Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
If you keep the controller latch signal high at all times, the controller should in theory continuously output the active state of player 1's A button to $4016.D0. This would help improve throughput, although from my testing just now, it seems like SubNESHawk doesn't actually support changing input mid-frame in this case.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
Interestingly, this submission is old enough to feature this now-incorrect comment:
Submission text wrote:
Furthermore, so far, nobody in the world has been able to write a computer program that plays Go better than an average club player.
There is this completion of level 14 (as opposed to the level 10 chosen by Bisqwit):
Link to video
It is slower than Bisqwit's run, but not by much - possibly faster if you were to discount menuing and thinking time. Although it is played on some English-language version, which could conceivably have a different AI.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
The movement could be made faster, but then why stop there? There's really not much limit as to what you can speed up, and eventually you'd just be ending the game as soon as it starts.
The hack strikes a decent balance as-is, I think. Maybe the choices Acmlm made were arbitrary, but they make for a fresh experience and at this point the hack is over a decade old and much better known on the internet than even some of the games we have published runs of on the site.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
There have been some issues with getting this movie to sync on devices other than the original one used. With debug logging enabled, the line "Loaded SWF version 8" is logged with a timestamp of 02:04.001 when the movie is run on the original PC with -g vulkan, and 2:04.000 in all other cases I've tested. The 1ms difference seems to cause a difference in the initial RNG seed. This can actually be remedied by setting the initial system time forward by 1ms, but then, later on, another issue arises with the GetTime action returning a value that's different by 1, which eventually causes another discrepancy, and not one that can be fixed by changing the system time. However, I'm not sure why all this happens in the first place, or if the movie can still be salvaged. Hopefully someone else will be able to figure this out.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
Arcorann wrote:
GJTASer2018 wrote:
adelikat wrote:
Avoiding this potential crash to get to that point I think is inherently interesting, especially since no human or bot has done this (yet), so it shows off something unique to TAS.
That was done using a Lua script to avoid the crash, so it doesn't count.
Since the crash is related to obtaining high scores in a single line clear, the easiest way to avoid the crash is to stop scoring Tetrises after a certain point. I think it'd be more interesting to investigate if it's possible to route around it and still score Tetrises, though. (In fact, this submission continues to score Tetrises until 6 levels after the video with the AI crashed, proving that it is indeed possible to some extent.)
Crashing actually has to do with an interrupt happening right in the middle of some code that handles jump tables; it's just that this doesn't happen until quite late into the game, when the score addition logic starts taking a significant amount of time. This is a delicate enough occurrance that just making small changes like delaying the line clear or toggling the next piece visibility can be enough to avoid the crash. There is a legitimate movie that gets the level counter to wrap around and even scores a tetris almost every time. Well, somewhat legitimate - I'm pretty sure FCEUX is not accurate enough to emulate this rather timing-sensitive behavior correctly, and that's backed up by the fact that the movie does run into a crash on Mesen.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
Sigma 7 wrote:
There's already a minus world run. See: https://tasvideos.org/1365M
The other minus world run is slightly faster (after adjusting for FDS loading time), and also qualifies as a complete run.
These two runs are technically incomparable, since they go through a different set of levels.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
The supposed claim seems to be that versions specifically chosen by the developer as betas can be downloaded fine, but the more general way to get any past version of the game via eg. DepotDownloader won't work.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
It's just when there's a lot of PPU data updates queued up (like a new row of blocks getting loaded, an update to the blinking palette, score/coins change). The buffer reserved for these updates is not always big enough, so some corruption happens beyond it.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
I've developed a TASing utility for this game called Fling, which Snowball has now taken over and has been updating for a while. For now it is private between a few trusted members of the community and closed source, because it includes source code of the game and is a somewhat playable version of the game.
As of recently, a TAS exists for every maingame individual level, thanks to the efforts of Snowball, JSL555, Mijzical and others.
List of times
Video of an "all chapters" run (already slightly outdated, because the project is constantly being worked on):
Link to video
Comparison between the game and Fling:
Link to video
More specific TAS discussion about this game goes on in the game's Discord server.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
Link to videoLink to videoLink to video
Seems to me like most of optimizing this game is just getting the scroll speed ($d0) to its maximum of 217 as quickly as possible and keeping it there.
Also, there have been somezips found.
Experienced Forum User, Published Author, Active player
(354)
Joined: 11/25/2012
Posts: 102
Location: Europe
This should have been posted a long, long, LONG time ago, but at least better late than never. Mudkip and I won't be finishing the Street Fighter 2010 run. There wasn't really much progress at all - for posterity, here's the movie file: link. Many apologies to anyone who wanted to take on the work during this time and/or was anticipating the finished result.
