Oh, that's strange. So you can actually read (just 1) past len?

Yeah, it's made to feel like strlen in a way. Sentinel types are mostly used with C code. That way strlen(some_string) == some_string.len.

Ah, I hadn't thought of that analogue, but it makes sense, then. It sort of recreates the opportunity to make the common bug, though, of allocating enough for the data but not the sentinel (or enough for the string but not the null terminator).

Not really, the sentinel is part of the type and the allocator interface in Zig is type-aware, meaning that you can't simply "forget" about sentinels. The type system will fight you.

So creating a pointer to a type with a fixed length and a sentinel and creating a new one will automatically allocate enough for the length + the sentinel? That's nice.

I'm still not sure I like the .len 'misreporting' since it would be harder to manually optimize around, e.g. packing a struct with a 8 byte + sentinel array would take 16 bytes that I might not notice if I am going by `.len`. This is all speculation, though. I need to actually play around with Zig.