todo.txt

Changes

BASE

- tg_040: fixed rtc bug in MinUI.pak
- all: GB.pak now uses the first/lightest palette color for the Grid effect (SDL2 only) 

EXTRAS

- tg_040: fixed rtc bug in Clock.pak
- all: Super Game Boy borders are now optional

---

gambatte
	light grid that matches palette
	proof of concept added to tg5040 with patch for gambatte core
		added patch
		need to rebuild all gambatte cores (manually because building cores is disabled locally)
		need to copy effect color changes to other sdl2 platforms
			+ magicmini (took longer than expected because of faux-brightness)
			+ my282
			+ rg28xx (rotation/resolution is same as my282 but the math isn't adding up here?! fffffuuuuuu)
			+ rgb30?
	people are already asking for more
		light grid for gbc (looks awful)
		someone will ask for the same for ngp/ngpc
	this is really entering shader territory...
		started dabbling in OpenGL and shaders
		hit a wall
	
minput
	seems to be designed top,left aligned for 640x480

it's time to retire some supported devices
	devices running SDL1.2
		gkdpixel
		trimuismart
		rg35xx (non-h700)
		miyoomini (and plus)
	devices that have been superceded
		rgb30
	finish updating README (don't forget BASE readme!)

rg_xx
	usb controller
		look into SDL2 controller API
		stock supports bluetooth right? how?
			enabling bluetooth via stock breaks wifi on MinUI

tg3040
	- 
		
next
	standard show/say/confirm.elf?
		sdl2 implementations require a delay arg 
		because they clear the framebuffer on exit
		so we can't use show + sleep or + confirm
		
		trimui's await_input makes it kind of like 
		a confirm.elf which I don't see in modern
		MinUI...was that in MiniUI maybe? yes

		rg35xxplus		sdl2	show.elf image.png delay
		trimuismart		sdl		show.elf image.png
		miyoomini		fb0		show.elf image.png
		trimui			sdl		show.elf image await_input
		tg5040/tg3040	sdl2	show.elf image.png delay (uses window surface as screen)
		rgb30			sdl2	show.elf image.png delay (uses window surface as screen)
		rg35xxplus		sdl2	show.elf image.png delay
		
		show.elf <image path> <delay>
			all official uses of show.elf are used to display 
			text probably better handled by confirm.elf
	
		show.elf <image_path or message> <delay or type>
			(default) 	A confirms, B cancels (both CANCEL and OKAY buttons drawn)
			<delay>		nothing confirms or cancels, quits after <delay> (no buttons drawn)
			only 		only A confirms, none cancel (only OKAY button drawn)
			wait		nothing confirms or cancels, must be killed (no buttons drawn)
		
	magicmini has no intel toolchain
	
	is Developer.pak supposed to be in the tg5040 extras?
		I didn't think I released that since it's mostly useless to others
		and even to me now that everything is built out and working...
		
	slow buildroot evtest
		https://discord.com/channels/741895796315914271/1232969624417730630/1284606079614849084
	
	mgba (at least on the mm+) doesn't like zipped roms
		try to reproduce on something with adb or ssh
		couldn't reproduce on
			trimuismart
			my282
		I guess maybe tmpfs is just super slow on miyoomini?
	
	redo audio pipeline
		I think fceumm has issues on more devices than not at this point (thanks h700)
		400 SAMPLES fix isn't a true fix, it just minimizes the side effect
		I wonder if picarch's dynamic rate control would help here
			that seems to be a frame pacing fix, not an audio fix
		can I let SDL handle resampling (on newer hardware?)
			pass NULL for received lets SDL resample invisibly (I think)
			would that even help?
		what about SDL_AudioStream? (added 2.0.7)
			https://wiki.libsdl.org/SDL2/Tutorials-AudioStream
			what if the changes to make audio stream public
			in SDL2 is what introduced the issue? :sweat_smile:

