init: data-driven boot service list via /etc/init.csv
Replace init's hardcoded boot_services array with an authoritative, human-readable service list read at boot, mirroring /etc/devices.csv. Each row is a service binary path followed by its argv; startup order is file order, shutdown the reverse. There is no hardcoded fallback — a missing file starts nothing (the no-ramdisk isolation behavior). - library/csv: shared CSV helpers (comment strip, field iteration) with unit tests; device-registry is refactored onto them so both /etc/*.csv files parse through one place. - init reads /etc/init.csv into fixed-max static tables (the same pattern as the device registry) and passes each row's argv straight to spawnSupervised. This also makes boot-time modes (e.g. device-manager test-usb-restart) expressible as data rather than hardcoded. - Diagnose mode (-Ddiagnose omits the display stack) becomes build-time file selection between etc/init.csv and etc/init-diagnose.csv, so init carries no comptime service logic; the diagnose build option is dropped from init. - build.zig: csv module wired; /etc/init.csv bundled into the initrd; the device-registry tests move to a dedicated block since they now import csv. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KJqSiLLchDUUCoXn5jsiwd
This commit is contained in:
@@ -0,0 +1,12 @@
|
||||
# /etc/init.csv — diagnose variant (-Ddiagnose), bundled at /etc/init.csv.
|
||||
#
|
||||
# The display stack (display, display-demo) is omitted so the kernel's timestamped
|
||||
# on-screen boot transcript is never suppressed — the bring-up timeline (USB,
|
||||
# storage, logger) stays readable on real hardware with no serial. See etc/init.csv
|
||||
# for the format; this file must otherwise track it.
|
||||
#
|
||||
# service args...
|
||||
/system/services/input
|
||||
/system/services/device-manager
|
||||
/system/services/fat
|
||||
/system/services/logger
|
||||
|
Reference in New Issue
Block a user