BNF

<toplevel>    ::= <function>*
<function>    ::= <type> ' ' <identifier> '(' (<arg> (',' <arg>)*)? ')' <block>
<arg>         ::= <type> ' ' <identifier>
<block>       ::= '{' (<stmt> ';')* '}'
<stmt>        ::= <declr-stmt> | <assign-stmt> | <call-stmt>
<call-stmt>   ::= <identifier> '(' (<expr> (',' <expr>)*)? ')'
<declr-stmt>  ::= <type> ' ' <identifier> ('=' <expr>)?
<assign-stmt> ::= <identifier> '=' <expr>
<expr>        ::= <arith-expr>
<arith-expr>  ::= <arith-elm> (<arith-op> <arith-elm>)*
<arith-elm>   ::= <num> | <identifier> | '(' <expr> ')'
<arith-op>    ::= '+' | '-' | '*' | '/'
<num>         ::= 32-bit int
<identifier>  ::= a-z | A-Z
<type>        ::= 'int'


One data type and one type of expression isn't too complicated, but will keep figuring out codegen more straight forward.