Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
I'm currently working on a maximum score TAS of Sayonara Wild Hearts. During an specific level, I ran into issues attempting to recreate an specific source of points the RTA runners obtain.
I was able to produce a frame-by-frame replica of a video by an RTA runner demonstrating the trick, and despite matching inputs as closely as possible, TAS does not get the points:
https://youtu.be/mAxT4yj5JRQ
SWH has to be TASed from a Linux user-port of the Mac App Store build, which is not available on Steam. This had to be done since the Windows build requires Direct3D to run, which blocks it from being user portable. The official Steam Windows port released two months after the Mac, iOS and console ports. As such, there's a chance they could contain different code.
Would the fact that TAS cannot obtain this source of points hurt its judging due to suboptimality? Is the fact that the trick has been done on the Switch version going to hurt the argument for potential version differences?
Furthermore, since SWH is an unity game, reverse engineering the code is reasonably easy. Are there any issues with me using reverse-engineered code to potentially prove this discrepancy in behavior between versions? Would it be required to argue that the suboptimality comes from version differences?
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
feos wrote:
RTA runs are missing the noise blocking your action in the last 2 levels. Is it related to running the game before?
Link to video
Thank you for helping me look into this. Turns out our enemy foreground_sleep_ms is at it again!
The official build for PICOHOT assumes the old value of 2 ms for foreground_sleep_ms in ~/.lexaloffle/pico-8/config.txt. By deleting the file and then creating it naturally (by running the game outside of libTAS) I was able to fix the issue. Naturally, however, this causes the movie to desync.
I've uploaded a resynced movie, which is also a 91 frame improvement
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
I've just finished a TAS of The Lost Strawberry but I have a kind of important question:
The game has a very weird artstyle which is not helped by the honestly hideous color palette it uses by default. I can change the palette from the PICO-8 pause menu, but doing so would obviously be slower for no gameplay benefit so I choose not to do it. However, I'm bothered by the fact that the game is hard to follow (and at times hard to look at) in its default palette.
I have yet to test if the TAS will sync if starting from a previous visit to the pause Menu done solely to adjust the palette, then reset. Could this be discussed with the publishers in case of a potential publication?
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
GJTASer2018 wrote:
But why were the delay values different in the first place? Is it something user-configurable?
I suspect the default may have changed in a recent PICO-8 update. I've had PICO-8 installed on Linux for a while (as I did RTA speedruns of Celeste Classic and some of its mods) so that may have set it back then.
This screenshot is actually interesting, since for both kileye and myself, the SDL_Delay calls cause the game to sleep for 5ms each, but your print shows the SDL_delay calls on your side only last 2ms. They make the exact same amount of calls per frame it seems, but since they sleep for different amounts of time, they cause different behavior altogether, which is probably where the desync comes from.
Here's the information my debugger shows if I ask it to print SDL debug info frame 0.
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
Iconoclasts TASing has come a long way. It's just about a year since I started no airswim, and as the different categories keep coming, as we dissect more and more about the inner workings of the game, we keep producing better things. Serena's work in the game never ceases to amaze me.
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
CasualPokePlayer wrote:
ikuyo wrote:
I attempted to download [4594] Linux Iconoclasts by Serena in 22:47.36's ltm file and while the download finished, the zipped ltm file (serena-iconoclasts.ltm) was not recognized as a valid ltm file for libTAS or even as a .tar.gz file (it appeared to be either corrupted or empty).
Downloading the zipped submission from #7292: Serena's Linux Iconoclasts in 22:47.36 gave me the file she uploaded (iconoV2.ltm) which works as expected.
It seems to be a valid .tar.gz file as far as 7zip sees. The SHA-1 also matches the submission file. Might be something weird during that one time you tried? Try downloading the file again.
Just did, got the same results :(
here's the file as shown by Ark. It's worth mentioning it cannot recognize the method for the file, altho the checksum matches the submission file
here's submission for reference
wouldn't be surprised if this is just ark sucking tho, so I'll do an unzip thru terminal just to make sure
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
I attempted to download [4594] Linux Iconoclasts by Serena in 22:47.36's ltm file and while the download finished, the zipped ltm file (serena-iconoclasts.ltm) was not recognized as a valid ltm file for libTAS or even as a .tar.gz file (it appeared to be either corrupted or empty).
Downloading the zipped submission from #7292: Serena's Linux Iconoclasts in 22:47.36 gave me the file she uploaded (iconoV2.ltm) which works as expected.
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
Noxxa wrote:
Cool run. I remember playing this back in the Newgrounds days and really not liking it. Funny to see it's still as broken as I remembered.
Does Illusion not do more damage vs Master/Crazy hand than up smash?
