Best Practice for globals in clojure, (refs vs alter-var-root)? - clojure

I've found myself using the following idiom lately in clojure code.
(def *some-global-var* (ref {}))
(defn get-global-var []
#*global-var*)
(defn update-global-var [val]
(dosync (ref-set *global-var* val)))
Most of the time this isn't even multi-threaded code that might need the transactional semantics that refs give you. It just feels like refs are for more than threaded code but basically for any global that requires immutability. Is there a better practice for this? I could try to refactor the code to just use binding or let but that can get particularly tricky for some applications.

I always use an atom rather than a ref when I see this kind of pattern - if you don't need transactions, just a shared mutable storage location, then atoms seem to be the way to go.
e.g. for a mutable map of key/value pairs I would use:
(def state (atom {}))
(defn get-state [key]
(#state key))
(defn update-state [key val]
(swap! state assoc key val))

Your functions have side effects. Calling them twice with the same inputs may give different return values depending on the current value of *some-global-var*. This makes things difficult to test and reason about, especially once you have more than one of these global vars floating around.
People calling your functions may not even know that your functions are depending on the value of the global var, without inspecting the source. What if they forget to initialize the global var? It's easy to forget. What if you have two sets of code both trying to use a library that relies on these global vars? They are probably going to step all over each other, unless you use binding. You also add overheads every time you access data from a ref.
If you write your code side-effect free, these problems go away. A function stands on its own. It's easy to test: pass it some inputs, inspect the outputs, they'll always be the same. It's easy to see what inputs a function depends on: they're all in the argument list. And now your code is thread-safe. And probably runs faster.
It's tricky to think about code this way if you're used to the "mutate a bunch of objects/memory" style of programming, but once you get the hang of it, it becomes relatively straightforward to organize your programs this way. Your code generally ends up as simple as or simpler than the global-mutation version of the same code.
Here's a highly contrived example:
(def *address-book* (ref {}))
(defn add [name addr]
(dosync (alter *address-book* assoc name addr)))
(defn report []
(doseq [[name addr] #*address-book*]
(println name ":" addr)))
(defn do-some-stuff []
(add "Brian" "123 Bovine University Blvd.")
(add "Roger" "456 Main St.")
(report))
Looking at do-some-stuff in isolation, what the heck is it doing? There are a lot of things happening implicitly. Down this path lies spaghetti. An arguably better version:
(defn make-address-book [] {})
(defn add [addr-book name addr]
(assoc addr-book name addr))
(defn report [addr-book]
(doseq [[name addr] addr-book]
(println name ":" addr)))
(defn do-some-stuff []
(let [addr-book (make-address-book)]
(-> addr-book
(add "Brian" "123 Bovine University Blvd.")
(add "Roger" "456 Main St.")
(report))))
Now it's clear what do-some-stuff is doing, even in isolation. You can have as many address books floating around as you want. Multiple threads could have their own. You can use this code from multiple namespaces safely. You can't forget to initialize the address book, because you pass it as an argument. You can test report easily: just pass the desired "mock" address book in and see what it prints. You don't have to care about any global state or anything but the function you're testing at the moment.
If you don't need to coordinate updates to a data structure from multiple threads, there's usually no need to use refs or global vars.

Related

In Clojure, why do you have to use parenthesis when def'ing a function and use cases of let

