Most of Clojure programming involves creating, manipulating, and transforming immutable values. However, as soon as your values become more complicated than a simple map or list – like a list of maps of maps – transforming these data structures becomes extremely cumbersome.
Specter is a library for querying and updating nested data structures. One way to think of it is "get-in" and "assoc-in" on steroids, though Specter works on any data structure, not just maps. It is similar to the concept of a "lens" in functional programming, though it has some important extensions.
Specter is fully extensible. At its core, its just a protocol for how to navigate within a data structure. By extending this protocol, you can use Specter to navigate any data structure or object you have.
Specter is a very high performance library. For example: the Specter equivalent to get-in runs 30% faster than get-in, and the Specter equivalent to update-in runs 5x faster than update-in. In each case the Specter code is equally as convenient.
The usage of Specter will be explained via example. Suppose you have a sequence of maps, and you want to extract all the even values for :a keys. Here's how you do it:
`select` extracts a sequence of results from a data structure. It takes in a "selector", which is a sequence of steps on how to navigate into that data structure. In this case, `ALL` looks at every element in the sequence, `:a` looks at the :a key for each element currently navigated to, and `even?` filters out any elements that aren't an even value.
Another function called `update` is used to perform a transformation on a data structure. In addition to a selector, it takes in an "update function" which specifies what to do with each element navigated to. For example, here's how to increment all the even values for :a keys in a sequence of maps:
`filterer` navigates you to a view of the sequence currently being looked at. `LAST` navigates you to the last element of whatever sequence you're looking at. But of course during updates, the updates are performed on the original data structure.
`srange` is a selector for looking at or replacing a subsequence of a sequence. For example, here's how to increment all the odd numbers between indexes 1 (inclusive) and 4 (exclusive):
`srange` can also be used to replace that subsequence entirely with a new sequence. For example, here's how to replace the subsequence from index 2 to 4 with [-1 -1 -1]:
When doing more involved transformations, you often find you lose context when navigating deep within a data structure and need information "up" the data structure to perform the transformation. Specter solves this problem by allowing you to collect values during navigation to use in the update function. Here's an example which transforms a sequence of maps by adding the value of the :b key to the value of the :a key, but only if the :a key is even:
The update function receives as arguments all the collected values followed by the navigated to value. So in this case `+` receives the value of the :b key followed by the value of the :a key, and the update is performed to :a's value.
The four built-in ways for collecting values are `VAL`, `collect`, `collect-one`, and `putval`. `VAL` just adds whatever element it's currently on to the value list, while `collect` and `collect-one` take in a selector to navigate to the desired value. `collect` works just like `select` by finding a sequence of values, while `collect-one` expects to only navigate to a single value. Finally, `putval` adds an external value into the collected values list.
As an example, here is how Clojure keywords implement this protocol:
```clojure
(extend-type clojure.lang.Keyword
StructurePath
(select* [kw structure next-fn]
(next-fn (get structure kw)))
(update* [kw structure next-fn]
(assoc structure kw (next-fn (get structure kw)))
))
```
`next-fn` represents the rest of the select or update, respectively. As you can see, this implementation perfectly captures what it means to navigate via a keyword within a data structure. In the select case, it completes the select by calling `next-fn` on the value of the keyword. In the update case, it updates the nested data structure using next-fn and then replaces the current value of the keyword with that updated data structure.
Finally, you can make `select` and `update` work much faster by precompiling your selectors using the `comp-paths` function. There's about a 3x speed difference between the following two invocations of update: