Skip to content

Latest commit

 

History

History
504 lines (412 loc) · 17.6 KB

from-clojure.md

File metadata and controls

504 lines (412 loc) · 17.6 KB

Learning Fennel from Clojure

Fennel takes a lot of inspiration from Clojure. If you already know Clojure, then you'll have a good head start on Fennel. However, there are still a lot of differences! This document will guide you thru those differences and get you up to speed from the perspective of someone who already knows Clojure.

Fennel and Lua are minimalist languages, and Clojure is not. So it may take some getting used to when you make assumptions about what should be included in a language and find that it's not. There's almost always still a good way to do what you want; you just need to get used to looking somewhere different. With that said, Fennel is easier to learn since the conceptual surface area is much smaller.

Runtime

Clojure and Fennel are both languages which have very close integration with their host runtime. In the case of Clojure it's Java, and in the case of Fennel it's Lua. However, Fennel's symbiosis goes beyond that of Clojure. In Clojure, every function implements the interfaces needed to be callable from Java, but Clojure functions are distinct from Java methods. Clojure namespaces are related to Java packages, but namespaces still exist as a distinct concept from packages. In Fennel you don't have such distinctions. Every Fennel function is indistinguishable from a Lua function, and every Fennel module is indistinguishable from a Lua module.

Clojure runs on the JVM, but it also has its own standard library: the clojure.core namespace as well as supplemental ones like clojure.set or clojure.java.io provide more functions. In Fennel, there are no functions whatsoever provided by the language; it only provides macros and special forms. Since the Lua standard library is quite minimal, it's common to pull in 3rd-party things like Lume, LuaFun, or Penlight for things you might expect to be built-in to the language, like merge or keys. There's also an experimental Cljlib library, that implements a lot of functions from clojure.core namespace, and has a set of macros to make writing code more familiar to Clojure programmers, like adding syntax for defining multi-arity functions, or multimethods, also providing deep comparison semantics, sequence abstraction, and some addition data structures, like sets.

In Clojure it's typical to bring in libraries using a tool like Leiningen. In Fennel you can use LuaRocks for dependencies, but it's often overkill. Usually it's better to just check your dependencies into your source repository. Deep dependency trees are very rare in Fennel and Lua. Even tho Lua's standard library is very small, adding a single file for a 3rd-party library into your repo is very cheap. Checking a jar into a git repository in Clojure is strongly discouraged (for good reasons) but those reasons usually don't apply with Lua libraries.

Deploying Clojure usually means creating an uberjar that you launch using an existing JVM installation, because the JVM is a pretty large piece of software. Fennel deployments are much more varied; you can easily create self-contained standalone executables that are under a megabyte, or you can create scripts which rely on an existing Lua install, or code which gets embedded inside a larger application where the VM is already present.

Functions and locals

Clojure has two types of scoping: locals and vars. Fennel uses lexical scope for everything. (Globals exist, but they're mostly used for debugging and repl purposes; you don't use them in normal code.) This means that the "unit of reloading" is not the clojure.lang.Var, but the module. Fennel's repl includes a ,reload module-name command for this. Inside functions, let is used to introduce new locals just like in Clojure. But at the top-level, local is used, which declares a local which is valid for the entire remaining chunk instead of just for the body of the let.

Like Clojure, Fennel uses the fn form to create functions. However, giving it a name will also declare it as a local rather than having the name be purely internal, allowing it to be used more like defn. Functions declared with fn have no arity checking; you can call them with any number of arguments. If you declare with lambda instead, it will throw an exception when you provide too few arguments.

Fennel supports destructuring similarly to Clojure. The main difference is that rather than using :keys Fennel has a notation where a bare : is followed by a symbol naming the key. One main advantage of this notation is that unlike :keys, the same notation is used for constructing and destructuring.

;; clojure
(defn my-function [{:keys [msg abc def]}]
  (println msg)
  (+ abc def))

(my-function {:msg "hacasb" :abc 99 :def 523})
;; fennel
(fn my-function [{: msg : abc : def}]
  (print msg)
  (+ abc def))

(my-function {:msg "hacasb" :abc 99 :def 523})

Like Clojure, normal locals cannot be given new values. However, Fennel has a special var form that will allow you to declare a special kind of local which can be given a new value with set.

Fennel also uses #(foo) notation as shorthand for anonymous functions. There are two main differences; the first is that it uses $1, $2, etc instead of %1, %2 for arguments. Secondly while Clojure requires parens in this shorthand, Fennel does not. #5 in Fennel is the equivalent of Clojure's (constantly 5).

;; clojure
(def handler #(my-other-function %1 %3))
(def handler2 (constantly "abc"))
;; fennel
(local handler #(my-other-function $1 $3))
(local handler2 #"abc")

Fennel does not have apply; instead you unpack arguments into function call forms:

;; clojure
(apply add [1 2 3])
;; fennel
(add (table.unpack [1 2 3])) ; unpack instead of table.unpack in older Lua

In Clojure, you have access to scoping information at compile time using the undocumented &env map. In Fennel and Lua, environments are first-class at runtime.

