The ugliness of the preceding columnar struct left us in a very unsatisfied state. Not only do we need to create a new data type with tons of Vector fields, we also have to create a constructor function to convert our array of structs into the new type.

We can recognize the power of Julia when we get to use powerful packages in its ecosystem. To fully implement this pattern, we will introduce the StructArrays.jl package, which automates most of the mundane tasks in turning an array of structs into a struct of arrays.

In fact, the usage of StructArrays is embarrassingly simple:

using StructArrays
sa = StructArray(records)

Let's take a quick look at the content. First of all, we can treat sa just like the original array—for example, we can take the first three elements of the array as before:

If we pick just one record, it comes back with the original TripPayment object:

Just to make sure that there is no mistake, we can also check the type of the first record:

Hence, the new sa object works just like before. Now, the difference comes in when we need to access all of the data from a single field. For example, we can get the fare_amount field as follows:

Because the type is already materialized as a dense array, we can expect superb performance when doing numerical or statistical analysis on this field, as follows: