Eldred Habert

> You would expect `db "{s}"` to declare `STRLEN("{s}")` many bytes, but actually `STRLEN` undercounts since there are multi-byte UTF-8 characters. `STRLEN("héllo") == 5`, but `db "héllo"` declares 6 bytes,...

Given https://hsivonen.fi/string-length, I'm for option 4 as well.

Alright, this looks like a good start! Should it go before the memory map? I think terms introduced in the architecture could then be used to better explain the memory...

Yeah, I think this should instead be superseded / expanded on by a more global fix to #194.

According to `perf annotate` on pokecrystal's `main.asm` file, 31% of the CPU time is spent copying the `yylval` type, which is 264 bytes large (0x108). Moving to variable-size strings (#650)...

Imo that's a bit of a crapshoot given that `-fanalyzer` is GCC-specific, clashing with #283. Possible alternatives: - Only run `-fanalyzer` in CI, as a separate job - Only run...

Clang does not warn about unknown warning flags thanks to `-Wno-unknown-warning-option`, but the opposite is not true of GCC (it warns unconditionally of unknown warning flags, unless it's a `-Wno-`)....

That's not compatible with BSD Make.

As it turns out, at least GCC 12's analyzer is over-eager, for example not understanding that this does not leak the contents of `sectionList`: https://github.com/gbdev/rgbds/blob/d51ab3520328fe7bd82f37da276180c3d7d4059d/src/asm/section.c#L385-L387 I'm not sure we want...

Plus, building the parser with the static analyzer OOMs on my machine (and I have upwards of 3 GiB of RAM available!), so, uh...