Yep, too much C/FFI here :) when there is none already.

Haven't checked much the autosuspend plugin, that I have disabled (so, I won't benefit from autoshutdown, which is fine for me). Haven't checked much of the ffi/rtc stuff either, but reading the other links, I trust you got that right :)

Is that uimanager alarm management abstracted enough from its shutdown use so it can eventually be used for other features (like update rss/news, autoupdate koreader,...)?

Yep, too much C/FFI here :) when there is none already.

Much easier to prototype. ;-) Actually this is the abstracted version of the proof of concept done purely in kobo/device. For the moment I've still got a branch with that here. Perhaps I should add a diff here for posterity. ^_^

Is that uimanager alarm management abstracted enough from its shutdown use so it can eventually be used for other features (like update rss/news, autoupdate koreader,...)?

Sure, it's (purposefully) completely agnostic as to what's done with the wakeup. I'll update it with a wakeup_action or some such later to finalize that.

Although I should add that unless someone writes a wakeup scheduler that manages things, this implementation is one wakeup action only. So that would be either update Wallabag/whatever or autoshutdown. But that's basically unrelated to the methods for scheduling the wakeup.

To clarify, you can only schedule one wakeup. So if you wake up to sync, then afterward you have to schedule either the next sync or the autoshutdown depending on which would come first. I suppose the scheduler could be as simple as a table of upcoming actions in proper order, holding a reference to the scheduled action.

I have no interest in implementing such a thing right now, and I think I'd rather keep it for a follow-up PR even if I do it later.

Diff for the part of the v1 prototype that didn't carry over into this PR:

diff --git a/frontend/device/kobo/device.lua b/frontend/device/kobo/device.lua
index 35edf5d3..6a124352 100644
--- a/frontend/device/kobo/device.lua
+++ b/frontend/device/kobo/device.lua
@@ -534,7 +534,13 @@ local function getProductId()
 end
 
 local unexpected_wakeup_count = 0
+local wakeup_scheduled = false
 local function check_unexpected_wakeup()
+    if wakeup_scheduled then
+        logger.dbg("Kobo suspend: scheduled wakeup for shutdown")
+        local UIManager = require("ui/uimanager")
+        UIManager:poweroff_action()
+    end
     logger.dbg("Kobo suspend: checking unexpected wakeup:",
                unexpected_wakeup_count)
     if unexpected_wakeup_count == 0 or unexpected_wakeup_count > 20 then
@@ -553,7 +559,49 @@ end
 
 function Kobo:getUnexpectedWakeup() return unexpected_wakeup_count end
 
+local ffi = require("ffi")
+local C = ffi.C
+-- for ioctl header definition:
+local dummy = require("ffi/posix_h")
+local rtc = require("ffi/rtc_h")
+
 function Kobo:suspend()
+    local t = ffi.new("time_t[1]")
+    t[0] = C.time(NULL)
+    t[0] = t[0] + 12
+    print(t[0])
+
+    local ptm = ffi.new("struct tm")
+    ptm = C.gmtime(t)
+    print(ptm.tm_sec)
+
+    local wake = ffi.new("struct rtc_wkalrm")
+    wake.time.tm_sec = ptm.tm_sec
+    wake.time.tm_min = ptm.tm_min
+    wake.time.tm_hour = ptm.tm_hour
+    wake.time.tm_mday = ptm.tm_mday
+    wake.time.tm_mon = ptm.tm_mon
+    wake.time.tm_year = ptm.tm_year
+    -- wday, yday, and isdst fields are unused by Linux
+    wake.time.tm_wday = -1;
+    wake.time.tm_yday = -1;
+    wake.time.tm_isdst = -1;
+    print(wake.time.tm_sec)
+
+    wake.enabled = 1
+    --wake.time = rtc_time
+
+    local rtc0 = C.open("/dev/rtc0", C.O_RDONLY)
+    print(rtc0, ffi.string(C.strerror(ffi.errno())))
+    local err = C.ioctl(rtc0, C.RTC_WKALM_SET, wake)
+    print(err, ffi.string(C.strerror(ffi.errno())))
+    local err2 = C.close(rtc0)
+    print(err2, ffi.string(C.strerror(ffi.errno())))
+
+    if err == 0 then
+        wakeup_scheduled = true
+    end
+
     logger.info("Kobo suspend: going to sleep . . .")
     local UIManager = require("ui/uimanager")
     UIManager:unschedule(check_unexpected_wakeup)

Maybe some logic to point out suspend/shutdown will occur first? Only devices with Device.wakeup_mgr (or some equivalent check) will be able to shutdown after suspend.

I initially stumble on that, "task" being assigned an timestamp, while intuitively, a task is a callback.
So, may be just a comment saying "a task here is a combo of (epoch timestamp, callback).

Nothing more to add about WakeupMrg, looks quite fine and explicite.

You save into G_reader_settings. Is there still some reason to read this koboautosuspend.lua and loop thru candidates below? (alt: should you not save them into that plugin setting file, and forger about G_reader_settings here?)

Now that you mention it, one of them is probably intended to be legacy. Given that the plugin-specific settings seem to stem from a migration out of the UIManager core, I'm guessing G_reader_settings is the one that should be deprecated.

good riddance:)

This was convoluted and hard to follow. The changed method Just Works™ (thanks to the scheduleIn/unschedule methods I wrote in 2014 I might add with some pride).

Now you're talking koreaderish.

No pb with that, but autoshutdown delay applies after autosuspend delay, so it's actually autosuspend_set + autoshutdown_sec ?
Does all that work when suspend is disabled, and only shutdown enabled?

The scheduling part above is only for suspending/shutting down while the device is active. On suspend that's all unscheduled (see two lines above), while the WakeupMgr takes care of scheduling the wakeup and executing the scheduled task.

@poire-z This is the answer to your question below in https://github.com/koreader/koreader/pull/5335/files#r322896802