**Toto je starší verze dokumentu!**

Task: Create a Hindley-Milner type-checker (60 points) and a simple interpreter (30 points) for a simple functional language. Write the Euclid's algorithm for searching the greatest common divisor in the language (10 points).

A term of the language is one of:

- A variable - any sequence of letters (except one of the keywords).
- A lambda application - two terms separated by a white space.
- A lambda abstraction - written as
`\variable . term`

- A let expression - written as
`let variable = term1 in term2`

, where the variable doesn't occur in term1. - A fix expression - written as
`fix variable . term`

, where the variable may occur in the term. Its semantic is given byfix x . M --> M[x := fix x . M]

- A parenthesised expression - written as
`(term)`

. Allows expressions like`(x z) (y z)`

. - Integer (decimal) constant.

You're encouraged to use the provided YACC-like parser description Parser.y so that you don't need to create your own. It is suited for the happy parser generator, however it should be easy to adapt for a different language. If you improve it or adapt it for another language, please send it to me and feel free to share it with the other students. (But of course, share only the parser description, nothing more!)

The type checker should compute the most general type for a given context and a program in the language.

The language contains a primitive (built-in) type `N`

for natural numbers, and several primitive functions:

`ifzero : Λα . N → α → α → α`

- if the first argument is zero, return the second argument, otherwise the third.`plus : N → N → N`

- addition.`minus : N → N → N`

- subtraction.`mul : N → N → N`

- multiplication.`div : N → N → N`

- integral division.

These functions with their types are always present in the default context when the type checker starts.

If a program is successfully type checked and its type is `N`

, the interpreter
is run on it and the result is printed. The interpreter doesn't have to be
efficient at all, but it must be correct. You must be sure that if the
interpreter is run on a program whose type is `N`

, the interpreter will not
crash and will return the correct integer result (or loop forever, if the
program doesn't terminate). You can also implement the interpreter by
converting programs into some standard suitable language and running the
result.

Example program:

(fix fact . \x . ifzero x 1 (mul x (fact (minus x 1)))) 10

should yield „3628800“ as the result.