslices); the schedstat file is gated by sched_info_on() at proc_pid_schedstat (fs/proc/base.c:511-523).
• "CONFIG_SCHED_CORE" — gates the core-scheduling subsystem (__account_forceidle_time).
• "CONFIG_SCHED_CLASS_EXT" — gates the sched_ext class. When off, no task can land on ext, so ext_enabled reads false uniformly.
• "CONFIG_TASK_DELAY_ACCT" — gates the delayacct accounting path that populates the taskstats genetlink delay-family fields (cpu_delay_*, blkio_delay_*, etc.).
• "CONFIG_TASK_IO_ACCOUNTING" — gates the per-task I/O accounting fields exposed by /proc/<tid>/io (rchar, wchar, syscr, syscw, read_bytes, write_bytes, cancelled_write_bytes). The kernel emits all 7 fields under one do_io_accounting call, and CONFIG_TASK_IO_ACCOUNTING depends on CONFIG_TASK_XACCT in init/Kconfig — so from the procfs-reader perspective the file is all-or-nothing.

is_dead: bool

True for kernel counters that are exposed in /proc but never incremented anywhere in the kernel tree — always reads zero. Operators reading the rendered table see the [dead] flag and stop chasing the always-zero cell. The registry is currently empty of is_dead: true entries: the previously-registered dead counters (nr_wakeups_idle, nr_wakeups_passive, nr_migrations_cold) were dropped from ThreadState and the registry; the kernel still emits the lines so the parser silently ignores them. The flag remains as infrastructure: a future kernel that resurrects a dead counter (or exposes a new always-zero one) registers with is_dead: true and the [dead] rendering path fires.

description: &'static str

One-line operator-facing description of what this metric counts. Surfaced by the ctprof metric-list subcommand alongside the bracketed tag suffix so an operator scanning a rendered table can map an unfamiliar metric name to its semantics without leaving the CLI. Plain ASCII. “Cumulative” is load-bearing — use it to distinguish counters from gauges; the AggRule only names the per-group reduction, not the per-thread counter shape.

section: Section

Section this metric belongs to for the --sections per-row filter. Most rows tag Section::Primary; taskstats-sourced rows (the eight delay-accounting categories plus the two memory watermarks) carry Section::TaskstatsDelay so an operator can scope the rendered table down to (or away from) the taskstats rows. The primary-table emitter checks DisplayOptions::is_section_enabled per row before rendering — --sections taskstats-delay keeps only taskstats rows, --sections primary excludes them, and either alone keeps the primary table open. The default (empty filter) renders every row regardless of section.

Trait Implementations

impl Clone for CtprofMetricDef

fn clone(&self) -> CtprofMetricDef

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for CtprofMetricDef

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Copy for CtprofMetricDef