API reference

This chapter describes most of Hy's public-facing macros, functions, and classes. It refers to Python's own documentation when appropriate rather than recapitulating the details of Python semantics.

Core macros

The following macros are automatically imported into all Hy modules as their base names, such that hy.core.macros.foo can be called as just foo. Macros that are also available as functions are described as functions under Python operators.

Fundamentals

macro(do #* body)

do (called progn in some Lisps) takes any number of forms, evaluates them, and returns the value of the last one, or None if no forms were provided.

(+ 1 (do (setv x (+ 1 1)) x))  ; => 3
macro(do-mac #* body)

do-mac evaluates its arguments (in order) at compile time, and leaves behind the value of the last argument (None if no arguments were provided) as code to be run. The effect is similar to defining and then immediately calling a nullary macro, hence the name, which stands for "do macro".

(do-mac `(setv ~(hy.models.Symbol (* "x" 5)) "foo"))
  ; Expands to:   (setv xxxxx "foo")
(print xxxxx)
  ; => "foo"

Contrast with eval-and-compile, which evaluates the same code at compile-time and run-time, instead of using the result of the compile-time run as code for run-time. do-mac is also similar to Common Lisp's SHARPSIGN DOT syntax (#.), from which it differs by evaluating at compile-time rather than read-time.

macro(eval-and-compile #* body)

eval-and-compile takes any number of forms as arguments. The input forms are evaluated as soon as the eval-and-compile form is compiled, then left in the program so they can be executed at run-time as usual; contrast with eval-when-compile. So, if you compile and immediately execute a program (as calling hy foo.hy does when foo.hy doesn't have an up-to-date byte-compiled version), eval-and-compile forms will be evaluated twice. For example, the following program

(eval-when-compile
  (print "Compiling"))
(print "Running")
(eval-and-compile
  (print "Hi"))

prints

Compiling
Hi
Running
Hi

The return value of eval-and-compile is its final argument, as for do.

One possible use of eval-and-compile is to make a function available both at compile-time (so a macro can call it while expanding) and run-time (so it can be called like any other function):

(eval-and-compile
  (defn add [x y]
    (+ x y)))

(defmacro m [x]
  (add x 2))

(print (m 3))     ; prints 5
(print (add 3 6)) ; prints 9

Had the defn not been wrapped in eval-and-compile, m wouldn't be able to call add, because when the compiler was expanding (m 3), add wouldn't exist yet.

While eval-and-compile executes the same code at both compile-time and run-time, bear in mind that the same code can have different meanings in the two contexts. Consider, for example, issues of scoping:

(eval-when-compile
  (print "Compiling"))
(print "Running")
(eval-and-compile
  (setv x 1))
(defn f []
  (setv x 2)
  (eval-and-compile
    (setv x 3))
  (print "local x =" x))
(f)
(eval-and-compile
  (print "global x =" x))

The form (setv x 3) above refers to the global x at compile-time, but the local x at run-time, so the result is:

Compiling
global x = 3
Running
local x = 3
global x = 1
macro(eval-when-compile #* body)

eval-when-compile executes the given forms at compile-time, but discards them at run-time and simply returns None instead; contrast eval-and-compile. Hence, while eval-when-compile doesn't directly contribute code to the final program, it can change Hy's state while compiling, as by defining a function:

(eval-when-compile
  (defn add [x y]
    (+ x y)))

(defmacro m [x]
  (add x 2))

(print (m 3))     ; prints 5
(print (add 3 6)) ; raises NameError: name 'add' is not defined
macro(py string)

py parses the given Python code at compile-time and inserts the result into the generated abstract syntax tree. Thus, you can mix Python code into a Hy program. Only a Python expression is allowed, not statements; use pys if you want to use Python statements. The value of the expression is returned from the py form.

(print "A result from Python:" (py "'hello' + 'world'"))

The code must be given as a single string literal, but you can still use macros, hy.eval, and related tools to construct the py form. If having to backslash-escape internal double quotes is getting you down, try a bracket string. If you want to evaluate some Python code that's only defined at run-time, try the standard Python function eval().

The code is implicitly wrapped in parentheses so Python won't give you grief about indentation. After all, Python's indentation rules are only useful for grouping statements, whereas py only allows an expression.

Python code need not syntactically round-trip if you use hy2py on a Hy program that uses py or pys. For example, comments will be removed.

macro(pys string)

As py, but the code can consist of zero or more statements, including compound statements such as for and def. pys always returns None.

(pys "myvar = 5")
(print "myvar is" myvar)

Unlike py, no parentheses are added, because Python doesn't allow statements to be parenthesized. Instead, the code string is dedented with textwrap.dedent() before parsing. Thus you can indent the code to match the surrounding Hy code when Python would otherwise forbid this, but beware that significant leading whitespace in embedded string literals will be removed.

macro(pragma #* args)

pragma is used to adjust the state of the compiler. It's called for its side-effects, and returns None. The arguments are key-value pairs, like a function call with keyword arguments:

(pragma :prag1 value1 :prag2 (get-value2))

Each key is a literal keyword giving the name of a pragma. Each value is an arbitrary form, which is evaluated as ordinary Hy code but at compile-time.

The effect of each pragma is locally scoped to its containing function, class, or comprehension form (other than for), if there is one.

Only one pragma is currently implemented:

  • :warn-on-core-shadow: If true (the default), defmacro and require will raise a warning at compile-time if you define a macro with the same name as a core macro. Shadowing a core macro in this fashion is dangerous, because other macros may call your new macro when they meant to refer to the core macro.

Quoting

macro(quote model)

Return the given model without evaluating it. Or to be more pedantic, quote complies to code that produces and returns the model it was originally called on. Thus quote serves as syntactic sugar for model constructors:

(quote a)
  ; Equivalent to:  (hy.models.Symbol "a")
(quote (+ 1 1))
  ; Equivalent to:  (hy.models.Expression [
  ;   (hy.models.Symbol "+")
  ;   (hy.models.Integer 1)
  ;   (hy.models.Integer 1)])

quote itself is conveniently abbreviated as the single-quote character ', which needs no parentheses, allowing one to instead write:

'a
'(+ 1 1)

See also:

  • quasiquote to substitute values into a quoted form

  • hy.eval to evaluate models as code

  • hy.repr to stringify models into Hy source text that uses '

macro(quasiquote model)
macro(unquote model)
macro(unquote-splice model)

quasiquote is like quote except that it treats the model as a template, in which certain special expressions indicate that some code should be evaluated and its value substituted there. The idea is similar to C's sprintf or Python's various string-formatting constructs. For example:

(setv x 2)
(quasiquote (+ 1 (unquote x)))  ; => '(+ 1 2)

unquote indicates code to be evaluated, so x becomes 2 and the 2 gets inserted in the parent model. quasiquote can be abbreviated as a backtick (`), with no parentheses, and likewise unquote can be abbreviated as a tilde (~), so one can instead write simply

`(+ 1 ~x)

(In the bulk of Lisp tradition, unquotation is written ,. Hy goes with Clojure's choice of ~, which has the advantage of being more visible in most programming fonts.)

Quasiquotation is convenient for writing macros:

(defmacro set-foo [value]
  `(setv foo ~value))
(set-foo (+ 1 2 3))
(print foo)  ; => 6

Another kind of unquotation operator, unquote-splice, abbreviated ~@, is analogous to unpack-iterable in that it splices an iterable object into the sequence of the parent sequential model. Compare the effects of unquote to unquote-splice:

(setv X [1 2 3])
(hy.repr `[a b ~X c d ~@X e f])
  ; => '[a b [1 2 3] c d 1 2 3 e f]

If unquote-splice is given any sort of false value (such as None), it's treated as an empty list. To be precise, ~@x splices in the result of (or x []).

