Jonathan Coates

> The existing code only supported using snbt files. Ahh, you're entirely right. I think I had registered this at some point, but forgot by the time I came to...

Looking at this patch again, it occurs to me that falling back to vanilla behaviour isn't correct either, as the template name now contains the mod id, so this will...

I guess the problem with `StructureTemplateManager.Provider` is that it also doesn't provide the paths it has tried. I don't know how much that matters?

@timofonic This is mostly blocked by #12 (and me needing to do some research in how [Eris](https://github.com/fnuecke/eris) stores its data).

Well, this is embarrassing. It only uses the array part. I suspect the fix here is to shift everything to use `rawget`/`rawset` instead of indexing `array` directly. https://github.com/SquidDev/Cobalt/blob/859205c40f1366dcf1bfff71ef7cd38038bcb221/src/main/java/org/squiddev/cobalt/LuaTable.java#L645-L675

Technically we should actually be supporting `__index`/`__newindex` here. Ughrhrhrrh.

A minimal reproduction: ``` lua for i=1, 1 do local ca for v = 1, 1 do if v then ca = v end end assert(ca); (function() return ca end)()...

It appears that the upvalue is being opened on the for loop, resulting in it being cleared each iteration. We either need to defer opening the variable or move opening...

Gotta love how broken GH actions is at times.

This is looking really nice, thank you. I'll try to have a fiddle later today, maybe load the logs within JITWatch and see if there's anything obvious which is problematic....