FRLG Roamer: fix leave_pokecenter never actually exiting - #1424
Conversation
travel_from_celio_to_kanto() sends a single 1280ms left-hold before calling leave_pokecenter(), assuming the game is immediately controllable right after activate_roamer()'s dialogue finishes. In practice (confirmed via testing, including on Spanish game text) the character never even attempts to turn left -- it goes straight into leave_pokecenter()'s blind "walk down" phase and misses the exit every time, eventually throwing "Failed to exit PokeCenter." Root cause: a leftover dialogue box from activate_roamer() blocks movement input until dismissed with A. Idle waiting (tested up to 30s) does not help since it never closes the dialogue; blindly mashing A does have an effect, confirming the game is responsive, but risks re-triggering a fresh conversation with Celio if no dialogue is actually left. Fix: - Check via screenshot whether a dialogue box is present before moving, and press A only once if so, re-checking after each single press (never pressing blind). - Move left in 5 discrete steps instead of one continuous hold, so a single dropped input doesn't eat the whole movement. Not RNG-sensitive: this all happens after activate_roamer() has already fixed the roamer's PID, so the extra checks/time here do not affect the RNG result.
|
The problem with this fix is that, if the final dialogue box isn't closed, that means we're not generating the roaming legendary Pokemon at the target time, making it much harder (if not impossible) to actually do the intended RNG manipulation. This is why there wasn't already a mash B step to exit the dialogue, since that would hide any issues like this that come up Are you sure it's still hitting the right targets? Have you gotten a shiny? The underlying problem is likely that some languages (including Spanish) have an extra dialogue box compared to English. The better fix would be to account for it during the blind button press sequence. If you don't want to figure this out yourself, I'd be happy to take care of it if you can provide a video of all of Celio's dialogue |
|
Yeah I got a shiny with the fix in, 4th automated attempt, so it's actually landing hits not just skipping the crash. Don't have the exact seed/advance numbers for that one though so can't say for sure it hit the exact PID Can't get you a video unfortunately, my save's already past that point with Celio so the dialogue doesn't come up again for me, and I couldn't find existing footage clean enough either Did find one thing that might help - an old guide for this method (manual, not language specific) says to watch for Celio's last text box specifically, "Thanks to you my dream came true" or whatever the exact line is, and time the final press off that instead of counting presses. Which lines up with what you're saying, a fixed count is gonna be fragile across languages no matter what number we land on. Might be worth having activate_roamer detect that last box with OCR instead Can help if useful, or someone with a save before that point could probably get you the footage faster than me at this point |
Could you check the logs for what your in-game advances calibration was when you hit your target? If we're right about what is happening, your calibration should be somewhat negative.
Our FRLG programs do everything that is timing sensitive completely blind to make things as consistent as possible. My first attempts at RNG manipulation programs were much less reliable due to trying to detect a couple of things from the video capture. My preferred fix is to tweak the sequence as needed on a per-language basis |
2 participants
travel_from_celio_to_kanto() does a single 1280ms hold left before calling leave_pokecenter(), assuming you can already move right after activate_roamer()'s dialogue ends. Turns out that's not reliable -- in testing (with Spanish game text) the character never even tries to turn left, it just falls straight into leave_pokecenter()'s blind walk-down and misses the door every time, eventually throwing the exit exception.
Turns out there's a leftover dialogue box still open after activate_roamer() finishes that eats the movement input. Waiting longer before moving doesn't help (tried up to 30s) since waiting alone doesn't close a dialogue box, but mashing A does close it -- which also confirms the game isn't actually frozen, just still showing text. Problem is mashing blind risks re-opening a new conversation with Celio if you overshoot.
So now it checks a screenshot for a dialogue box before moving, and only presses A once if it's actually there, re-checking in between presses. Also switched the left move to 5 separate steps instead of one long hold, since the continuous hold kept getting eaten somewhere.
None of this touches anything before the PID gets fixed in activate_roamer(), so it shouldn't affect the RNG result -- tested it and it's still landing on the right targets.
Heads up on the 220ms/step value for the movement: that number was reverse-calibrated from a single test session on my own hardware/capture setup (measured 400ms/step as ~9 tiles for 5 steps, scaled down to land on ~5). It's possible the dialogue fix alone was the actual bug and the original 1280ms hold would've been fine once unblocked -- discrete taps and a continuous hold don't necessarily cover distance at the same rate per ms, so this timing may need retuning on other hardware. Flagging this in case it needs adjustment rather than presenting it as a settled number.
Tested and working on FRLG (Spanish).