Note that while a symbol name can begin with @ in Hy, ~@ takes precedence in the parser, so if you want to unquote the symbol @foo with ~, you must use whitespace to separate ~ and @, as in ~ @foo.

Assignment, mutation, and annotation

macro(setv #* args)

setv compiles to an assignment statement (see setx for assignment expressions), which sets the value of a variable or some other assignable expression. It requires an even number of arguments, and always returns None. The most common case is two arguments, where the first is a symbol:

(setv websites 103)
(print websites)  ; => 103

Additional pairs of arguments are equivalent to several two-argument setv calls, in the given order. Thus, the semantics are like Common Lisp's setf rather than psetf.

(setv  x 1  y x  x 2)
(print x y)  ; => 2 1

All the same kinds of complex assignment targets are allowed as in Python. So, you can use list assignment to assign in parallel. (As in Python, tuple and list syntax are equivalent for this purpose; Hy differs from Python merely in that its list syntax is shorter than its tuple syntax.)

(setv [x y] [y x])  ; Swaps the values of `x` and `y`

Unpacking assignment looks like this (see unpack-iterable):

(setv [letter1 letter2 #* others] "abcdefg")
(print letter1 letter2 (hy.repr others))
  ; => a b ["c" "d" "e" "f" "g"]

See let to simulate more traditionally Lispy block-level scoping.

macro(setx target value)

setx compiles to an assignment expression (PEP 572). Thus, unlike setv, it returns the assigned value. It takes exactly two arguments, and the target must be a bare symbol.

(when (> (setx x (+ 1 2)) 0)
  (print x "is greater than 0"))
    ; => 3 is greater than 0
macro(let bindings #* body)

let is a macro for simulating traditional block scoping as seen in other Lisps. Since it coexists with ordinary Python scoping, its consequences can be complex, so it's wise to get a solid understanding of Python scoping before you use it. Beginners to Python should note particularly that setv inside a function or class typically creates a local variable, so let isn't required for local variables or closures as it is in many other Lisps.

That disclaimer aside, let creates local variables with lexically scoped names. The macro takes a list of binding pairs followed by a body which gets executed. A let-bound name ceases to refer to that local outside the let form, but arguments in nested functions, and bindings in nested let forms, can shadow these names.

(let [x 5  y 6]   ; Create `x` and `y`
  (print x y)     ; => 5 6
  (let [x 7]      ; Create a variable that shadows the earlier `x`
    (print x y))  ; => 7 6
  (print x y))    ; => 5 6

The left-hand item of a binding pair is typically a plain symbol, but it can also use extended iterable unpacking (PEP 3132):

(let [[head #* tail] #(0 1 2)]
  [head tail])  ; => [0 [1 2]]

Basic assignments, as with setv or +=, will update the local variable named by a let binding when they assign to a let-bound name. But assignments via import are always hoisted to normal Python scope, and likewise, defn or defclass will assign the function or class in the Python scope, even if it shares the name of a let binding. To avoid this hoisting, use importlib.import_module(), fn, or type (or whatever metaclass) instead.

If lfor, sfor, dfor, or gfor (but not for) is in the body of a let, assignments in iteration clauses and :setv clauses will create a new variable in the comprehenion form's own scope, without touching any outer let-bound variable of the same name.

Like the let* of many other Lisps, let executes the variable assignments one-by-one, in the order written:

(let [x 5
      y (+ x 1)]
   (print x y)) ; => 5 6

(let [x 1
      x (fn [] x)]
   (x)) ; => 1

Note that let-bound variables continue to exist in the surrounding Python scope. As such, let-bound objects may not be eligible for garbage collection as soon as the let ends. To ensure there are no references to let-bound objects as soon as possible, use del at the end of the let, or wrap the let in a function.

macro(global #* syms)

global compiles to a global statement, which declares one or more names as referring to global (i.e., module-level) variables. The arguments are symbols; with no arguments, global has no effect. The return value is always None.

(setv  a 1  b 10)
(print a b)  ; => 1 10
(defn f []
  (global a)
  (setv  a 2  b 20))
(f)
(print a b)  ; => 2 10
macro(nonlocal #* syms)

Similar to global, but names can be declared in any enclosing scope. nonlocal compiles to a global statement for any names originally defined in the global scope, and a nonlocal statement for all other names.

(setv  a 1  b 1)
(defn f []
  (setv  c 10  d 10)
  (defn g []
    (nonlocal a c)
    (setv  a 2  b 2
           c 20 d 20))
  (print a b c d)  ; => 1 1 10 10
  (g)
  (print a b c d)) ; => 2 1 20 10
(f)
macro(del #* args)

del compiles to a del statement, which deletes variables or other assignable expressions. It always returns None.

(del  foo  (get mydict "mykey")  myobj.myattr)
macro(annotate value type)

annotate and its shorthand form #^ are used to denote annotations, including type hints, in three different contexts:

  • Standalone variable annotations (PEP 526)

  • Variable annotations in a setv call

  • Function-parameter annotations (PEP 3107)

The difference between annotate and #^ is that annotate requires parentheses and takes the name to be annotated first (like Python), whereas #^ doesn't require parentheses (it only applies to the next two forms) and takes the name second:

(setv (annotate x int) 1)
(setv #^ int x 1)

The order difference is not merely visual: #^ actually evaluates the type first.

Here are examples with #^ for all the places you can use annotations:

; Annotate the variable `x` as an `int` (equivalent to `x: int`).
#^ int x
; You can annotate with expressions (equivalent to `y: f(x)`).
#^(f x) y

; Annotations with an assignment: each annotation `(int, str)`
; covers the term that immediately follows.
; Equivalent to `x: int = 1; y = 2; z: str = 3`
(setv  #^ int x 1  y 2  #^ str z 3)

; Annotate `a` as an `int`, `c` as an `int`, and `b` as a `str`.
; Equivalent to `def func(a: int, b: str = None, c: int = 1): ...`
(defn func [#^ int a  #^ str  [b None] #^ int  [c 1]] ...)

; Function return annotations come before the function name (if
; it exists).
(defn #^ int add1 [#^ int x] (+ x 1))
(fn #^ int [#^ int y] (+ y 2))

For annotating items with generic types, the of macro will likely be of use.

An issue with type annotations is that, as of this writing, we know of no Python type-checker that can work with ast objects or bytecode files. They all need Python source text. So you'll have to translate your Hy with hy2py in order to actually check the types.

macro(deftype args)

deftype compiles to a type statement, which defines a type alias. It requires Python 3.12. Its arguments optionally begin with :tp and a list of type parameters (as in defn), then specify the name for the new alias and its value.

(deftype IntOrStr (| int str))
(deftype :tp [T] ListOrSet (| (get list T) (get set T)))

Subsetting

.

The dot macro . compiles to one or more attribute references, which select an attribute of an object. The first argument, which is required, can be an arbitrary form. With no further arguments, . is a no-op. Additional symbol arguments are understood as a chain of attributes, so (. foo bar) compiles to foo.bar, and (. a b c d) compiles to a.b.c.d.

As a convenience, . supports two other kinds of arguments in place of a plain attribute. A parenthesized expression is understood as a method call: (. foo (bar a b)) compiles to foo.bar(a, b). A bracketed form is understood as a subscript: (. foo ["bar"]) compiles to foo["bar"]. All these options can be mixed and matched in a single . call, so

(. a (b 1 2) c [d] [(e 3 4)])

compiles to

a.b(1, 2).c[d][e(3, 4)]

Dotted identifiers provide syntactic sugar for common uses of this macro. In particular, syntax like foo.bar ends up meaning the same thing in Hy as in Python. Also, get is another way to subscript in Hy.

macro(unpack-iterable form)
macro(unpack-mapping form)

(Also known as the splat operator, star operator, argument expansion, argument explosion, argument gathering, and varargs, among others...)

unpack-iterable and unpack-mapping allow an iterable or mapping object (respectively) to provide positional or keywords arguments (respectively) to a function.