	
	clean out this rat nest of a todo doc
	figure out Clock.pak timezone issue
	
	magicmini
		check out silent hill (PS) for simple versus gaussian interpolation
			supposedly awful on simple
		screen does not really turn off during sleep
			it doesn't blank either
			it does if I manually set it to 4 with 351Files :thinking_face:
		
		move amixer cset name='Playback Path' SPK to libmsettings
			see rg35xx,rgb30,m17 for keymon poll examples
			/sys/class/gpio/gpio86/value is meant to be 0 when headphones are connected (only at boot?)
				but its always 1
			so we can only detect headphone change events not initial state
				which makes separate speaker/headphone volumes useless RIP
				
		reduce min brightness
			might require kernel hacking
			added software stopgap
		audio is choppy in places
			even after 400 sample fix
		
	h700
		implement cpu speed (or figure out why it wasn't already implemented)
		
	my282
		ldd isn't working but this is
			LD_TRACE_LOADED_OBJECTS=1 /lib/ld-2.23.so /mnt/SDCARD/.system/my282/bin/minarch.elf
		
	emu pak detection isn't working when two folders are combined
		eg. SFC and SUPA folder but no SUPA.pak still shows games from SUPA
	all
		look into how many cores support 8888
			this might be a bottleneck (if it's having to go 8888 to 565 and back to 8888 for every pixel, every frame)
			not a single BASE or EXTRA core requests 8888, so that won't help even if it is a bottleneck
			
		Thread Core
			needs changes to support systems that change resolution (eg. PS)
			doesn't seem to have any affect on a30 (mostly because the problems are hi res games and the copy from cpu to gpu)
			disable and remove this feature entirely?

		revisit bub181's core # vs freq ideas for other devices
		
		look into RETRO_ENVIRONMENT_GET_CURRENT_SOFTWARE_FRAMEBUFFER for zero copy framebuffer
			not sure it's really a zero copy because LockTexture has to copy the texture from the gpu to the cpu...
			
			times on a30's SDL
				GB: SDL_UnlockTexture blocked for 3ms (160,144) (TLOZ:LA)
				PS: SDL_UnlockTexture blocked for 5ms (368,480) (TEKKEN3)
				PS: SDL_UnlockTexture blocked for 8ms (512,480) (BR2)
				PS: SDL_UnlockTexture blocked for 11ms (640,480) (CMR2)
				try using SDL_UpdateTexture instead of Lock/Unlock
					might be saving 1-2ms
				could this slowness be related to vram speed? (DMC?)
		
		update all cores
			call out breaking changes (gpsp save states) in release notes 
		add Clear Recently Played.pak, copy to all platforms
			maybe include confirmation?
			requires show.elf
		enable SELECT/START+L/R for volume/brightness
		limit option name replacements based on core lib name
		noui
			quit doesn't work as expected on trimuismart
			possibly because of absense of shutdown?
	h700
		hijack boot earlier?
			hangs boot
			reflashing a fresh stock image takes too long to happily trial and error this
			was able to "takeover" /etc/init.d/launcher.sh and inject it into systemd's basic.target
				but it didn't improve boot time
		look into using fbdev directly
			I was able to get this working
				SDL_NOMOUSE=1 required for SDL1.2 to work
				WAITFORVSYNC doesn't appear to work
				more notes in separate 28xx/notes.txt
		
		muos uses sdl 1.2 fbdev_mali+alsa for retroarch...which what
			refried beans uses SDL2
			no underruns in fceumm without threaded video...even with the sdl2 instead of default gl driver...
			i tried copying muos libsdl2 to minui but I think I grabbed the 64bit version, is there a 32 bit version?
				yes but it's built with a newer version of glibc than stock
				so I'm back to having to build my own rootfs
				or building sdl2 from muos with my toolchain
		stock uses sdl2 fbdev_mali+alsa for retroarch
			stock retroarch has frame pacing issues in AoS/gpsp 
				without threaded video
				fceumm has underruns without threaded video
		minarch has the same issues on refried muos
			needs modified lib path (because muos is 64bit by default)
				export LD_LIBRARY_PATH=$SYSTEM_PATH/lib:/usr/lib32
				might have also had to hide MinUI's libSDL*
				SDL_BLITTER_DISABLED=1 has no effect on underruns
		
