table is special cased to {any:any}, but should it allow any record?

[mtdowling]

table is an alias for {any:any} based on this test case: https://github.com/teal-language/tl/blob/master/spec/subtyping/table_spec.lua#L3. I assumed that record is a subtype of table, allowing me to pass in any record, {any}, and {any:any}. If it's relevant, my use case is that I want to accept any value that could have a metatable.

local function f(x: table) -- < wanted this to accept records too
end

local record R
end

local r: R = {}

f(r)

Fails with:

argument 1: can't match a record to a map with non-string keys

If it's desirable to distinguish between records and tables, then if there was a type like record that accepts any record, then I suppose I could do table | record.

[hishamhm]

The table notation was a very early addition back from when Teal was still bootstrapping itself. It ended up kind of sticking around. I don't even use myself anymore, and just use {any:any} in similar situations.

Mixing such general table types with records is often a sign of something tricky going on. When such use of the table is read-only, of course things are safe, but often we're using {any:any} to do some smart manipulation of the data (e.g. copying any object generically, etc.) — it's easy to end up breaking some of the compiler's assumptions when doing that (e.g. setting a record field with a value of an invalid type). So I didn't want to make it too "natural" for such dynamic uses to pop up in the language. I think it's a good thing that whenever something unusual (in statically-typed terms) like that happens, the code has a cast somewhere that acts as a warning to the reader.

So, right now the alternative to do such general-table manipulation on record is really to do f(r as {any:any}).

In my experience, this has been a good thing: when I'm hunting my own bugs in Teal, the presence of as is often a good hint that I might have tripped something up with my tables.