Mercurial > repos > other > SevenLanguagesInSevenWeeks

Install Clojure with "curl -O https://download.clojure.org/install/linux-install-1.10.0.442.sh; chmod +x linux-install-1.10.0.442.sh; sudo ./linux-install-1.10.0.442.sh" (it's a downloader that downloads the real package)

Clojure has a repl (clj), which needs rlwrap ("sudo zypper in rlwrap"). This then downloads some MORE files from Maven. Don't you just LOVE the Java ecosystem?!

Clojure is a Lisp dialect. It feels a little like Prolog in that it's logic and brackets galore, but Prolog is about predicates and pattern matching whereas Lisp has more of an imperative language with odd syntax
but sensible things like "if" statements. It uses strong, dynamic typing.

Running commands at the repl and prints output *and* the return value (which can be "nil").

Commands are prefix, not infix:
  "Normal" languages: 1 + 1
  Clojure: (+ 1 1)

Division of integers creates a (normalised) "ratio" type to maintain accuracy:
  (/ 2 4) → 1/2
  (/ 2.0 4) → 0.5

Prefix does have its advantages, though - it's more like a function and can have variable numbers of parameters:
  (+ 2 2 2 2) → 8

And it can be used to check order:
  (< 1 2 3) → true
  (< 1 3 2 4) → false

Strings are double-quoted, but characters can be escaped with a backslash outside quotes to become a string. When done with "str" it makes a string:
  (str \h \e \l \l \o) → "hello"
This seems like it would rarely be useful.

Strings and characters are different (just like in Java)
  (= "a" \a) → false

Clojure can pass code as arguments, e.g.:
  (if true (println "It's true!"))
but that's quite normal for functional languages.

The "if" function is basically "if <condition> <true>[ <false>]".

Note: 0 and "" are true but nil is false, so only false and the null value are falsey.

Clojure has four main data structures - lists (for code), vectors (for data), sets and maps. These are types of sequence. Other things may also be sequences (strings, file system structures, etc).
Sequences can then be operated on with "first", "last", "rest" and "cons" (construct from head and tail) functions.

You can't build a list Python style with "(1 2 3)". You need either "(list 1 2 3)" or the odd syntax "'(1 2 3)".
There are also functions like "nth" (nth <list> <0-based index>)

Vectors use sequare brackets. They're like a list, but you can implicitly index them (i.e. you don't need "nth") by doing ([vector] pos) because they're treated as functions.

first, last and rest can be used for pattern matching.

Sets use hash followed by squiggly brackets (which is easily confused with a map). Sorted sets use "(sorted-set <values> …)".
Manipulating sets is less clean - (clojure.set/union <set1> <set2>) for unioning sets and (clojure.set/difference <set1> <set2>) for difference.
Sets are also functions and so you can check if an item is in a set with (#{set} <value>). Returns "nil" if not in the set, else returns the value.

Maps are defined with squiggly brackets and have implicit key-value positions, so you can write"{:chewie :wookie :lea :human}" for keys :chewie and :lea with values :wookie and :human.
You can also put in commas, though.


":text" appears to be a unique identifier ("keyword"), but the book doesn't mention this until three pages after it started using it. They're like Ruby symbols and Prolog/Erland atoms.
Symbols point to something else and keywords point to themselves (although it isn't made clear what that means).
Keywords are also functions, so (map key) and (key map) are equivalent.

Maps can be merged with "merge" or with "merge-with", which takes a function to calculate the merged value.

Variables are assigned to with "(def <variable-name> <value>)". Functions are defined with "(defn <func-name> [<params list>] (<body>))".
There's also an optional docstring after <func-name>. Docs can be read with "(doc <func-name>)".

Function arguments can be quite complex and use pattern matching, e.g. getting the second value in a vector with:
  (defn line-end [[_ second]] second)
which takes a single arg (the vector "[_ second]") and the body just returns "second". Clojure calls this "destructuring". It ignores any additional values (e.g. "[_ c]" matches "[1 2 3 4 5]" and puts "2" in c and ignores the trailing 3,4,5).

Where "def" is used for long-term variables (which are created in user-space - the return result is #'user/<def-name>), "let" is used for local variables, which is useful within function definitions, e.g.:
  (defn centre_space [board] (let [[_ [_ c]] board] c))
which is a function with one argument (board) where the function body uses "let" with a vector containing a pattern to match and a variable to match against, and then a function that uses that vector.

There are also anonymous functions that use "(fn [args] (body))" - like a defn, but without a name. Alternatively, "#(function-body-with-%-as-a-variable)" defines an anonymous "reader macro" function with % bound to each value.
  (map (fn [w] (* 2 (count w))) my_list)
is equivalent to:
  (map #(* 2 (count %)) my_list)

As well as the map function, there's "apply" and "filter".

Also, some helper functions exist, like "odd?". Terminating in a question mark seems to be the Clojure approach to predicates. Sometimes you end up with multiple question marks:

(every? number? [1 2 3 :four])
;false

not-every? and not-any? both have question marks as well. But not everything is question marked:

(some nil? [1 2 nil])
; true

This seems oddly inconsistent. The footnote explains that it's because "some returns the first value that is not nil or false", so "nil?" return false for 1 and 2 and then return true for nil and so some returns true.
With other "some" filters, like "(some odd? …)", it works because nil (if it doesn't find anything) is falsey and numbers (if it finds an odd one) are truthy.

It's not a predicate because "(some nil? [1 2]) returns nil, not false.

