Skip to main content

Slicing

Slice syntax mirrors Python: [start:end] or [start:end:step].

Forms​

FormMeaning
[a:b]Rows a (inclusive) through b (exclusive)
[:b]First b rows
[a:]From row a to the end
[:]Every row (rare; same as no slice)
[a:b:s]Same as [a:b] but step s

Negative indices​

Negative numbers count from the end:

shop.customers[-10:]    // last 10
shop.customers[:-1]     // all but the last

Filters, slices, and aggregates combine in any order​

You can chain as many as you want, and the optimizer figures out the best execution plan. These are all equivalent:

shop.orders[status = "paid"]._desc(placed_at)[0:20]
shop.orders._desc(placed_at)[status = "paid"][0:20]
shop.orders[total > 100][status = "paid"]._desc(placed_at)[0:20]

Multiple filters act like and:

shop.orders[status = "paid"][total > 100]
// same as
shop.orders[status = "paid" and total > 100]

Use whichever form reads better. For supported patterns this becomes a single indexed range read in the storage layer.

Slice on a relation​

You can slice the related side of a projection too:

shop.customers{
  id,
  orders[0:5]{id, total}
}

Response:

[
  {
    "id": "c1",
    "orders": [
      { "id": "o1", "total": 99.00 },
      { "id": "o2", "total": 45.50 }
    ]
  }
]

Each customer gets their first 5 orders.

Out-of-range​

Slices that overshoot the data simply return what's there:

shop.customers[1000:2000]    // empty array on a 50-row table

This never throws.