=> (defn f [a b c d] [a b c d])
=> (f (unpack-iterable [1 2]) (unpack-mapping {"c" 3 "d" 4}))
[1 2 3 4]

unpack-iterable is usually written with the shorthand #*, and unpack-mapping with #**.

=> (f #* [1 2] #** {"c" 3 "d" 4})
[1 2 3 4]

Unpacking is allowed in a variety of contexts, and you can unpack more than once in one expression (PEP 3132, PEP 448).

=> (setv [a #* b c] [1 2 3 4 5])
=> [a b c]
[1 [2 3 4] 5]
=> [#* [1 2] #* [3 4]]
[1 2 3 4]
=> {#** {1 2} #** {3 4}}
{1 2  3 4}
=> (f #* [1] #* [2] #** {"c" 3} #** {"d" 4})
[1 2  3 4]

Conditionals and basic loops

macro(if test true-value false-value)

if compiles to an if expression (or compound if statement). The form test is evaluated and categorized as true or false according to bool. If the result is true, true-value is evaluated and returned. Othewise, false-value is evaluated and returned.

(if (has-money-left account)
  (print "Let's go shopping!")
  (print "Back to work."))

See also:

  • do, to execute several forms as part of any of if's three arguments.

  • when, for shorthand for (if condition (do …) None).

  • cond, for shorthand for nested if forms.

macro(hy.core.macros.when test #* body)

Shorthand for (if test (do …) None). See if. For a logically negated version, see Hyrule's unless.

(when panic
  (log.write panic)
  (print "Process returned:" panic.msg)
  (return panic))
macro(hy.core.macros.cond #* args)

Shorthand for a nested sequence of if forms, like an if-elif-else ladder in Python. Syntax such as

(cond
  condition1 result1
  condition2 result2)

is equivalent to

(if condition1
  result1
  (if condition2
    result2
    None))

Notice that None is returned when no conditions match; use True as the final condition to change the fallback result. Use do to execute several forms as part of a single condition or result.

With no arguments, cond returns None. With an odd number of arguments, cond raises an error.

macro(while condition #* body)

while compiles to a while statement, which executes some code as long as a condition is met. The first argument to while is the condition, and any remaining forms constitute the body. It always returns None.

(while True
  (print "Hello world!"))

The last form of a while loop can be an else clause, which is executed after the loop terminates, unless it exited abnormally (e.g., with break). So,

(setv x 2)
(while x
   (print "In body")
   (-= x 1)
   (else
     (print "In else")))

prints

In body
In body
In else

If you put a break or continue form in the condition of a while loop, it will apply to the very same loop rather than an outer loop, even if execution is yet to ever reach the loop body. (Hy compiles a while loop with statements in its condition by rewriting it so that the condition is actually in the body.) So,

(for [x [1]]
   (print "In outer loop")
   (while
     (do
       (print "In condition")
       (break)
       (print "This won't print.")
       True)
     (print "This won't print, either."))
   (print "At end of outer loop"))

prints

In outer loop
In condition
At end of outer loop
macro(break)

break compiles to a break statement, which terminates the enclosing loop. The following example has an infinite while loop that ends when the user enters "k":

(while True
  (if (= (input "> ") "k")
    (break)
    (print "Try again")))

In a loop with multiple iteration clauses, such as (for [x xs y ys] …), break only breaks out of the innermost iteration, not the whole form. To jump out of the whole form, enclose it in a block and use block-ret instead of break. In the case of for, but not lfor and the other comprehension forms, you may also enclose it in a function and use return.

macro(continue)

continue compiles to a continue statement, which returns execution to the start of a loop. In the following example, (.append output x) is executed on each iteration, whereas (.append evens x) is only executed for even numbers.

(setv  output []  evens [])
(for [x (range 10)]
  (.append output x)
  (when (% x 2)
    (continue))
  (.append evens x))

In a loop with multiple iteration clauses, such as (for [x xs y ys] …), continue applies to the innermost iteration, not the whole form. To jump to the next step of an outer iteration, try rewriting your loop as multiple nested loops and interposing a block, as in (for [x xs] (block (for [y ys] …))). You can then use block-ret in place of continue.

Comprehensions

macro(for #* args)

for compiles to one or more for statements, which execute code repeatedly for each element of an iterable object. The return values of the forms are discarded and the for form returns None.

(for [x [1 2 3]]
  (print "iterating")
  (print x))
; Output: iterating 1 iterating 2 iterating 3

The first argument of for, in square brackets, specifies how to loop. A simple and common case is [variable values], where values is an iterable object (such as a list) and variable is a symbol specifiying the name for each element. Subsequent arguments to for are body forms to be evaluated for each iteration of the loop.

More generally, the first argument of for allows the same types of clauses as lfor:

(for [x [1 2 3]  :if (!= x 2)  y [7 8]]
  (print x y))
; Output:  1 7  1 8  3 7  3 8

In particular, you can use an :async clause to get the equivalent of Python's async for:

(import asyncio)
(defn :async numbers []
  (yield 1)
  (yield 2))
(asyncio.run ((fn :async []
  (for [:async x (numbers)]
    (print x)))))

The last argument of for can be an (else …) form. This form is executed after the last iteration of the for's outermost iteration clause, but only if that outermost loop terminates normally. If it's jumped out of with e.g. break, the else is ignored.

(for [x [1 2 3]]
  (print x)
  (when (= x 2)
    (break))
  (else (print "loop finished")))
macro(lfor #* args)

The comprehension forms lfor, sfor, dfor, gfor, and for are used to produce various kinds of loops, including Python-style comprehensions. lfor in particular can create a list comprehension. A simple use of lfor is:

(lfor  x (range 5)  (* 2 x))  ; => [0 2 4 6 8]

x is the name of a new variable, which is bound to each element of (range 5). Each such element in turn is used to evaluate the value form (* 2 x), and the results are accumulated into a list.

Here's a more complex example:

(lfor
  x (range 3)
  y (range 3)
  :if (!= x y)
  :setv total (+ x y)
  [x y total])
; => [[0 1 1] [0 2 2] [1 0 1] [1 2 3] [2 0 2] [2 1 3]]

When there are several iteration clauses (here, the pairs of forms x (range 3) and y (range 3)), the result works like a nested loop or Cartesian product: all combinations are considered in lexicographic order.

The general form of lfor is:

(lfor CLAUSES VALUE)

where the VALUE is an arbitrary form that is evaluated to produce each element of the result list, and CLAUSES is any number of clauses. There are several types of clauses:

  • Iteration clauses, which look like LVALUE ITERABLE. The LVALUE is usually just a symbol, but could be something more complicated, like [x y].

  • :async LVALUE ITERABLE, which is an asynchronous form of iteration clause per Python's async for.

  • :do FORM, which simply evaluates the FORM. If you use (continue) or (break) here, it will apply to the innermost iteration clause before the :do.

  • :setv LVALUE RVALUE, which is equivalent to :do (setv LVALUE RVALUE).

  • :if CONDITION, which is equivalent to :do (when (not CONDITION) (continue)).