I'm starting to learn clojure and I've stumbled upon the following, when I found myself declaring a "sum" function (for learning purposes) I wrote the following code
(def sum (fn [& args] (apply + args)))
I have understood that I defined the symbol sum as containing that fn, but why do I have to enclose the Fn in parenthesis, isn't the compiler calling that function upon definition instead of when someone is actually invoking it? Maybe it's just my imperative brain talking.
Also, what are the use cases of let? Sometimes I stumble on code that use it and other code that don't, for example on the Clojure site there's an exercise to use the OpenStream function from the Java Interop, I wrote the following code:
(defn http-get
[url]
(let [url-obj (java.net.URL. url)]
(slurp (.openStream url-obj))))
(http-get "https://www.google.com")
whilst they wrote the following on the clojure site as an answer
(defn http-get [url]
(slurp
(.openStream
(java.net.URL. url))))
Again maybe it's just my imperative brain talking, the need of having a "variable" or an "object" to store something before using it, but I quite don't understand when I should use let or when I shouldn't.
To answer both of your questions:
1.
(def sum (fn [& args] (apply + args)))
Using def here is very unorthodox. When you define a function you usually want to use defn. But since you used def you should know that def binds a name to a value. fn's return value is a function. Effectively you bound the name sum to the function returned by applying (using parenthesis which are used for application) fn.
You could have used the more traditional (defn sum [& args] (apply + args))
2.
While using let sometimes makes sense for readability (separating steps outside their nested use) it is sometimes required when you want to do something once and use it multiple times. It binds the result to a name within a specified context.
We can look at the following example and see that without let it becomes harder to write (function is for demonstration purposes):
(let [db-results (query "select * from table")] ;; note: query is not a pure function
;; do stuff with db-results
(f db-results)
;; return db-results
db-results)))
This simply re-uses a return value (db-results) from a function that you usually only want to run once - in multiple locations. So let can be used for style like the example you've given, but its also very useful for value reuse within some context.
Both def and defn define a global symbol, sort of like a global variable in Java, etc. Also, (defn xxx ...) is a (very common) shortcut for (def xxx (fn ...)). So, both versions will work exactly the same way when you run the program. Since the defn version is shorter and more explicit, that is what you will do 99% of the time.
Typing (let [xxx ...] ...) defines a local symbol, which cannot be seen by code outside of the let form, just like a local variable (block-scope) in Java, etc.
Just like Java, it is optional when to have a local variable like url-obj. It will make no difference to the running program. You must answer the question, "Which version makes my code easier to read and understand?" This part is no different than Java.

Creating clojure atoms with a function

I want to 1) create a list of symbols with the function below; then 2) create atoms with these symbols/names so that the atoms can be modified from other functions. This is the function to generate symbols/names:
(defn genVars [ dist ]
(let [ nms (map str (range dist)) neigs (map #(apply str "neig" %) nms) ]
(doseq [ v neigs ]
(intern *ns* (symbol v) [ ] ))
))
If dist=3, then 3 symbols, neig0, ... neig2 are created each bound with an empty vector. If it is possible to functionally create atoms with these symbols so that they are accessible from other functions. Any help is much appreciated, even if there are other ways to accomplish this.
your function seems to be correct, just wrap the value in the intern call with atom call. Also I would rather use dotimes.
user>
(defn gen-atoms [amount prefix]
(dotimes [i amount]
(intern *ns* (symbol (str prefix i)) (atom []))))
#'user/gen-atoms
user> (gen-atoms 2 "x")
nil
user> x0
#atom[[] 0x30f1a7b]
user> x1
#atom[[] 0x2149efef]
The desire to generate names suggests you would be better served by a single map instead:
(def neighbours (atom (make-neighbours)))
Where the definition of make-neigbours might look something like this:
(defn make-neighbours []
(into {} (for [i (range 10)]
[(str "neig" i) {:age i}])))
Where the other namespace would look values up using something like:
(get-in #data/neighbours ["neig0" :age])
Idiomatic Clojure tends to avoid creating many named global vars, preferring instead to collocating state into one or a few vars governed by Clojure's concurrency primitives (atom/ref/agent). I encourage you to think about whether your problem can be solved with a single atom in this way instead of requiring defining multiple vars.
Having said that, if you really really need multiple atoms, consider storing them all in a single map var instead of creating many global vars. Personally, I have never encountered a situation where creating many atoms was better than a single big atom (so I would be interested to hear about situations where this would be important).
If you really really need many vars, be aware that defining vars inside a function is actually bad style (https://github.com/bbatsov/clojure-style-guide#dont-def-vars-inside-fns). With good reason too! The beauty of using functions and data comes from the purity of the functions. def inside a function is particularly nasty as it is not only a side-effect, but is an potentially execution flow altering side-effect.
Of course yes there is a way to achieve it, as another answer points out.
Where it comes to defining things that goes beyond def and defn, there is quite a lot of precedence to using macros. For example defroutes from compojure, defschema from Schema, deftest from clojure.test. Generally anything that is a convenience form for creating vars. You could use a macro solution to create defs for your atoms:
(defmacro defneighbours [n]
`(do
~#(for [sym (for [i (range n)]
(symbol (str "neig" i)))]
`(def ~sym (atom {}))))
In my opinion this is actually less offensive than a functional version, only because it is creating global defs. It is a little more obvious about creating global defs by using the regular def syntax. But I only bring it up as a strawman, because this is still bad.
The reason functions and data work best is because they compose.
There are tangible considerations that make a single atom governing state very convenient. You can iterate over all neighbors conveniently, you can add new ones dynamically. Also you can do things like concatenating neighbors with other neighbors etc. Basically there are lots of function/data abstractions that you lock yourself out of if you create many global vars.
This is the reason that macros are generally considered useful for syntactic tricks, but best avoided in favor of functions and data. And it has a real impact on the flexibility of your code. For example going back to compojure; the macro syntax is actually very limiting, and for that reason I prefer not to use defroutes at all.
In summary:
Don't make lots of global defs if you can avoid it.
Prefer 1 atom over many atoms where possible.
Don't def inside a function.
Macros are best avoided in favor of functions and data.
Regardless of these guidelines, it is always good to explore what is possible, and I can't know your circumstances, so above all I hope you overcome your immediate problem and find Clojure a pleasant language to use.