		diff older version of stock that doesn't underrun so bad
			compare with rg35xxplus (I don't remember having audio issues there)
		the problem is the panel is being driven at 59fps
			nope, fixed that with boot_package dtb patch and it still underruns
				wtf
			two things mitigate the effect of the problem
				change SAMPLES to 400
					makes underruns much less frequent and much smaller
					because fceumm fills the buffer ~799 samples at a time
					works
					also bumped buffer to 8 seconds then reverted to original 5 seconds
						see if improved underruns persists without or if that a component of the fix 
				instead of zero filling on underrun fill with last frame values
					makes the more regular 1-2 sample underruns almost unnoticeable
			
			is the batching logic responsible for lag somehow?
				there's an SDL_Delay and a loop
				but doubling BATCH_SIZE and switching to usleep(500)
					didn't seem to help
					SDL_Delay(1) actually delays for 1ms on this chip
			tried renicing to -20, no help
			played with taskset to no help
			gba AoS judder persists even if audio is silenced
			GFX_Flip blocks on tg5040 too, so many red herrings
			I disabled audio entirely (didn't even initialize it) and the judder persisted
			there's something rotten inside SDL_RenderPresent()
				retry compiling sdl2
					figure out where SDL2-2.30.4 with mali came from (knulli? rocknix?)
					no good, this one works https://github.com/JohnnyonFlame/SDL-malifbdev-rot
						solves judder
		
		rebuild using opengl?
			based on steward-fu's work
				https://steward-fu.github.io/website/handheld/rg28xx_cpp_es_triangle.htm
				https://steward-fu.github.io/website/handheld/rg28xx_cpp_es_texture.htm
			
			# add stock libs and includes to toolchain
			# what was sysroot.tar.gz created from? just /usr on the device? probably
			
				tar -czf sysroot.tar.gz /usr/include /usr/lib /usr/lib32
				
			# will convert to xz in docker
				
			mv sysroot.tar.gz /opt
			cd /opt
			mkdir h700
			tar -xvzf sysroot.tar.gz -C h700
			rm sysroot.tar.gz
			rsync -a --ignore-existing /opt/h700/ /opt/rg35xxplus-toolchain/usr/arm-buildroot-linux-gnueabihf/sysroot/

			# build flags
			-lGLESv2 -lEGL -lmali -ldl
		
		tried building sdl2 myself it crashed and burned spectacularly
		tried using taskset to bump priority but it already appears maxed
		
		

	/tmp/noui
		show poweroff on quit in minarch if present
		tricky because of thread core option...

	pretty sure nothing uses the various platform.h BUTTON_* macros (because SDLK_* mappings are so spotty on these devices)
		pretty sure they all use CODE_* or JOY_* now

	macos (for modern compiler and debugging tools, not for release!)
		minarch
	
	rg35xxplus/h
		figure out
			audio crackle on quit sdl
				setting raw volume to 0 doesn't fix
			implement jack
				sysfs state doesn't update
			check hdmi
				nothing
				/dev/hdmi?

	unzip60
		rename Makefile.trimuismartpro to just Makefile.minui?
	gkdpixel
		usb mode
			rndis? related? /etc/default/configfs
		bootlogo?
			haven't located on disk or in memory
		figure out dual sdcard support for mini?
		non-plus snes9x2005 for perf?
		codify alt paks to use stock standalone emus in MinUI
		light bleed
			experiement with /sys/devices/platform/jz-pwm-dev.0/jz-pwm/pwm0/period
				default is 10000
				nothing
		graphics
			jitter while scrolling
			but only on randon launches
			othertimes it's perfectly smooth (same core, same game)
		subpixel scaler work underway
				gb/gbc
				gba
				nes/snes
	
