Re: From dynamic to static

[David Given <dg@...>]

On Monday 31 October 2005 18:24, Chris Marrin wrote:
[...]
> But this cannot be known at compile time in Lua (or any other
> dynamically typed language that I know of). At compile time o may not
> have a method1 function. But it hopefully will by the time it gets called.

Hence why I said you'd have to use a subset of the language. You'd have to 
tell the type analyser that method1 was defined on *some* class. (In fact, if 
you didn't, the type analyser should fail, because no object type would be 
compatible with it.)

The type analyser would have to know (a) what a constructor looks like; (b) 
what a method declaration on a particular class looks like; and (c) how to 
find out what the type of a value stored in a table is (statically). It might 
be necessary to decree that if x.foo is of one particular type, then y.foo 
must be that same type, unless you were willing to find a way of declaring 
typed structures. (Of course, class instances are already typed, so you could 
manage it in the case of instance variables.)

You'd also need a limitation on variable assignments to one particular type. 
So:

local foo = 7
foo = "bar"

...would fail a type constraint. However, this also means that:

local foo = 7
foo = bar()

...would mean that, for it to be valid, bar() must return a number.

In the interests of sanity, it would probably be necessary to disallow 
table[key] syntax, or else decree that it bypasses the type safety --- I 
don't know how you'd manage to work that work sensibly otherwise. However, if 
you were going to be using this strongly-typed Lua, you'd probably want to 
use container classes for data storage anyway, and wouldn't need it... this 
would also allow such nifty features as a polymorphic Array class that leaves 
the type of its members undefined, so that each instance can store a 
different strongly-typed value...

You know, this actually sounds *possible*. Good grief.

-- 
+- David Given --McQ-+ 
|  dg@cowlark.com    | "Never attribute to malice what can be adequately
| (dg@tao-group.com) | explained by stupidity." --- Nick Diamos
+- www.cowlark.com --+

Attachment: pgpaPbcgLqhRU.pgp
Description: PGP signature