A. Sundararajan's Weblog

SML subset for TEDI?

I came across "Do we need a new kind of schema language?" from Tim Bray's blog. After reading it, I can't resist this: why not use a subset of Standard ML for this purpose?

• Basic types - SML has int, real, bool, char, string
• Tuples - SML has t1*t2*t3... (for example, int*bool, int*string*real and so on)
• Lists - 'a list - int list, string list and so on
• Records - {name1:t1, name2:t2 } (for example, {name:string, age:int})
• type for type abbreviations - useful when defining "bigger" schema.
• datatype for algebraic data types
• For larger schemas, we could use parameterized data types.
• For larger schemas, we could also use SML structures to group related types.

In the above, we have every type expression except for function types! (IMO, it is not a big list!). Annotations for bindings may be specified in SML comments [this needs more thought].

 { url: String, width: Integer?, height: Integer?, title: String }
 

 { url: string, width: Maybe int, height: Maybe int, title: string }

Maybe is

  datatype 'a Maybe = Just of 'a | Nothing

Pros:

• We can have parametric types and module system - may be useful, for defining larger, generic schema.
• Tuples, Lists, Records could be mapped to parametrized classes (like in Scala, Java etc.) or native data types in scripting languages.
• sum-of-product types can be translated as classes (like case classes in Scala)
• Type expressions are proper set of one particular language - at least few people would feel at home JSON is proper subset of JavaScript object literals, functional folks could have their turn . BTW, I am okay with Haskell as well. Personally, I've played with SML little bit more
• If we want to have schema specified default values of various elements of data, we can include proper subset of value definitions in SML.

( Apr 10 2007, 07:40:43 PM IST ) Permalink | | | | |

---