For lfor, sfor, gfor, and dfor, variables defined by an iteration clause or :setv are not visible outside the form. However, variables defined within the body, as with a setx expression, will be visible outside the form. In for, by contrast, iteration and :setv clauses share the caller's scope and are visible outside the form.

macro(dfor #* args)

dfor creates a dictionary comprehension. Its syntax is the same as that of lfor except that it takes two trailing arguments. The first is a form producing the key of each dictionary element, and the second produces the value. Thus:

(dfor  x (range 5)  x (* x 10))
  ; => {0 0  1 10  2 20  3 30  4 40}
macro(gfor #* args)

gfor creates a generator expression. Its syntax is the same as that of lfor. The difference is that gfor returns an iterator, which evaluates and yields values one at a time:

(import itertools [count take-while])
(setv accum [])
(list (take-while
  (fn [x] (< x 5))
  (gfor x (count) :do (.append accum x) x)))
    ; => [0 1 2 3 4]
accum
    ; => [0 1 2 3 4 5]
macro(sfor #* args)

sfor creates a set comprehension. (sfor CLAUSES VALUE) is equivalent to (set (lfor CLAUSES VALUE)). See lfor.

Context managers and pattern-matching

macro(with managers #* body)

with compiles to a with or an async with statement, which wraps some code with one or more context managers. The first argument is a bracketed list of context managers, and the remaining arguments are body forms.

The manager list can't be empty. If it has only one item, that item is evaluated to obtain the context manager to use. If it has two, the first argument (a symbol) is bound to the result of the second. Thus, (with [(f)] …) compiles to with f(): and (with [x (f)] …) compiles to with f() as x: .

(with [o (open "file.txt" "rt")]
  (print (.read o)))

If the manager list has more than two items, they're understood as variable-manager pairs; thus

(with [v1 e1  v2 e2  v3 e3] ...)

compiles to

with e1 as v1, e2 as v2, e3 as v3: ...

The symbol _ is interpreted specially as a variable name in the manager list: instead of binding the context manager to the variable _ (as Python's with e1 as _: ), with will leave it anonymous (as Python's with e1: ).

Finally, any variable-manager pair may be preceded with the keyword :async to use an asynchronous context manager:

(with [:async v1 e1] )

with returns the value of its last form, unless it suppresses an exception (because the context manager's __exit__ method returned true), in which case it returns None. So, the first example could also be written

(print (with [o (open "file.txt" "rt")] (.read o)))
macro(match subject #* cases)

match compiles to a match statement. It requires Python 3.10 or later. The first argument should be the subject, and any remaining arguments should be pairs of patterns and results. The match form returns the value of the corresponding result, or None if no case matched.

(match (+ 1 1)
  1 "one"
  2 "two"
  3 "three")
; => "two"

You can use do to build a complex result form. Patterns, as in Python match statements, are interpreted specially and can't be arbitrary forms. Use (| …) for OR patterns, PATTERN :as NAME for AS patterns, and syntax like the usual Hy syntax for literal, capture, value, sequence, mapping, and class patterns. Guards are specified with :if FORM. Here's a more complex example:

(match #(100 200)
  [100 300]               "Case 1"
  [100 200] :if flag      "Case 2"
  [900   y]               f"Case 3, y: {y}"
  [100 (| 100 200) :as y] f"Case 4, y: {y}"
  _                       "Case 5, I match anything!")

This will match case 2 if flag is true and case 4 otherwise.

match can also match against class instances by keyword (or positionally if its __match_args__ attribute is defined; see PEP 636):

(import  dataclasses [dataclass])
(defclass [dataclass] Point []
  #^ int x
  #^ int y)
(match (Point 1 2)
  (Point 1 x) :if (= (% x 2) 0) x)  ; => 2

It's worth emphasizing that match is a pattern-matching construct rather than a generic switch construct, and retains all of Python's limitations on match patterns. For example, you can't match against the value of a variable. For more flexible branching constructs, see Hyrule's branch and case, or simply use cond.

Exception-handling

macro(raise exception :from other)

raise compiles to a raise statement, which throws an exception. With no arguments, the current exception is reraised. With one argument, an exception, that exception is raised.

(try
  (raise KeyError)
  (except [KeyError]
    (print "gottem")))

raise supports one other syntax, (raise EXCEPTION_1 :from EXCEPTION_2), which compiles to raise EXCEPTION_1 from EXCEPTION_2.

macro(try #* body)

try compiles to a try statement, which can catch exceptions and run cleanup actions. It begins with any number of body forms. Then follows any number of except or except* (PEP 654) forms, which are expressions that begin with the symbol in question, followed by a list of exception types, followed by more body forms. Finally there are an optional else form and an optional finally form, which again are expressions that begin with the symbol in question and then comprise body forms. Note that except* requires Python 3.11, and except* and except may not both be used in the same try.

Here's an example of several of the allowed kinds of child forms:

(try
  (error-prone-function)
  (another-error-prone-function)
  (except [ZeroDivisionError]
    (print "Division by zero"))
  (except [[IndexError KeyboardInterrupt]]
    (print "Index error or Ctrl-C"))
  (except [e ValueError]
    (print "ValueError:" (repr e)))
  (except [e [TabError PermissionError ReferenceError]]
    (print "Some sort of error:" (repr e)))
  (else
    (print "No errors"))
  (finally
    (print "All done")))

Exception lists can be in any of several formats:

  • [] to catch any subtype of Exception, like Python's except:

  • [ETYPE] to catch only the single type ETYPE, like Python's except ETYPE:

  • [[ETYPE1 ETYPE2 …]] to catch any of the named types, like Python's except ETYPE1, ETYPE2, …:

  • [VAR ETYPE] to catch ETYPE and bind it to VAR, like Python's except ETYPE as VAR:

  • [VAR [ETYPE1 ETYPE2 …]] to catch any of the named types and bind it to VAR, like Python's except ETYPE1, ETYPE2, as VAR:

The return value of try is the last form evaluated among the main body, except forms, except* forms, and else.

Functions

macro(defn name #* args)

defn compiles to a function definition (or possibly to an assignment of a lambda expression). It always returns None. It requires two arguments: a name (given as a symbol; see fn for anonymous functions) and a "lambda list", or list of parameters (also given as symbols). Any further arguments constitute the body of the function:

(defn name [params] bodyform1 bodyform2…)

An empty body is implicitly (return None). If there are at least two body forms, and the first of them is a string literal, this string becomes the docstring of the function. The final body form is implicitly returned; thus, (defn f [] 5) is equivalent to (defn f [] (return 5)). There is one exception: due to Python limitations, no implicit return is added if the function is an asynchronous generator (i.e., defined with (defn :async …) or (fn :async …) and containing at least one yield).

defn accepts a few more optional arguments: a literal keyword :async (to create a coroutine like Python's async def), a bracketed list of decorators, a list of type parameters (see below), and an annotation (see annotate) for the return value. These are placed before the function name (in that order, if several are present):

(defn :async [decorator1 decorator2] :tp [T1 T2] #^ annotation name [params] )

defn lambda lists support all the same features as Python parameter lists and hence are complex in their full generality. The simplest case is a (possibly empty) list of symbols, indicating that all parameters are required, and can be set by position, as in (f value), or by name, as in (f :argument value). To set a default value for a parameter, replace the parameter with the bracketed list [pname value], where pname is the parameter name as a symbol and value is an arbitrary form. Beware that, per Python, value is evaluated when the function is defined, not when it's called, and if the resulting object is mutated, all calls will see the changes.

Further special lambda-list syntax includes:

/

If the symbol / is given in place of a parameter, it means that all the preceding parameters can only be set positionally.

*

If the symbol * is given in place of a parameter, it means that all the following parameters can only be set by name.

#* args

If the parameter list contains #* args or (unpack-iterable args), then args is set to a tuple containing all otherwise unmatched positional arguments. The name args is merely cherished Python tradition; you can use any symbol.

#** kwargs

#** kwargs (a.k.a. (unpack-mapping kwargs)) is like #* args, but collects unmatched keyword arguments into a dictionary.