Unused args in watcher function

Here is a code example from Clojure Programming
(def history (atom ()))
(defn log->list
[dest-atom key source old new]
(when (not= old new)
(swap! dest-atom conj new)))
(def sarah (atom {:name "Sarah" :age 25}))
(add-watch sarah :record (partial log->list history))
(swap! sarah update-in [:age] inc)
(swap! sarah identity)
(pprint #history)
The log->list function has two args key and source that never gets used. What are they for here?
the key is the tag used to later remove it:
(remove-watch reference key)
If the key is then passed to the watching function then it has the ability to later remove it's self once it has been triggered. Though you can build that by agreeing on the key name in advance as well, though having this feature can make things more elegant.
From the docstring on add-watch:
Adds a watch function to an agent/atom/var/ref reference. The watch
fn must be a fn of 4 args: a key, the reference, its old-state, its
new-state...
Because we're adding a watch on the atom, log->list needs to be a 4 arity function. We will get passed the key, reference, old value, and new value. Even though we only use the old and new values, we need to have 4 args in the function so it has the right arity.
It is often idiomatic in Clojure to use an underscore in place of an unused argument in a function definition to show it is unused, and to reduce visual noise. However in this case, I think keeping the unused variables and naming them makes it clearer what's happening and makes it easier to modify this in the future if necessary. This is particularly important as we need to keep the arity of this function exactly the same for it to work with add-watch. In a book, I would lean towards this option.
In actual code it's borderline and I could go either way on underscore vs naming the args. What I chose would probably depend on the context and whatever struck the right balance of clarity and conciseness.

Clojure, redeclear variable causes warning "already refers to"

I began learning Clojure today and I ran into a problem that I would not solve with cleaver Googeling.
I have a simple script where I'd like to increase a counter when a condition is met. I've learned that variables are immutble in Clojure and the way around increasing this is to redeclear it, however this throws a warning.
(defn main[]
(def num 0)
(if [...]
(def num (+ num 1))
)
)
However this throws the following warning:
WARNING: num already refers to: #'clojure.core/num in namespace: user, being replaced by: #'user/num
There are two problems here:
One, you are shadowing a function in clojure.core. This gets a warning because it can lead to unexpected behavior. If you know you won't be using clojure.core/num, you can include the following in your namespace declaration:
(ns my.ns
(:refer-clojure :exclude [num])
....)
Next problem: def is not for creating local values. Using def as anything other than a top level form is almost always a mistake, and any exceptions to this should require very explicit justification. It can only create global mutable vars. Use let for bindings that are specific to one scope, like inside a function.
(defn -main
[& args]
(let [num 0
num (if ... num (inc num))]
...))
Here, num is not mutated, and not created as a global var, but it is a local binding that is shadowed by the second binding.
Short Version:
In Clojure there are special abstractions to represent something that changes over time. The simplest one is called an atom. You start the atom with a value 0, and then you change the atom by applying the function inc to its value.
(def n (atom 0))
(def -main
[& [args]]
(if condition
(swap! n inc))
...)
Long Version:
If I understand correctly, you are asking for a way around immutability. This means that you are modelling some concept (i.e. site visitors) that takes different values (i.e 42) over time. Clojure is very opinionated on how to do this and the techniques it offers (STM) are central to the language. The reason Clojure works differently is to make concurrency easier but you do not need to think about to concurrency in order to increment a counter. To understand the rationale behind Clojure I recommend Rich Hickey's talks, in this case The Value of Values.

