Why Hy?¶
Hy (or "Hylang" for long; named after the insect order Hymenoptera, since Paul Tagliamonte was studying swarm behavior when he created the language) is a multi-paradigm general-purpose programming language in the Lisp family. It's implemented as a kind of alternative syntax for Python. Compared to Python, Hy offers a variety of new features, generalizations, and syntactic simplifications, as would be expected of a Lisp. Compared to other Lisps, Hy provides direct access to Python's built-ins and third-party Python libraries, while allowing you to freely mix imperative, functional, and object-oriented styles of programming.
Hy versus Python¶
The first thing a Python programmer will notice about Hy is that it has Lisp's traditional parenthesis-heavy prefix syntax in place of Python's C-like infix syntax. For example,
print("The answer is", 2 + object.method(arg))
could be written
(print "The answer is" (+ 2 (.method object arg)))
in Hy. Consequently, Hy is free-form: structure is indicated by punctuation rather than whitespace, making it convenient for command-line use.
As in other Lisps, the value of a simplistic syntax is that it facilitates Lisp's signature feature: metaprogramming through macros, which are functions that manipulate code objects at compile time to produce new code objects, which are then executed as if they had been part of the original code. In fact, Hy allows arbitrary computation at compile-time. For example, here's a simple macro that implements a C-style do-while loop, which executes its body for as long as the condition is true, but at least once.
(defmacro do-while [condition #* body]
`(do
~@body
(while ~condition
~@body)))
(setv x 0)
(do-while x
(print "This line is executed once."))
Hy also removes Python's restrictions on mixing expressions and statements,
allowing for more direct and functional code. For example, Python doesn't allow
with
blocks, which close a resource once you're done using it,
to return values. They can only execute a set of statements:
with open("foo") as o:
f1 = o.read()
with open("bar") as o:
f2 = o.read()
print(len(f1) + len(f2))
In Hy, with
returns the value of its last body form, so you can use
it like an ordinary function call:
(print (+
(len (with [o (open "foo")] (.read o)))
(len (with [o (open "bar")] (.read o)))))
To be even more concise, you can put a with
form in a gfor
:
(print (sum (gfor
filename ["foo" "bar"]
(len (with [o (open filename)] (.read o))))))
Finally, Hy offers several generalizations to Python's binary operators.
Operators can be given more than two arguments (e.g., (+ 1 2 3)
), including
augmented assignment operators (e.g., (+= x 1 2 3)
). They are also provided
as ordinary first-class functions of the same name, allowing them to be passed
to higher-order functions: (sum xs)
could be written (reduce + xs)
,
after importing the function +
from the module hy.pyops
.
The Hy compiler works by reading Hy source code into Hy model objects and compiling them into Python abstract syntax tree (ast
)
objects. Python AST objects can then be compiled and run by Python itself,
byte-compiled for faster execution later, or rendered into Python source code.
You can mix Python and Hy code in the same project, or even the same
file, which can be a good way to get your feet wet in Hy.
Hy versus other Lisps¶
At run-time, Hy is essentially Python code. Thus, while Hy's design owes a lot to Clojure, it is more tightly coupled to Python than Clojure is to Java; a better analogy is CoffeeScript's relationship with JavaScript. Python's built-in functions and data structures are directly available:
(print (int "deadbeef" :base 16)) ; 3735928559
(print (len [1 10 100])) ; 3
The same goes for third-party Python libraries from PyPI and elsewhere. Here's a tiny CherryPy web application in Hy:
(import cherrypy)
(defclass HelloWorld []
(defn [cherrypy.expose] index [self]
"Hello World!"))
(cherrypy.quickstart (HelloWorld))
You can even run Hy on PyPy for a particularly speedy Lisp.
Like all Lisps, Hy is homoiconic. Its syntax is represented not
with cons cells or with Python's basic data structures, but with simple
subclasses of Python's basic data structures called models.
Using models in place of plain list
s, set
s, and so on has two
purposes: models can keep track of their line and column numbers for the
benefit of error messages, and models can represent syntactic features that the
corresponding primitive type can't, such as the order in which elements appear
in a set literal. However, models can be concatenated and indexed just like
plain lists, and you can return ordinary Python types from a macro or give them
to hy.eval
and Hy will automatically promote them to models.
Hy takes much of its semantics from Python. For example, functions use the same
namespace as objects that aren't functions, so a variable named globals
can shadow the Python built-in function globals()
. In
general, any Python code should be possible to literally translate to Hy. At
the same time, Hy goes to some lengths to allow you to do typical Lisp things
that aren't straightforward in Python. For example, Hy provides the
aforementioned mixing of statements and expressions, name mangling that transparently converts symbols with names like valid?
to
Python-legal identifiers, and a let
macro to provide block-level scoping
in place of Python's usual function-level scoping.
What Hy is not¶
Hy isn't minimal or elegant. Hy is big and ugly and proud of it; it's an unopinionated big-tent language that lets you do what you want. It has all of Python's least-motivated semantic features, plus more features, plus various kinds of syntactic sugar. (The syntax isn't as complex as Python's, but there are a lot of details beyond plain old S-expressions.) If you're interested in a more small-and-beautiful approach to Lisp, in the style of Scheme, check out Hissp, another Lisp embedded in Python that was created by a Hy developer.
Also, Hy isn't a reimplementation of an older Lisp. It is its own language. It looks kind of like Clojure and kind of like Common Lisp, but nontrivial programs that run in one of these langauges can't be expected to run on another unaltered.