Each of these special constructs is allowed only once, and has the same restrictions as in Python; e.g., #* args must precede #** kwargs if both are present. Here's an example with a complex lambda list:

(defn f [a / b [c 3] * d e #** kwargs]
  [a b c d e kwargs])
(print (hy.repr (f 1 2 :d 4 :e 5 :f 6)))
  ; => [1 2 3 4 5 {"f" 6}]

Type parameters require Python 3.12, and have the semantics specified by PEP 695. The keyword :tp introduces the list of type parameters. Each item of the list is a symbol, an annotated symbol (such as #^ int T), or an unpacked symbol (such as #* T or #** T). As in Python, unpacking and annotation can't be used with the same parameter.

macro(fn args)

As defn, but no name for the new function is required (or allowed), and the newly created function object is returned. Decorators and type parameters aren't allowed, either. However, the function body is understood identically to that of defn, without any of the restrictions of Python's lambda. :async is also allowed.

macro(return object)

return compiles to a return statement. It exits the current function, returning its argument if provided with one, or None if not.

(defn f [x]
  (for [n (range 10)]
    (when (> n x)
      (return n))))
(f 3.9)  ; => 4

Note that in Hy, return is necessary much less often than in Python. The last form of a function is returned automatically, so an explicit return is only necessary to exit a function early. To get Python's behavior of returning None when execution reaches the end of a function, just put None there yourself:

(defn f []
  (setv d (dict :a 1 :b 2))
  (.pop d "b")
  None)
(print (f))  ; Prints "None", not "2"
macro(yield arg1 arg2)

yield compiles to a yield expression, which returns a value as a generator. For a plain yield, provide one argument, the value to yield, or omit it to yield None.

(defn naysayer []
  (while True
    (yield "nope")))
(list (zip "abc" (naysayer)))
  ; => [#("a" "nope") #("b" "nope") #("c" "nope")]

For a yield-from expression, provide two arguments, where the first is the literal keyword :from and the second is the subgenerator.

(defn myrange []
  (setv r (range 10))
  (while True
    (yield :from r)))
(list (zip "abc" (myrange)))
  ; => [#("a" 0) #("b" 1) #("c" 2)]
macro(await obj)

await creates an await expression. It takes exactly one argument: the object to wait for.

(import asyncio)
(defn :async main []
  (print "hello")
  (await (asyncio.sleep 1))
  (print "world"))
(asyncio.run (main))

Macros

macro(defmacro name lambda-list #* body)

Define a macro, at both compile-time and run-time. The syntax is a subset allowed of that by defn: no decorator or return-type annotations are allowed, and the only types of parameter allowed are symbol, [symbol default-value], /, and #* args. See Macros for details and examples.

macro(hy.core.macros.defreader _hy-compiler key #* body)

Define a reader macro, at both compile-time and run-time. After the name, all arguments are body forms: there is no parameter list as for defmacro, since it's up to the reader macro to decide how to parse the source text following its call position. See Reader macros for details and examples.

macro(hy.core.macros.get-macro _hy-compiler arg1 arg2)

Get the function object used to implement a macro. This works for all sorts of macros: core macros, global (i.e., module-level) macros, local macros, and reader macros. For regular (non-reader) macros, get-macro is called with one argument, a symbol or string literal, which can be premangled or not according to taste. For reader macros, this argument must be preceded by the literal keyword :reader (and note that the hash mark, #, is not included in the name of the reader macro).

(get-macro my-macro)
(get-macro :reader my-reader-macro)

Except when retrieving a local macro, get-macro expands to a get form on the appropriate object, such as _hy_macros, selected at the time of expanding get-macro. This means you can say (del (get-macro …)), perhaps wrapped in eval-and-compile or eval-when-compile, to delete a macro, but it's easy to get confused by the order of evaluation and number of evaluations. For more predictable results in complex situations, use (del (get …)) directly instead of (del (get-macro …)).

macro(hy.core.macros.local-macros _hy-compiler)

Expands to a dictionary mapping the mangled names of local macros to the function objects used to implement those macros. Thus, local-macros provides a rough local equivalent of _hy_macros.

(defn f []
  (defmacro m []
    "This is the docstring for the macro `m`."
    1)
  (help (get (local-macros) "m")))
(f)

The equivalency is rough in the sense that local-macros returns a literal dictionary, not a preexisting object that Hy uses for resolving macro names. So, modifying the dictionary will have no effect.

See also get-macro.

Classes

macro(defclass arg1 #* args)

defclass compiles to a class statement, which creates a new class. It always returns None. Only one argument, specifying the name of the new class as a symbol, is required. A list of decorators (and type parameters, in the same way as for defn) may be provided before the class name. After the name comes a list of superclasses (use the empty list [] for the common case of no superclasses) and any number of body forms, the first of which may be a docstring.

A simple class declaration and its uses might look like this:

(defclass MyClass []
  "A simple example class."

  (setv i 12345)

  (defn f [self]
    "hello world"))

(setv instance (MyClass))
(print instance.i)        ; => 12345
(print (.f instance))     ; => hello world

A more complex declaration might look like this:

(defclass [decorator1 decorator2] :tp [T1 T2] MyClass [SuperClass1 SuperClass2]
  "A class that does things at times."

  (setv
    attribute1 value1
    attribute2 value2)

  (defn method1 [self arg1 arg2]
    )

  (defn method2 [self arg1 arg2]
    ))

Modules

macro(import #* forms)

import compiles to an import statement, which makes objects in a different module available in the current module. It always returns None. Hy's syntax for the various kinds of import looks like this:

;; Import each of these modules.
;; Python: import sys, os.path
(import sys os.path)

;; Import several names from a single module.
;; Python: from os.path import exists, isdir as is_dir, isfile
(import os.path [exists  isdir :as dir?  isfile])

;; Import a module with an alias for the whole module.
;; Python: import sys as systest
(import sys :as systest)

;; Import all objects from a module into the current namespace.
;; Python: from sys import *
(import sys *)

;; You can list as many imports as you like of different types.
;; Python:
;;     from tests.resources import kwtest, function_with_a_dash
;;     from os.path import exists, isdir as is_dir, isfile as is_file
;;     import sys as systest
;;     from math import *
(import tests.resources [kwtest function-with-a-dash]
        os.path [exists
                 isdir :as dir?
                 isfile :as file?]
        sys :as systest
        math *)

__all__ can be set to control what's imported by (import module-name *), as in Python, but beware that all names in __all__ must be mangled. The macro export is a handy way to set __all__ in a Hy program.

macro(require #* args)

require is a version of import for macros. It allows all the same syntax as import, and brings the requested macros into the current scope at compile-time as well as run-time. The following are all equivalent ways to call a macro named foo in the module mymodule:

(require mymodule)
(mymodule.foo 1)

(require mymodule :as M)
(M.foo 1)

(require mymodule [foo])
(foo 1)

(require mymodule *)
(foo 1)

(require mymodule [foo :as bar])
(bar 1)

There's a bit of a trick involved in syntax such as mymodule.foo. Namely, there is no object named mymodule. Instead, (require mymodule) assigns every macro foo in mymodule to the name (hy.mangle "mymodule.foo") in _hy_macros.

Reader macros have a different namespace from regular macros, so they need to be specified with the added syntax :readers […]. You could require a reader macro named spiff with the call (require mymodule :readers [spiff]), or star-require reader macros with (require mymodule :readers *). For legibility, a regular-macros specification may analogously be prefixed :macros:

(require mymodule :macros [foo] :readers [spiff])

require with reader macros is more limited than with regular macros. You can't access reader macros with dotted names, and you can't rename them with :as.