Functional languages do lazy tail recursion. Unless they're Clojure, because the JVM doesn't support it. Clojure does it with a  "loop" and a "recur" function. "loop" takes x and y with initial values and a function to call. See loop_recur.clj.

For loops take the form (for [val collection<, val2 collection2<, …>>] (body)), which is a bit like "for X in Collection", but doesn't look like it at first.
Multiple val/collection pairs give nested for loops (so every val with every val2).

But then for loops can take a ":when" test keyword (or :let or :while). Which is odd, because :xxx has only ever been a user atom, but this one already has meaning. And they can be mixed anywhere in the parameters and Clojure knows what to do.

Reduce is more familiar: (reduce func list)

(reduce + [1 2 3 4]) ; sums
(reduce * [1 2 3 4 5]) ; factorial

As well as sorting a list with (sort list) you can use a custom funcion with (sort-by function list) where "function" takes a single parameter and generates a key.

Infinite sequences can be built with (repeat obj), (cycle [list]) and (iterate func start_obj). (take count list) takes the number of items that you want from a potentially infinite sequence (which is lazily evaluated). There's also a (drop count list) function to skip items.
If you don't like the "inside function happens first" Lisp-like function order then the slightly bizarre "->>" function (left-to-right operator) applies a number of functions in order, e.g.:

(->> [:later :rinse : repeat] (cycle) (drop 2) (take 5))
; equivalent to (take 5 (drop 2 (cycle [:later :rinse : repeat])))

You can also (interpose obj list) to insert something between each item and (interleave list list) to interleave two lists.

The book seems to use "(take n (iterate inc m))" rather than "(range m n+1)" but doesn't say why. Range is lazy, as is iterate. The only benefit to iterate is if you're doing an 'infinite' list with a high start, as it starts high rather than spending time generating and dropping.

Clojure doesn't have types/classes and interfaces. It has `defrecord` (types) and `defprotocol` (interfaces or contracts). This wasn't explained particularly well in the book (too much shown without explanation, or with explanation a page later) so bearings.clj reworks it with more documentation (and some better implementations).

Clojure agressively executes code, so an "unless" function won't work because the function will always be execute *before* it checks whether the "unless" condition is false. This can apparently be fixed with macros, which are expanded before execution.
Three paragraphs later it explains(?) that you use a macro to turn code into lists and that macro substitution doesn't evaluate the contents. But you've got to quote your functions and build lists.
";" for comments, "'" for quotes and "#" for anonymous functions are pre-existing "reader macros".
"Data as code" (your data can be executed) is powerful, but it looks messy and confusing. And it's still not clear why "code" is all evaluated straight awaybut "code put in by macro in first parsing pass" is not.

(when boolean expression1 expression2 expression…) is a macro that effectively becomes (if boolean (do expression1…)) - https://www.braveclojure.com/writing-macros/ - but is actually constructed as (list 'if test (cons 'do body)).  "Clojure for the Brave and True" makes this a *little* clearer.

Clojure does concurrency without locking and without actors. It uses "software transactional memory" (STM). You create a (ref thing) to a thing and then pass it around (def myvar (ref "refd string")). myvar then holds the ref and (deref myvar) or @myvar gets the original value back (note: deref does not remove the reference, it just follows it to the value).
Changes to the <Ref> object need to be in a transaction, such as (dosync …). (alter myref func args) calls func with @myref and args. (ref-set myref new_value) swaps the value.

Thread safety without transactions comes from (atom thing_to_atom) rather than (ref …) but is used in a similar way. You can replace an atom's value using (reset! myatom value), but it is more Clojure-y (Clojonic?) to use (swap! myatom func extra_args). Note that "atom" doesn't have to mean "simple" - maps can be atoms, but atom gives you the ability to edit a variable with thread safety.

Agents are like atoms but with async updates. You (send agent func) to update their value. If the update is slow then @agent will return its 'old' value until the update is complete, but the update will be transactional.
The book says "If you think you want the latest value of something, you have already failed". Presumably that's where you should use atoms, although it doesn't tell you that. Applying functions to the value through (swap! …) means it implicitly gets the latest value at execution time.

From https://stackoverflow.com/a/9136699/283242:
    Refs are for Coordinated Synchronous access to "Many Identities".
    Atoms are for Uncoordinated synchronous access to a single Identity.
    Agents are for Uncoordinated asynchronous access to a single Identity.
    Vars are for thread local isolated identities with a shared default value.

Futures are the same as in other languages. (future <some sequence of expressions>) creates a future that returns the last value when dereferenced with @my_future and the app blocks (unlike actors, which need (await …) to block, and that's only until your changes are complete)

Clojure can also call all the standard Java methods on a Java object using (.javaMethodName object), e.g. (.toUpperCase "Using Clojure")

Day 3 exercise says to find a queue that blocks when it is empty. The best I can find is a wrapper on java.util.concurrent.LinkedBlockingDeque (https://gist.github.com/mjg123/1305115/72434dd5da89e5a83c91facb0b7687d6b7e66836). I've tried creating one but can't get the logic right.
It *feels* like an atom job, then an agent to do an async push, but then you're just adding an empty check and Thread/sleep to a loop. Or you use actors with (await …) rather than atoms, but await only waits for changes in your thread, which is of minimal use with a blocking queue.
author	IBBoard <dev@ibboard.co.uk>
date	Sun, 14 Jul 2019 13:44:13 +0100
parents	920b50be0fe5
children