Tables

Clojure ships with a rich selection of data structures for all kinds of situations. Lua (and thus Fennel) has exactly one data structure: the table. Under the hood, tables with sequential integer keys are of course implemented using arrays for performance reasons, but the table itself does not "know" whether it's a sequence table or a map-like table. It's up to you when you iterate thru the table to decide; you iterate on sequence tables using ipairs and map-like tables using pairs. Note that you can use pairs on sequences just fine; you just won't get the results in order.

The other big difference is that tables are mutable. It's possible to use metatables to implement immutable data structures on the Lua runtime, but there's a significant performance overhead beyond just the inherent immutability penalty. Using the LuaFun library can get you immutable operations on mutable tables without as much overhead. However, note that generational garbage collection is still a very recent development on the Lua runtime, so purely-functional approaches that generate a lot of garbage may not be a good choice for libraries which need to run on a wide range of versions.

Like Clojure, any value can serve as a key. However, since tables are mutable data, two tables with identical values will not be = to each other as per Baker and thus will act as distinct keys. Clojure's :keyword notation is used in Fennel as a syntax for strings; there is no distinct type for keywords.

Note that nil in Fennel is rather different from Clojure; in Clojure it has many different meanings, ("nil punning") but in Fennel it always represents the absence of a value. As such, tables cannot contain nil. Attempting to put nil in a table is equivalent to removing the value from the table, and you never have to worry about the difference between "the table does not contain this key" vs "the table contains a nil value at this key". And setting key to a nil in a sequential table will not shift all other elements, and will leave a "hole" in the table. Use table.remove instead on sequences to avoid these holes.

Tables cannot be called like functions, (unless you set up a special metatable) nor can :keyword style strings. If a string key is statically known, you can use tbl.key notation; if it's not, you use the . form in cases where you can't destructure: (. tbl key).

;; clojure
(dissoc my-map :abc)
(when-not (contains? my-other-map some-key)
  (println "no abc"))
;; fennel
(set my-map.abc nil)
(when (= nil (. my-other-map some-key))
  (print "no abc"))

Dynamic scope

As was mentioned previously, Clojure has two types of scoping: lexical and dynamic. Clojure vars can be declared in the dynamic scope with the special metadata attribute, supported by def and its derivatives, to be later altered with the binding macro:

;; clojure
(def ^:dynamic *foo* 32)
(defn bar [x]
  (println (+ x *foo*)))
(println (bar 10)) ;; => 42
(binding [*foo* 17]
  (println (bar 10))) ;; => 27
(println (bar 10)) ;; => 42

Fennel doesn't have dynamic scope. Instead we can use table mutability to alter values held, to be later dynamically looked up:

;; fennel
(local dynamic {:foo 32})
(fn bar [x]
  (print (+ dynamic.foo x)))
(print (bar 10)) ;; => 42
(set dynamic.foo 17)
(print (bar 10)) ;; => 27

In contrast to Clojure's binding, which only binds var to a given value in the scope created by the binding macro, the modification of the table here is permanent, and table values have to be restored manually.

In Clojure, similarly to variables, dynamic functions can be defined:

;; clojure
(defn ^:dynamic *fred* []
  "Hi, I'm Fred!")
(defn greet []
  (println (*fred*)))
(greet) ;; prints: Hi, I'm Fred!
(binding [*fred* (fn [] "I'm no longer Fred!")]
  (greet)) ;; prints: I'm no longer Fred!

In Fennel we can simply define a function as part of the table, either by assigning anonymous function to a table key, as done in the variable example above, or by separating function name and table name with a dot in the fn special:

;; fennel
(local dynamic {})
(fn dynamic.fred []
  "Hi, I'm Fred!")
(fn greet []
  (print (dynamic.fred)))
(greet) ;; prints: Hi, I'm Fred!
(set dynamic.fred (fn [] "I'm no longer Fred!"))
(greet) ;; prints: I'm no longer Fred!

Another alternative is to use the var special. We can define a variable holding nil, use it in some function, and later set it to some other value:

;; fennel
(var foo nil)
(fn bar []
  (foo))
(set foo #(print "foo!"))
(bar) ;; prints: foo!
(set foo #(print "baz!"))
(bar) ;; prints: baz!

This can also be used for forward declarations like Clojure's declare.

Iterators

In Clojure, we have this idea that "everything is a seq". Lua and Fennel, not being explicitly functional, have instead "everything is an iterator". The book Programming in Lua has a detailed explanation of iterators. The each special form consumes iterators and steps thru them similarly to how doseq does.

;; clojure
(doseq [[k v] {:key "value" :other-key "SHINY"}]
  (println k "is" v))
;; fennel
(each [k v (pairs {:key "value" :other-key "SHINY"})]
  (print k "is" v))

When iterating thru maps, Clojure has you pull apart the key/value pair thru destructuring, but in Fennel the iterators provide you with them as separate values.