Note that (require mymodule :readers [spiff]) doesn't imply (require mymodule); that is, mymodule.foo won't be made available. If you want that, use something like

(require mymodule
         mymodule :readers [spiff])

To define which macros are collected by (require mymodule *), set the variable _hy_export_macros (analogous to Python's __all__) to a list of mangled macro names, which is accomplished most conveniently with export. The default behavior is analogous to (import mymodule *): all macros are collected other than those whose mangled names begin with an underscore (_),

macro(hy.core.macros.export #* args)

A convenience macro for defining __all__ and _hy_export_macros, which control which Python objects and macros (respectively) are collected by * imports in import and require (respectively). export allows you to provide the names as symbols instead of strings, and it calls hy.mangle for you on each name.

The syntax is (export objects macros), where objects refers to Python objects and macros to macros. Keyword arguments are allowed. For example,

(export
  :objects [my-fun MyClass]
  :macros [my-macro])

exports the function my-fun, the class MyClass, and the macro my-macro.

Miscellany

macro(chainc #* args)

chainc creates a comparison expression. It isn't required for unchained comparisons, which have only one comparison operator, nor for chains of the same operator. For those cases, you can use the comparison operators directly with Hy's usual prefix syntax, as in (= x 1) or (< 1 2 3). The use of chainc is to construct chains of heterogeneous operators, such as x <= y < z. It uses an infix syntax with the general form

(chainc ARG OP ARG OP ARG…)

Hence, (chainc x <= y < z) is equivalent to (and (<= x y) (< y z)), including short-circuiting, except that y is only evaluated once.

Each ARG is an arbitrary form, which does not itself use infix syntax. Use py if you want fully Python-style operator syntax. You can also nest chainc forms, although this is rarely useful. Each OP is a literal comparison operator; other forms that resolve to a comparison operator are not allowed.

At least two ARGs and one OP are required, and every OP must be followed by an ARG.

As elsewhere in Hy, the equality operator is spelled =, not == as in Python.

macro(assert condition [label None])

assert compiles to an assert statement, which checks whether a condition is true. The first argument, specifying the condition to check, is mandatory, whereas the second, which will be passed to AssertionError, is optional. The whole form is only evaluated when __debug__ is true, and the second argument is only evaluated when __debug__ is true and the condition fails. assert always returns None.

(assert (= 1 2) "one should equal two")
  ; AssertionError: one should equal two

Placeholder macros

There are a few core macros that are unusual in that all they do, when expanded, is crash, regardless of their arguments:

  • else

  • except

  • except*

  • finally

  • unpack-mapping

  • unquote

  • unquote-splice

The purpose of these macros is merely to reserve their names. Each symbol is interpreted specially by one or more other core macros (e.g., else in while) and thus, in these contexts, any definition of these names as a function or macro would be ignored. If you really want to, you can override these names like any others, but beware that, for example, trying to call your new else inside while may not work.

Hy

A few core functions, mostly related to the manipulation of Hy code, are available through the module hy.

(hy.read stream filename reader)

Like hy.read-many, but only one form is read, and shebangs are forbidden. The model corresponding to this specific form is returned, or, if there are no forms left in the stream, EOFError is raised. stream.pos is left where it was immediately after the form.

(hy.read-many stream  [filename <string>] reader  [skip-shebang False])

Parse all the Hy source code in stream, which should be a textual file-like object or a string. filename, if provided, is used in error messages. If no reader is provided, a new hy.HyReader object is created. If skip_shebang is true and a shebang line is present, it's detected and discarded first.

Return a value of type hy.models.Lazy. If you want to evaluate this, be careful to allow evaluating each model before reading the next, as in (hy.eval (hy.read-many o)). By contrast, forcing all the code to be read before evaluating any of it, as in (hy.eval `(do [~@(hy.read-many o)])), will yield the wrong result if one form defines a reader macro that's later used in the same stream to produce new forms.

Warning

Thanks to reader macros, reading can execute arbitrary code. Don't read untrusted input.

(hy.eval model globals locals module macros)

An equivalent of Python's eval() for evaluating Hy code. The chief difference is that the first argument should be a model rather than source text. If you have a string of source text you want to evaluate, convert it to a model first with hy.read or hy.read-many:

(hy.eval '(+ 1 1))             ; => 2
(hy.eval (hy.read "(+ 1 1)"))  ; => 2

The optional arguments globals and locals work as in the case of eval().

Another optional argument, module, can be a module object or a string naming a module. The module's __dict__ attribute can fill in for globals (and hence also for locals) if module is provided but globals isn't, but the primary purpose of module is to control where macro calls are looked up. Without this argument, the calling module of hy.eval is used instead.

(defmacro my-test-mac [] 3)
(hy.eval '(my-test-mac))                 ; => 3
(import hyrule)
(hy.eval '(my-test-mac) :module hyrule)  ; NameError
(hy.eval '(list-n 3 1) :module hyrule)   ; => [1 1 1]

Finally, finer control of macro lookup can be achieved by passing in a dictionary of macros as the macros argument. The keys of this dictionary should be mangled macro names, and the values should be function objects to implement those macros. This is the same structure as is produced by local-macros, and in fact, (hy.eval :macros (local-macros)) is useful to make local macros visible to hy.eval, which otherwise doesn't see them.

(defn f []
  (defmacro lmac [] 1)
  (hy.eval '(lmac))     ; NameError
  (print (hy.eval '(lmac) :macros (local-macros)))) ; => 1
(f)

In any case, macros provided in this dictionary will shadow macros of the same name that are associated with the provided or implicit module. You can shadow a core macro, too, so be careful: there's no warning for this as there is in the case of defmacro.

(hy.repr obj)

This function is Hy's equivalent of Python's repr(). It returns a string representing the input object in Hy syntax.

(hy.repr [1 2 3])  ; => "[1 2 3]"
(repr [1 2 3])     ; => "[1, 2, 3]"

Like repr in Python, hy.repr can round-trip many kinds of values. Round-tripping implies that given an object x, (hy.eval (hy.read (hy.repr x))) returns x, or at least a value that's equal to x. A notable exception to round-tripping is that if a model contains a non-model, the latter will be promoted to a model in the output:

(setv
  x (hy.models.List [5])
  output (hy.repr x)
  y (hy.eval (hy.read output)))
(print output)            ; '[5]
(print (type (get x 0)))  ; <class 'int'>
(print (type (get y 0)))  ; <class 'hy.models.Integer'>

When hy.repr doesn't know how to represent an object, it falls back on repr(). Use hy.repr-register to add your own conversion function for a type instead.

(hy.repr-register types f placeholder)

hy.repr-register lets you set the function that hy.repr calls to represent a type:

(defclass C)
(hy.repr-register C (fn [x] "cuddles"))
(hy.repr [1 (C) 2])  ; => "[1 cuddles 2]"

Registered functions often call hy.repr themselves. hy.repr will automatically detect self-references, even deeply nested ones, and output "..." for them instead of calling the usual registered function. To use a placeholder other than "...", pass a string of your choice as the placeholder argument:

(defclass Container)
(hy.repr-register Container :placeholder "HY THERE"
  (fn [x] f"(Container {(hy.repr x.value)})"))
(setv container (Container))
(setv container.value container)
(hy.repr container)   ; => "(Container HY THERE)"
(hy.mangle s)

Stringify the argument (with str, not repr() or hy.repr) and convert it to a valid Python identifier according to Hy's mangling rules.

(hy.mangle 'foo-bar)   ; => "foo_bar"
(hy.mangle "🦑")       ; => "hyx_XsquidX"

If the stringified argument is already both legal as a Python identifier and normalized according to Unicode normalization form KC (NFKC), it will be returned unchanged. Thus, hy.mangle is idempotent.

