Let

Let expressions allow to introduce local variables. Such expressions have the following syntax:

\let | p_1 => e_1
     ...
     | p_n => e_n
\in e_{n+1}

where p_1, … p_n are patterns and e_1, … e_{n+1} are expressions.

Every pattern is either a variable or an expression of the form (p_1’, … p_k’), where p_1’, … p_k’ are patterns. This implies that if p_i is a tuple of k subpatterns, then the type of e_i must be either a k-fold Sigma type or a record with k fields. Note that because of the eta-equivalence for Sigma types and records the structure of e_i does not matter: for example, expression \let (x,y) => z \in e evaluates to e[z.1/x,z.2/y] if type of z is Sigma type (and with fields instead of projections z.1 and z.2 in case of a record). In general, if pattern p_i contains variables x_i^1, … x_i^{n_i} expression \let | p_1 => e_1 … | p_n => e_n \in e evaluates to e[… proj_i^j(e_i)/x_i^j …], where proj_i^j is the sequence of projections and field access expressions, corresponding to j-th variable of p_i.

The expression \let | x_1 => e_1 … | x_n => e_n \in e has type \let | x_1 => e_1 … | x_n => e_n \in E, where E is the type of e.

The type of e_i can be explicitly specified as follows: | p_i : E_i => e_i.

It is also allowed to write lambda parameters after a pattern if it is a variable. That is, instead of | x_i => e_i, you can write | x_i p^i_1 … p^i_{n_i} => e_i, where p^i_1, … p^i_{n_i} are either variables or named parameters to which e_i can refer. Such a clause is equivalent to | x_i => \lam p^i_1 … p^i_{n_i} => e_i.

Let expressions also can be strict. This means that expressions e_1, … e_n will be evaluated immediately when the let expression is evaluated. To define a strict let expression, use the keyword \let! instead of \let.