Javier

> does not seem to be important or useful for loading values from fields sure it is! the generated code for a pointer _is_ different from the code for the...

as so many things in luajit.... by reading the code and writing small tests. did a talk on ctypes last year: https://youtu.be/IUtJJhQSLek?t=1939

I think that the distinction between `struct s` and `struct s*` are important, and it's the same as in C. removing the first would make something like `struct s arr[100]`...

i think (without any evidence whatsoever) that the high cost of FFI in the interpreter is not only because of allocations and constantly boxing/unboxing values, but also because any operation...

there are a few different lists that an object might belong to, depending on the type of object and/or the phase of the GC. all of them use the `.nextgc`...

depends on #337 to fix (remove) the `depguard` linter. until that, no PR is likely to pass.

> as a simpler approach? It seems like the first will imply the latter two are also going to fail. yes, that should work too > That aside, what's happening...