(setv x '♦-->♠)
(= (hy.mangle (hy.mangle x)) (hy.mangle x))  ; => True

Generally, the stringifed input is expected to be parsable as a symbol. As a convenience, it can also have the syntax of a dotted identifier, and hy.mangle will mangle the dot-delimited parts separately.

(hy.mangle "a.c!.d")  ; => "a.hyx_cXexclamation_markX.d"
(hy.unmangle s)

Stringify the argument and try to convert it to a pretty unmangled form. See Hy's mangling rules.

(hy.unmangle "hyx_XsquidX")  ; => "🦑"

Unmangling may not round-trip, because different Hy symbol names can mangle to the same Python identifier. In particular, Python itself already considers distinct strings that have the same normalized form (according to NFKC), such as hello and 𝔥𝔢𝔩𝔩𝔬, to be the same identifier.

It's an error to call hy.unmangle on something that looks like a properly mangled name but isn't. For example, (hy.unmangle "hyx_XpizzazzX") is erroneous, because there is no Unicode character named "PIZZAZZ" (yet).

(hy.macroexpand model module macros)

As hy.macroexpand-1, but the expansion process is repeated until it has no effect.

(defmacro m [x]
  (and (int x) `(m ~(- x 1))))
(print (hy.repr (hy.macroexpand-1 '(m 5))))
  ; => '(m 4)
(print (hy.repr (hy.macroexpand '(m 5))))
  ; => '0

Note that in general, macro calls in the arguments of the expression still won't expanded. To expand these, too, try Hyrule's macroexpand-all.

(hy.macroexpand-1 model module macros)

Check if model is an Expression specifying a macro call. If so, expand the macro and return the expansion; otherwise, return model unchanged.

(defmacro m [x]
 `(do ~x ~x ~x))
(print (hy.repr (hy.macroexpand-1 '(m (+= n 1)))))
  ; => '(do (+= n 1) (+= n 1) (+= n 1))

An exceptional case is if the macro is a core macro that returns one of Hy's internal compiler result objects instead of a real model. Then, you just get the original back, as if the macro hadn't been expanded.

The optional arguments module and macros can be provided to control where macros are looked up, as with hy.eval.

See also hy.macroexpand.

(hy.gensym [g ])

Generate a symbol with a unique name. The argument, if provided, will be included in the generated symbol name, as an aid to debugging.

The below example uses the return value of f twice but calls it only once, and uses hy.gensym for the temporary variable to avoid collisions with any other variable names.

(defmacro selfadd [x]
  (setv g (hy.gensym))
  `(do
     (setv ~g ~x)
     (+ ~g ~g)))

(defn f []
  (print "This is only executed once.")
  4)

(print (selfadd (f)))
(hy.as-model x)

Convert x and any elements thereof into models recursively. This function is called implicitly by Hy in many situations, such as when inserting the expansion of a macro into the surrounding code, so you don't often need to call it. One use is to ensure that models are used on both sides of a comparison:

(= 7 '7)                ; => False
(= (hy.as-model 7) '7)  ; => True

It's an error to call hy.as-model on an object that contains itself, or an object that isn't representable as a Hy literal, such as a function.

class (hy.I)

hy.I is an object that provides syntactic sugar for imports. It allows syntax like (hy.I.math.sqrt 2) to mean (import math) (math.sqrt 2), except without bringing math or math.sqrt into scope. (See hy.R for a version that requires a macro instead of importing a Python object.) This is useful in macros to avoid namespace pollution. To refer to a module with dots in its name, use slashes instead: hy.I.os/path.basename gets the function basename from the module os.path.

You can also call hy.I like a function, as in (hy.I "math"), which is useful when the module name isn't known until run-time. This interface just calls importlib.import_module(), avoiding (1) mangling due to attribute lookup, and (2) the translation of / to . in the module name. The advantage of (hy.I modname) over importlib.import_module(modname) is merely that it avoids bringing importlib itself into scope.

class (hy.R)

There is no actual object named hy.R. Rather, this syntax is recognized specially by the compiler as a shorthand for requiring and calling a macro.

Readers

Hy's reader (i.e., parser) classes are most interesting to the user in the context of reader macros.

class hy.HyReader(*, use_current_readers=False)

A modular reader for Hy source. It inherits from hy.Reader.

When use_current_readers is true, initialize this reader with all reader macros from the calling module.

fill_pos(model, start)

Set position information for model. start should be a (line number, column number) tuple for the start position, whereas the end position is set to the current cursor position.

parse(stream, filename=None, skip_shebang=False)

Yield all models in stream. The parameters are understood as in hy.read-many.

parse_forms_until(closer)

Yield models until the character closer is seen. This method is useful for reading sequential constructs such as lists.

parse_one_form()

Parse the next form in the stream and return its model. Any preceding whitespace and comments are skipped over.

read_default(key)

Try to read an identifier. If the next character after that is ", then instead parse it as a string with the given prefix (e.g., r"...").

(This method is the default reader handler, for when nothing in the read table matches.)

class hy.Reader

An abstract base class for reading input character-by-character.

See hy.HyReader for an example of creating a reader class.

ends_ident

The set of characters that indicate the end of an identifier

Type:

set[str]

reader_table

A dictionary mapping a reader-macro key to its dispatch function

Type:

dict[str, Callable]

pos

A read-only (line, column) tuple indicating the current cursor position of the source being read

Type:

tuple[int, int]

chars(eof_ok=False)

Consume and yield characters of the stream. If eof_ok is false (the default) and the end of the stream is reached, raise hy.PrematureEndOfInput.

dispatch(tag)

Call the handler for the reader macro with key tag (a string). Return the model it produces, if any.

end_identifier(character)

A context manager to temporarily add a new character to the ends_ident set.

getc()

Consume one character from the stream and return it. This method does the bookkeeping for position data, so all character consumption should go through it.

getn(n)

Consume and return n characters.

peek_and_getc(target)

Peek at the next character and check if it's equal to target, only consuming it if it's equal. A bool is returned.

peekc()

Peek at the next character, returning it but not consuming it.

peeking(eof_ok=False)

As chars(), but without consuming any of the returned characters. This method is useful for looking several characters ahead.

read_ident(just_peeking=False)

Read characters until we hit something in ends_ident. The characters are consumed unless just_peeking is true.

saving_chars()

A context manager to save all read characters. The value is a list of characters, rather than a single string.

slurp_space()

Consume and return zero or more whitespace characters.

exception hy.PrematureEndOfInput(message, expression=None, filename=None, source=None, lineno=1, colno=1)

Raised when input ends unexpectedly during parsing.

Python operators

Python provides various binary and unary operators. These are usually invoked in Hy using core macros of the same name: for example, (+ 1 2) calls the core macro named +, which uses Python's addition operator. There are a few exceptions to the names being the same:

  • == in Python is = in Hy.

  • ~ in Python is bnot in Hy.

  • is not in Python is is-not in Hy.

  • not in in Python is not-in in Hy.

For Python's subscription expressions (like x[2]), Hy has two named macros, get and cut.

By importing from the module hy.pyops (typically with a star import, as in (import hy.pyops *)), you can also use these operators as functions. Functions are first-class objects, so you can say things like (map - xs) to negate all the numbers in the list xs. Since macros shadow functions, forms like (- 1 2) will still call the macro instead of the function. The functions in hy.pyops have the same semantics as their macro equivalents, with one exception: functions can't short-circuit, so the functions for operators such as and and != unconditionally evaluate all arguments.

