type inference

[julochrobak]

The type inference for function arguments and result type would increase the code reuse of user defined Bandicoot functions dramatically.

Here is an example of a function which takes one relation parameter and returns a relation as well. The function requires the input argument to have a string attribute called "quality". It could work on any relations which match this criteria and therefor the following code reuse would be possible:

fn GoodProducts(b) {
    return select (quality == "good") b;
}

var books {title string, price real, quality string}
var cars {vendor string, name string, quality string}

fn GoodCars() {
    return GoodProducts cars;
}

fn GoodBooks() {
    return GoodProducts books;
}

This would make Bandicoot also very practical in scenarios where the data structure is not completely known, yet a calculation is needed or some information needs to be persisted and the rest of the data just flows through.

Here is an example of an HTTP service which captures the users and leaves the rest of the data untouched:

var users {user string};

fn CaptureUsers(in) {
    users += project (user) in;

    return in;
}

Any CSV tabular data could be piped through this function which contains the "user" column, for example capturing users for orders and processing them further:

cat orders.csv | \
    curl --data-binary @- http://localhost:12345/CaptureUsers | \
    ... some other processing of the orders.csv file ...