I put in a random 8 char code one time on SMB3, like 20 years ago, and got that extended 1-up jingle. Surprised the crap out my friend and I. Unfortunately didn't remember the code, really wasn't expecting any result from that.

cc65 is great, but I'd recommend llvm-mos or vbcc over it, these days. If performance matters.

Mapper 7 games are mostly by Rare, and they often do insanely tricky stuff requiring cycle perfect PPU timing.

Bigfoot hill climb event is another example.

Right now I planned 8-bit wide mapper register, vs the MMC1-like serial one, that's my last resort to free up I/Os.

My thinking is that lots of WRAM makes it practical to hold pattern tiles as code, load immediate / sta $2007, so for 8x8 modes one can still get by with limited CHR bankswitching.

You still have to populate all 32x30 entries, then it would fetch pattern data from 4 different banks for each metatile quadrant.

The only thing that worries me a bit is only having enough I/O for 32kB CHR RAM, hopefully that's not too limiting.

The only PPU inputs are PPU RD, A0, A5, it's crazy.

Getting back to my old habits, designing a strange new #nesdev mapper with features nobody has asked for. Accumulated ideas over the past 9 years since GTROM.

Notable features would be 128kB WRAM, no parallel flash, 16x16 pattern and 8x8 attribute table modes, PRG bank at $4080-$7FFF. Could be fun.