Hy also provides macros for Python's augmented assignment operators (but no equivalent functions, because Python semantics don't allow for this). These macros require at least two arguments even if the parent operator doesn't; for example, (-= x) is an error even though (- x) is legal. If the parent operator supports more than two arguments, though, so does the augmented-assignment version, using an aggregation operator to bind up all arguments past the first into a single rvalue. Typically, the aggregator is the same as the original operator: for example, (+= count n1 n2 n3) is equivalent to (+= count (+ n1 n2 n3)). Exceptions (such as -=, which uses the aggregator +, so (-= count n1 n2 n3) is equivalent to (-= count (+ n1 n2 n3))) are noted in the documentation for the parent operator (such as - for -=).

(hy.pyops.!= a1 a2 #* a-rest)

The inequality operator. Its effect can be defined by the equivalent Python:

  • (!= x y)x != y

  • (!= a1 a2 an)a1 != a2 != != an

(hy.pyops.% x y)

The modulus operator. Its effect can be defined by the equivalent Python:

  • (% x y)x % y

(hy.pyops.& a1 #* a-rest)

The bitwise AND operator. Its effect can be defined by the equivalent Python:

  • (& x)x

  • (& x y)x & y

  • (& a1 a2 an)a1 & a2 & & an

(hy.pyops.* #* args)

The multiplication operator. Its effect can be defined by the equivalent Python:

  • (*)1

  • (* x)x

  • (* x y)x * y

  • (* a1 a2 an)a1 * a2 * * an

(hy.pyops.** a1 a2 #* a-rest)

The exponentiation operator. Its effect can be defined by the equivalent Python:

  • (** x y)x ** y

  • (** a1 a2 an)a1 ** a2 ** ** an

(hy.pyops.+ #* args)

The addition operator. Its effect can be defined by the equivalent Python:

  • (+)0

  • (+ x)+x

  • (+ x y)x + y

  • (+ a1 a2 an)a1 + a2 + + an

(hy.pyops.- a1 #* a-rest)

The subtraction operator. Its effect can be defined by the equivalent Python:

  • (- x)-x

  • (- x y)x - y

  • (- a1 a2 an)a1 - a2 - - an

Aggregator for augmented assignment: +

(hy.pyops./ a1 #* a-rest)

The division operator. Its effect can be defined by the equivalent Python:

  • (/ x)1 / x

  • (/ x y)x / y

  • (/ a1 a2 an)a1 / a2 / / an

Aggregator for augmented assignment: *

(hy.pyops.// a1 a2 #* a-rest)

The floor division operator. Its effect can be defined by the equivalent Python:

  • (// x y)x // y

  • (// a1 a2 an)a1 // a2 // // an

(hy.pyops.< a1 #* a-rest)

The less-than operator. Its effect can be defined by the equivalent Python:

  • (< x)True

  • (< x y)x < y

  • (< a1 a2 an)a1 < a2 < < an

(hy.pyops.<< a1 a2 #* a-rest)

The left shift operator. Its effect can be defined by the equivalent Python:

  • (<< x y)x << y

  • (<< a1 a2 an)a1 << a2 << << an

Aggregator for augmented assignment: +

(hy.pyops.<= a1 #* a-rest)

The less-than-or-equal-to operator. Its effect can be defined by the equivalent Python:

  • (<= x)True

  • (<= x y)x <= y

  • (<= a1 a2 an)a1 <= a2 <= <= an

(hy.pyops.= a1 #* a-rest)

The equality operator. Its effect can be defined by the equivalent Python:

  • (= x)True

  • (= x y)x == y

  • (= a1 a2 an)a1 == a2 == == an

(hy.pyops.> a1 #* a-rest)

The greater-than operator. Its effect can be defined by the equivalent Python:

  • (> x)True

  • (> x y)x > y

  • (> a1 a2 an)a1 > a2 > > an

(hy.pyops.>= a1 #* a-rest)

The greater-than-or-equal-to operator. Its effect can be defined by the equivalent Python:

  • (>= x)True

  • (>= x y)x >= y

  • (>= a1 a2 an)a1 >= a2 >= >= an

(hy.pyops.>> a1 a2 #* a-rest)

The right shift operator. Its effect can be defined by the equivalent Python:

  • (>> x y)x >> y

  • (>> a1 a2 an)a1 >> a2 >> >> an

Aggregator for augmented assignment: +

(hy.pyops.@ a1 #* a-rest)

The matrix multiplication operator. Its effect can be defined by the equivalent Python:

  • (@ x y)x @ y

  • (@ a1 a2 an)a1 @ a2 @ @ an

(hy.pyops.^ x y)

The bitwise XOR operator. Its effect can be defined by the equivalent Python:

  • (^ x y)x ^ y

(hy.pyops.and #* args)

The logical conjuction operator. Its effect can be defined by the equivalent Python:

  • (and)True

  • (and x)x

  • (and x y)x and y

  • (and a1 a2 an)a1 and a2 and and an

(hy.pyops.bnot x)

The bitwise NOT operator. Its effect can be defined by the equivalent Python:

  • (bnot x)~x

(hy.pyops.cut coll [arg1 sentinel] [arg2 sentinel] [arg3 sentinel])

cut compiles to a slicing expression, which selects multiple elements of a sequential data structure. The first argument is the object to be sliced. The remaining arguments are optional, and understood the same way as in a Python slicing expression.

(setv x "abcdef")
(cut x)           ; => "abcdef"
(cut x 2)         ; => "ab"
(cut x 2 None)    ; => "cdef"
(cut x 3 5)       ; => "de"
(cut x -3 None)   ; => "def"
(cut x 0 None 2)  ; => "ace"

A call to the cut macro (but not its function version in hy.pyops) is a valid target for assignment (with setv, +=, etc.) and for deletion (with del).

(hy.pyops.get coll key1 #* keys)

get compiles to one or more subscription expressions, which select an element of a data structure. The first two arguments are the collection object and a key; for example, (get person name) compiles to person[name]. Subsequent arguments indicate chained subscripts, so (get person name "surname" 0) becomes person[name]["surname"][0]. You can assign to a get form, as in

(setv real-estate {"price" 1,500,000})
(setv (get real-estate "price") 0)

but this doesn't work with the function version of get from hy.pyops, due to Python limitations on lvalues.

If you're looking for the Hy equivalent of Python list slicing, as in foo[1:3], note that this is just Python's syntactic sugar for foo[slice(1, 3)], and Hy provides its own syntactic sugar for this with a different macro, cut.

See also

  • The dot macro ., which can also subscript

  • Hyrule's assoc, to easily assign multiple elements of a single collection

(hy.pyops.in a1 a2 #* a-rest)

The membership test operator. Its effect can be defined by the equivalent Python:

  • (in x y)x in y

  • (in a1 a2 an)a1 in a2 in in an

(hy.pyops.is a1 #* a-rest)

The identicality test operator. Its effect can be defined by the equivalent Python:

  • (is x)True

  • (is x y)x is y

  • (is a1 a2 an)a1 is a2 is is an

(hy.pyops.is-not a1 a2 #* a-rest)

The negated identicality test operator. Its effect can be defined by the equivalent Python:

  • (is-not x y)x is not y

  • (is-not a1 a2 an)a1 is not a2 is not is not an

(hy.pyops.not-in a1 a2 #* a-rest)

The negated membership test operator. Its effect can be defined by the equivalent Python:

  • (not-in x y)x not in y

  • (not-in a1 a2 an)a1 not in a2 not in not in an

(hy.pyops.or #* args)

The logical disjunction operator. Its effect can be defined by the equivalent Python:

  • (or)None

  • (or x)x

  • (or x y)x or y

  • (or a1 a2 an)a1 or a2 or or an

(hy.pyops.| #* args)

The bitwise OR operator. Its effect can be defined by the equivalent Python:

  • (|)0

  • (| x)x

  • (| x y)x | y

  • (| a1 a2 an)a1 | a2 | | an