	native Files.pak
		I might just need to accept some duplication with MinUI (eg. Array and Directory/Entry arrays)
	m17
		still haven't found way to disable volume controls
			I think it's controlled by /usr/bin/gpio-volctrl.sh
			but rootfs is a 500MB ro squashfs so patching isn't really ideal
	all
		investigate Clock.pak issues (timezone related primarily)
		evtest
			random note: doesn't know about /dev/input/js0?
		standarize /bin/shutdown
			while (1) pause(); prevents shutdown on some platforms
			just system("rm /tmp/minui_exec") and exit(0) ?
		.res
			use standard naming (replace ext with png instead of append png)
		better document the new Bootlogo.paks
		
		collections can't handle accented characters (eg. é) in path/rom name...
			wait, why can recents.txt but not collections.txt?
			because one is written by the device so it's using device native accented chars
		update stored rom path logic to only require TAG and not full folder name? (would simplify Collection authoring?)
		add make fetch to get all external sources before build?
		add Developer.pak
			only for devices with adb or ssh
			
		rewrite Shortcuts section of readme by device not the tangled mess it currently is
			maybe do the same for all?
			break out platforms into completely separate installers/updaters?
				will require revisiting unzipping system
				will need to inject not replace
					but then how to handle removing deleted files?
					seems like this is what currently happens on some (all?) platforms
		look into mednafen/beetle ngp core (neopop?)
			beetle-ngp-libretro
			this ran at about 1 frame per 10 seconds...on rgb30
			add logging
		wonderswan
			runs at 1fps
			possibly because minarch forces 60fps but this handheld runs at 75?
			
	tg5040
		create PSP.pak and NDS.pak and document how to setup by copying assets from stock?
			
		can't detect headphone state at boot?
		wifi
			Splore.pak can pull updated lists from bbs but cannot download carts
				no idea why
				wget works on stock so I'm failing to initiaize the network properly
					this gets ping working: udhcpc -i wlan0
					but wget is not compiled with ssl support
					the following did not work to enable opkg to install/update wget
					
						echo "src/gz openwrt_core_cortexa53 http://downloads.openwrt.org/snapshots/packages/aarch64_cortex-a53/packages" > /etc/opkg/customfeeds.conf
						echo "src/gz openwrt_core_sunxi http://downloads.openwrt.org/snapshots/targets/sunxi/cortexa53/packages" >> /etc/opkg/customfeeds.conf
						
						no valid architecture
					just remap wget calls to curl (which has ssl support)
			need to patch pico8_64
			
				xxd old > old.hex
				xxd new > new.hex
				diff -u old.hex new.hex > pico8_64.patch
						
		n64
			requires libretro opengl implementation
		psp
			after adding xrgb8888 support to MinUI it still chokes on a number of missing libraries 
			and eventually an unlinked function (something like XUnlockMutex)
		
	user reported
		random reports of minui not booting reliably 
			first smart
				can't reproduce
				already waits 3 seconds
			then rgb30
				related to orphaned clusters issue? likely
				if so it needs to either be solved in MOSS/JELOS (RIP)
				or the other platforms MinUI supports (the non-Plus Mini in particular has a really terrible shutdown process at least on older firmwares)
		
	move PLAT_isOnline() into PWR_* formally
			
	all
		revisit localization?
	
	minarch:
		crop save state pillar/letter-boxing?
		force aspect over hdmi? (or at least disable cropped?)
	
	rgb30
		reboots with time offset by timezone offset
		does jack detection work?
		could wifi support more than one access point?
			an additional two lines for each AP?
			then try them in order?
			