Since Fennel has no functions, it relies on macros to do things like map and filter. Similarly to Clojure's for, Fennel has a pair of macros that operate on iterators and produce tables. icollect walks thru an iterator and allows the body to return a value that's put in a sequential table to return. The collect macro is similar in that it returns a table, except the body should return two values, and the returned table is key/value rather than sequential. The body of either macro allows you to return nil to filter out that entry from the result table.

;; clojure
(for [x [1 2 3 4 5 6]
      :when (= 0 (% x 2))]
  x) ; => (2 4 6)

(into {} (for [[k v] {:key "value" :other-key "SHINY"}]
           [k (str "prefix:" v)]))
; => {:key "prefix:value" :other-key "prefix:SHINY"}
;; fennel
(icollect [i x (ipairs [1 2 3 4 5 6])]
  (if (= 0 (% x 2)) x)) ; => [2 4 6]

(collect [k v (pairs {:key "value" :other-key "SHINY"})]
  (values k (.. "prefix:" v)))
; => {:key "prefix:value" :other-key "prefix:SHINY"}

Note that filtering values out using icollect does not result in a table with gaps in it; each value gets added to the end of the table.

All these forms accept iterators. Though the table-based pairs and ipairs are the most common iterators, other iterators like string.gmatch or io.lines or even custom ones work just as well.

Tables cannot be lazy (again other than thru metatable cleverness) so to some degree iterators take on the role of laziness.

If you want the sequence abstraction from Clojure, the Cljlib library provides Clojure's mapv, filter, and other functions that work using a similar seq abstraction implemented for ordinary tables with linear runtime cost of converting tables to a sequential ones. In practice, using Cljlib allows porting most Clojure data transformations almost directly to Fennel, though their performance characteristics will vary a lot.

Pattern Matching

Tragically Clojure does not have pattern matching as part of the language. Fennel fixes this problem by implementing the case macro. Refer to the reference for details. Since if-let just an anemic form of pattern matching, Fennel omits it in favor of case.

;; clojure
(if-let [result (calculate-thingy)]
  (println "Got" result)
  (println "Couldn't get any results"))
;; fennel
(case (calculate-thingy)
  result (print "Got" result)
  _ (print "Couldn't get any results"))

Modules

Modules in Fennel are first-class; that is, they are nothing more than tables with a specific mechanism for loading them. This is different from namespaces in Clojure which have some map-like properties but are not really data structures in the same way.

Clojure makes you replace the dashes in namespace names with underscores in filenames; Fennel lets you name the files consistently with the modules they contain.

In Clojure, vars are public by default. In Fennel, all definitions are local to the file, but including a local in a table that is placed at the end of the file will cause it to be exported so other code can use it. This makes it easy to look in one place to see a list of everything that a module exports rather than having to read thru the entire file.

;; clojure
(ns my.namespace)

(def ^:private x 13)
(defn add-x [y] (+ x y))
;; fennel
(local x 13)
(fn add-x [y] (+ x y))

{: add-x}

Modules are loaded by require and are typically bound using local, but they are also frequently destructured at the point of binding.

;; clojure
(require '[clojure.pprint :as pp])
(require '[my.namespace :refer [add-x]])

(defn show-something []
  (pp/pprint {:a 1 :b (add-x 13)}))
;; fennel
(local fennel (require :fennel))
(local {: add-x} (require :my.module))

(fn show-something []
  (print (fennel.view {:a 1 :b (add-x 13)})))

Macros

In any lisp, a macro is a function which takes an input form and returns another form to be compiled in its place. Fennel makes this even more explicit; macros are loaded as functions from special macro modules which are loaded in compile scope. They are brought in using import-macros:

;; macros.fnl

{:flip (fn [arg1 arg2] `(values ,arg2 ,arg1))}
;; otherfile.fnl
(import-macros {: flip} :macros)

(print (flip :abc :def))

Instead of using ~ for unquote, Fennel uses the more traditional ,. At the end of a quoted form you can use table.unpack or unpack in place of ~@.

You can also define macros inline without creating a separate macro module using macro, but these macros cannot be exported from the module as they do not exist at runtime; also they cannot interact with other macros.

Lists and symbols are strictly compile-time concepts in Fennel.

Errors

There are two kinds of ways to represent failure in Lua and Fennel. The error function works a bit like throwing an ex-info in Clojure, except instead of try and catch we have pcall and xpcall to call a function in "protected" state which will prevent errors from bringing down the process. These can't be chained or seamlessly re-thrown in the same way as Exceptions on the JVM are.

See the tutorial for details.

Other

There is no cond in Fennel because if behaves exactly the same as cond if given more than three arguments.

Functions can return multiple values. This can result in surprising behavior, but it's outside the scope of this document to describe. You can use the values form in a tail position to return multiple values.

Operators like + and or, etc are special forms which must have the number of arguments fixed at compile time. This means you cannot do things like (apply + [1 2 3]) or call (* ((fn [] (values 4 5 6)))), though the latter would work for functions rather than special forms.