I could only get 1% damage per frame with grounded illusions, and you run into the issue of controlling movement so Fox doesn't fall off otherwise. I'll take a look at aerial Illusions anyway.
For the part where I follow Master Hand, I can at least say that Illusion would end up getting interrupted by the hand's hitbox and I'd take damage.
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
feos wrote:
Kibuka is the only boss that is slower than the human record (by 1 second), but I watched them side by side and couldn't figure out why. ikuyo halp
Ok I took at the videos side by side and you may have helped me find a significant optimization lol.
It appears that Royal will release his power burst on Kibuka faster in the RTA video, which makes the RTA record gain time after cycle 2 in Phase 1.
I'm not entirely sure what causes this. Footage seems to indicate that since Kibuka is closer to the target in the RTA record, Royal attacks faster. I'll have to verify with some changes. Main issue will be resyncing the run since this boss happens quite early and the RNG will quickly change across all remaining fights.
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
I feel it is also important to add this:
The Airswim route makes use of the fact that even tho Impact Zone is not supposed to be accessed until you've made it thru Midway (the space corridor), the loading zone for it is always loaded beneath the Spine Controller room in Blockrock. As such, by performing a good sequence of airswims, you can reach it way before intended. However, because you never went to Midway, the endgame state for Blockrock is never set. Since the cutscenes all make use of whatever flags are sets for their rooms, even at the end, the game breaks because it has the wrong version of Robin's house loaded.
The only way to get credits to trigger correctly would be to reach Midway (as the No Airswim movie does) which would add sizable differences. I'm not even sure how you'd reach rocket platform fast from leaving shard. The only way I can think of would be to maybe use the key zip found for no airswim but then you gotta leave airswim state and the route gets complicated
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
Also, in the context of IL discussion, this includes stuff that we would not necessarily consider ILs, such as:
- Mods and fangame content in general for example (I know Undertale fangame boss showcases are quite popular among Undertale TASers, to name an example). For example, while TASing the entire Spring Collab for Celeste would be a massive endeavor, TASing specific levels is quite accessible and has been done.
- Proofs of concept for specific setups (SM64 ABC stuff, general challenge run concepts, or even experiments for other TAS or RTA runs)
In general, stuff that is just "partially" part of a full game TAS that could take a long time to actually adapt into a full game TAS. Heck even something as let's say, cutting a frame from one of the RNG heavy ILs in Melee Target Test would probably require tons of effort in order to ensure RNG sync for a full game submission which would not necessarily be worthy for such an small improvement, but by allowing ILs, we give a proper place for documentation and archival, and allow people to see these improvements. We all win!
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
In order to obtain an standalone, you can get one using the proper PICO-8 console. We provide one thru our community downloads: https://celesteclassic.github.io/downloads.html
Note that this includes two executables, the main "vanilla" one and a practice mod used by RTA runners. Vanilla was used for this movie.
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
While it is true that most bosses die in similar fashions, the movie is really good at having interesting movement. It might be a bit repetitive, but I see no reason to not have it.
Yes vote.
Experienced Forum User, Judge, Published Author, Active player
(495)
Joined: 7/8/2021
Posts: 98
Samsara wrote:
feos wrote:
Haven't read the thread, only the OP. How about this:
Any user can make a submission and set it to a certain status (not just New or Canceled) that would mean the same as sending it to the Showcase class. After that it won't have to be judged, but would still have its own dedicated discussion thread (you can't even subscribe to userfile comments right now), properly working submission page with all the usual stats, etc.
In the relevant game related page like http://tasvideos.org/Game/nes-battletoads.html such submissions would show up in their own tab. I think it would be fitting to call this tab User Playground. Or just Playground. And different branches would need to be grouped together somehow.
Yeah I suggest calling this hole concept Playground, and make it fully managed by users.
I've added this to the OP for a little more exposure. This feels very strong to me, with my only concerns being similarities to current Moons and the nature of user judgement as a whole, though those aren't huge concerns and they should be able to be rooted out fairly easily.
How does this sound to everyone else?
I really like this proposal. It would allow to show stuff that would probably never get accepted as a regular submission, such as #7265: trapmix21's NES Tetris "Pause Manipulation Showcase" in 01:34.86, but it could also be extended to things like IL runs (which are common in several communities), proofs of concept, Human Theory TASes and others.
You'd probably have to introduce some specific guidelines for posting to Playground, more in lines of obeying site rules/CoC and proper attribution, but otherwise keep the process as free as possible. Ideally, encourage encodes to be submitted to make it easier for these posts to get attention.
here's submission for reference
wouldn't be surprised if this is just ark sucking tho, so I'll do an unzip thru terminal just to make sure