How do I get core clojure functions to work with my defrecords

I have a defrecord called a bag. It behaves like a list of item to count. This is sometimes called a frequency or a census. I want to be able to do the following
(def b (bag/create [:k 1 :k2 3])
(keys bag)
=> (:k :k1)
I tried the following:
(defrecord MapBag [state]
Bag
(put-n [self item n]
(let [new-n (+ n (count self item))]
(MapBag. (assoc state item new-n))))
;... some stuff
java.util.Map
(getKeys [self] (keys state)) ;TODO TEST
Object
(toString [self]
(str ("Bag: " (:state self)))))
When I try to require it in a repl I get:
java.lang.ClassFormatError: Duplicate interface name in class file compile__stub/techne/bag/MapBag (bag.clj:12)
What is going on? How do I get a keys function on my bag? Also am I going about this the correct way by assuming clojure's keys function eventually calls getKeys on the map that is its argument?
Defrecord automatically makes sure that any record it defines participates in the ipersistentmap interface. So you can call keys on it without doing anything.
So you can define a record, and instantiate and call keys like this:
user> (defrecord rec [k1 k2])
user.rec
user> (def a-rec (rec. 1 2))
#'user/a-rec
user> (keys a-rec)
(:k1 :k2)
Your error message indicates that one of your declarations is duplicating an interface that defrecord gives you for free. I think it might actually be both.
Is there some reason why you cant just use a plain vanilla map for your purposes? With clojure, you often want to use plain vanilla data structures when you can.
Edit: if for whatever reason you don't want the ipersistentmap included, look into deftype.
Rob's answer is of course correct; I'm posting this one in response to the OP's comment on it -- perhaps it might be helpful in implementing the required functionality with deftype.
I have once written an implementation of a "default map" for Clojure, which acts just like a regular map except it returns a fixed default value when asked about a key not present inside it. The code is in this Gist.
I'm not sure if it will suit your use case directly, although you can use it to do things like
user> (:earth (assoc (DefaultMap. 0 {}) :earth 8000000000))
8000000000
user> (:mars (assoc (DefaultMap. 0 {}) :earth 8000000000))
0
More importantly, it should give you an idea of what's involved in writing this sort of thing with deftype.
Then again, it's based on clojure.core/emit-defrecord, so you might look at that part of Clojure's sources instead... It's doing a lot of things which you won't have to (because it's a function for preparing macro expansions -- there's lots of syntax-quoting and the like inside it which you have to strip away from it to use the code directly), but it is certainly the highest quality source of information possible. Here's a direct link to that point in the source for the 1.2.0 release of Clojure.
Update:
One more thing I realised might be important. If you rely on a special map-like type for implementing this sort of thing, the client might merge it into a regular map and lose the "defaulting" functionality (and indeed any other special functionality) in the process. As long as the "map-likeness" illusion maintained by your type is complete enough for it to be used as a regular map, passed to Clojure's standard function etc., I think there might not be a way around that.
So, at some level the client will probably have to know that there's some "magic" involved; if they get correct answers to queries like (:mars {...}) (with no :mars in the {...}), they'll have to remember not to merge this into a regular map (merge-ing the other way around would work fine).