Memory Card save corruption with Crash Team Racing NTSC-U

[rubenllorente]

Platform

I am facing an issue on the following platform:

RetroArch 1.21.0 x86_64 (Steam build running on Steam Deck) - Frontend OS: Linux v 6.11
Beetle PSX HW 0.9.44.1
BIOS File: scph5501.bin
Memory card format: libretro

The game on which I can trigger the bug is Crash Team Racing (USA) which exists in bin/cue format.

Bug description

It seems like saving a game to (virtual) memory card has a chance of saving a corrupted game. This is specially noticeable when saving high scores for time trials and relic races: sometimes the save file within the memory card will record bugged information, ie. the times for a particular circuit will be bugged. Once this happens, trying to view the high scores for that track causes the game to crash. It is the game that crashes, not the emulator.

It seems like some sort of overflow takes place within the memory card data structure because you can tell the glitch is happening because unlockable items are unlocked at random when this is triggered - this is, characters and rewards you are supposed to unlock via play get flagged as available around the same time the high scores get messed up. Using a savefile editor shows the game information for affected tracks is all broken.

I am not using savestates at all. This bug triggers inconsistently. A search online does not reveal any known bug in the original game itself.

I have switched to Mednafen style memory cards for the time being. I will report if corruption is triggered under such mode.

Further help to debug the issue would be appreciated.

[rubenllorente]

I reproduced the bug using Mednafen-style saves.

[LibretroAdmin]

Dug into this one — diffed Beetle's memory card emulation against upstream Mednafen's memcard.cpp (write protocol, XOR checksum, bounds check, the NV read/write paths). It's a faithful port, nothing in the card code itself that'd corrupt a save.
That actually fits your report: you said it happens on both libretro and Mednafen save formats, and those are byte-identical internally in Beetle — so it's probably not the save handling, but something happening before the data gets written.
Quick question though — what are your CD and CPU settings?

CD Loading Speed — native (1x) or boosted?
CPU Overclock — on or off?

The docs warn the loading-speed speedhack can cause "texture corruption, timing glitches, or softlocks" in a lot of games, and what you're describing — garbage score blocks plus random unlocks popping at the same time — reads more like timing corruption hitting CTR mid-save than the card actually going bad.
If either of those is on, could you set CD speed back to native, turn off the overclock, and try to repro? That'd rule it out fast. Also worth testing against standalone Mednafen rather than SwanStation — since Beetle's a port of Mednafen, if it's clean there but broken in Beetle, that's a real regression and narrows it down a lot. If it still happens at native timing, I'll keep digging on the seek/CD-timing side.