perf(markdown-formatter): unwired producer builds telemetry payload before the opt-in guard (~90ms/edit on Windows)

[kyle-sexton]

Summary

When HOOK_TELEMETRY_SINK is unset (telemetry not wired — the default for consumers who have not opted in), the markdown-format producer still performs telemetry-only work before the opt-in guard short-circuits it. Low priority — bounded and dwarfed by markdownlint-cli2, but genuinely unnecessary work when telemetry is off.

Detail

hook::emit_telemetry (plugins/markdown-formatter/hooks/hook-utils.sh) guards on the first line:

[[ -n "${HOOK_TELEMETRY_SINK:-}" ]] || return 0

That guard itself is free. But markdown-format.sh builds the telemetry payload before calling it, unconditionally:

• the FILE_REL block — 2× cygpath calls on Windows;
• data_json — 1× jq -n.

So ~3 telemetry-only subprocesses run per .md edit even when no sink is configured: ≈90ms on Windows (MSYS2 fork() ≈30ms each), ≈9ms on Linux. FILE_REL and data_json are used only for the envelope.

Suggested fix

Gate the FILE_REL + data_json construction behind a cheap sink-presence check (e.g. a hook::telemetry_enabled helper that just tests [[ -n "${HOOK_TELEMETRY_SINK:-}" ]]), so the unwired path does zero telemetry-only subprocesses. REPO_ROOT stays (it is also used for markdownlint config discovery).

Severity

Low — telemetry is opt-in; the cost is small and dominated by the lint itself. Capture, not block.