		figure out why power suspends the device after MOSS install
			it appears to reboot but it only stops suspending after a manual reboot
			fewt has a fix?
				https://github.com/JustEnoughLinuxOS/distribution/pull/2195/files#diff-719d790f2d3c1d5a0884ae9eb69fb77956fcaaae235282acb6e6817339f35105

		moss
			build lib32 to try to fix snes9x save state compatibility
				first attempt did not go well :sweat_smile:

rpmini
	dead in the water waiting for upstream fixes
	evaluating batocera and rocknix as a base
	requires ota update to enable booting direct
		might also enable installing linux to emmc
	currently seeing scaling artifacts (variable width pixel columns/rows) when blitting 1:1
		causes grid effects to look terrible
		suspect it's the sytem level rotation
		hoping it's not the physical pixels in the panel
		see rpmini/tmp/sdltest
			not going to work on this anymore until kernel fix is released
				
known issues
	all: race core (NGP/C.pak) doesn't restore audio state correctly
	rgb30: light crackling in some cores (eg. fceumm and snes9x2005_plus)

all
	minarch
		GFX_blitMessage() screen w/h is 0,0 for some reason
		prevent MENU+button from triggering button
		prevent binding same MENU+button to control and shortcut at same time
			reset the newer input to None and show a message, ideal something like:
				MENU+<button> is already bound to (<name> shortcut|<name> control)
	clock
		factor in timezone offset?
	simple progress animation for installing/updating
		launch as a daemon with &
		draws in a corner
		kill after install/update complete
		if it's circular we don't even need to worry about rotation :sweat_smile:
	main makefile
		targets have too many dependencies
		getting a little better
	cores
		add optional per core patch folder
			currently looks for a global one in all/cores
		move all the CORE_* vars up to all/cores (since they ended up all being the same)
		there has got to be a better way to main these patches
			updating fake-08 was so much copy/pasting
	minui
		alt tag is getting truncated, should be added after truncation
	add a "res" (or "many"?) folder 
		for install/update and charging gfx?
	standard
		show
		say
		confirm

smart
	cores
		supa starfox perf bad

miyoomini
	separate settings file for mini and plus?
	battery overlay

miniplus
	-
	
rg35xx
	revisit hdmi based on rgb30 implementation?
	overclock.elf with userspace versus performance
		performance should be slower because it will max out to the max defined scaling frequency
		userspace should accept the specific scaling frequency
		re: NDS.pak
		
		XQuader figured out that CPUs are unlocked with performance but not userspace
			the following will unlock them:
				echo 0xf > /sys/devices/system/cpu/autoplug/plug_mask
				echo 1 > /sys/devices/system/cpu/cpu1/online
				echo 1 > /sys/devices/system/cpu/cpu2/online
				echo 1 > /sys/devices/system/cpu/cpu3/online
			but using performance instead of userspace has the same result so maybe just use that
				as long as it doesn't kill battery life?
			turns oout this was discovered back in may by another user:
				https://www.reddit.com/r/RG35XX/comments/13rtrqb/minuis_snes_performance/
				https://github.com/shauninman/MinUI-Legacy-RG35XX/compare/main...zoli0726:union-minui:main
				seems like this is sufficient to unlock the improved performance
					echo 0xF > /sys/devices/system/cpu/autoplug/plug_mask
					though I'm not sure I'm observing a difference...(at least when applying over adb)
		
	charging screen isn't working for someone on RH discord
		charging.png isn't being copied to MISC during install (it should only be omitted during update)
	how to handle updating on mini and missing TF1 updates as a result?
		could I write the version to the .minstalled file
		if it doesn't match the current one run the installer
		trimui smart has a similar issue now
			the bootstrap folder updates /etc/main

rg35xx/install and miyoomini/install have png/bmp versions of the same files
	as do miyoomini/dat/ and rg35xx/ramdisk/

---

FIXME: main makefile is a brittle mess
	each platform has its own default cores and extras
	so it's a tangle

---

1. makefile routes requests through makefile.toolchain
2. which clones the toolchain repo if necessary and
3. spins up a docker to run workspace/makefile build for each/the target platform
4. which builds commone elfs and punts to workspace/PLATFORM/makefile
5. which builds unique platform elfs

there's more to this now (eg. PLATFORM/makefile.copy)

---

replace 
	git clone repo && cd repo && git checkout branch 
with
	git clone --depth 1 --branch branch repo
note: doesn't work with commits only branches