Principled Meta Programming
Edit this page on GitHubPrincipled meta programming is a new framework for staging and for some forms of macros. It is expressed as strongly and statically typed code using two fundamental operations: quotations and splicing. A novel aspect of the approach is that these two operations are regulated by a phase consistency principle that treats quotes and splices in exactly the same way.
Overview
Quotes and Splices
Principled meta programming is built on two wellknown fundamental
operations: quotation and splicing. Quotation is expressed as
'(...)
or '{...}
for expressions (both forms are equivalent) and
as '[...]
for types. Splicing is expressed as a prefix ~
operator.
For example, the code below presents an inline function assert
which calls at compiletime a method assertImpl
with a boolean
expression tree as argument. assertImpl
evaluates the expression and
prints it again in an error message if it evaluates to false
.
import scala.quoted._
inline def assert(expr: => Boolean): Unit =
~ assertImpl('(expr))
def assertImpl(expr: Expr[Boolean]) = '{
if !(~expr) then
throw new AssertionError(s"failed assertion: ${~showExpr(expr)}")
}
def showExpr(expr: Expr[Boolean]): Expr[String] =
'("<some source code>") // Better implementation later in this document
If e
is an expression, then '(e)
or '{e}
represent the typed
abstract syntax tree representing e
. If T
is a type, then '[T]
represents the type structure representing T
. The precise
definitions of "typed abstract syntax tree" or "typestructure" do not
matter for now, the terms are used only to give some
intuition. Conversely, ~ e
evaluates the expression e
, which must
yield a typed abstract syntax tree or type structure, and embeds the
result as an expression (respectively, type) in the enclosing program.
Quotations can have spliced parts in them; in this case the embedded splices are evaluated and embedded as part of the formation of the quotation.
Quotes and splices are duals of each other. For arbitrary
expressions e
and types T
we have:
~'(e) = e
'(~e) = e
~'[T] = T
'[~T] = T
Types for Quotations
The type signatures of quotes and splices can be described using two fundamental types:
Expr[T]
: abstract syntax trees representing expressions of typeT
Type[T]
: type structures representing typeT
.
Quoting takes expressions of type T
to expressions of type Expr[T]
and it takes types T
to expressions of type Type[T]
. Splicing
takes expressions of type Expr[T]
to expressions of type T
and it
takes expressions of type Type[T]
to types T
.
The two types can be defined in package scala.quoted
as follows:
package scala.quoted
sealed abstract class Expr[T] {
def unary_~: T // splice operation
}
sealed abstract class Type[T] {
type unary_~ = T // splice type
}
Both Expr
and Type
are abstract and sealed, so all constructors for
these types are provided by the system. One way to construct values of
these types is by quoting, the other is by typespecific lifting
operations that will be discussed later on.
The Phase Consistency Principle
A fundamental phase consistency principle (PCP) regulates accesses to free variables in quoted and spliced code:
 For any free variable reference
x
, the number of quoted scopes and the number of spliced scopes between the reference tox
and the definition ofx
must be equal.
Here, this
references count as free variables. On the other
hand, we assume that all imports are fully expanded and that _root_
is
not a free variable. So references to global definitions are
allowed everywhere.
The phase consistency principle can be motivated as follows: First,
suppose the result of a program P
is some quoted text '{ ... x ... }
that refers to a free variable x
in P
This can be
represented only by referring to the original variable x
. Hence, the
result of the program will need to persist the program state itself as
one of its parts. We don’t want to do this, hence this situation
should be made illegal. Dually, suppose a toplevel part of a program
is a spliced text ~{ ... x ... }
that refers to a free variable x
in P
. This would mean that we refer during construction of P
to
a value that is available only during execution of P
. This is of
course impossible and therefore needs to be ruled out. Now, the
smallstep evaluation of a program will reduce quotes and splices in
equal measure using the cancellation rules above. But it will neither
create nor remove quotes or splices individually. So the PCP ensures
that program elaboration will lead to neither of the two unwanted
situations described above.
In what concerns the range of features it covers, principled meta programming is
quite close to the MetaML family of languages. One difference is that MetaML does
not have an equivalent of the PCP  quoted code in MetaML can access
variables in its immediately enclosing environment, with some
restrictions and caveats since such accesses involve serialization.
However, this does not constitute a fundamental gain in
expressiveness. Principled meta programming allows to define a Liftable
typeclass which can implement such accesses within the confines of the
PCP. This is explained further in a later section.
Details
From Expr
s to Functions and Back
The Expr
companion object contains an AsFunctionN
(for 0 <= N < 23) decorator that turns a tree
describing a function into a function mapping trees to trees.
object Expr {
...
implicit class AsFunction1[T, U](f: Expr[T => U]) extends AnyVal {
def apply(x: Expr[T]): Expr[U] = ???
}
}
This decorator gives Expr
the apply
operation of an applicative functor, where Expr
s
over function types can be applied to Expr
arguments. The definition
of AsFunction1(f).apply(x)
is assumed to be functionally the same as
'((~f)(~x))
, however it should optimize this call by returning the
result of betareducing f(x)
if f
is a known lambda expression.
The AsFunction1
decorator distributes applications of Expr
over function
arrows:
AsFunction1(_).apply: Expr[S => T] => (Expr[S] => Expr[T])
Its dual, let’s call it reflect
, can be defined as follows:
def reflect[T, U](f: Expr[T] => Expr[U]): Expr[T => U] = '{
(x: T) => ~f('(x))
}
Note how the fundamental phase consistency principle works in two
different directions here for f
and x
. The reference to f
is
legal because it is quoted, then spliced, whereas the reference to x
is legal because it is spliced, then quoted.
Types and the PCP
In principle, The phase consistency principle applies to types as well
as for expressions. This might seem too restrictive. Indeed, the
definition of reflect
above is not phase correct since there is a
quote but no splice between the parameter binding of T
and its
usage. But the code can be made phase correct by adding a binding
of a Type[T]
tag:
def reflect[T, U](f: Expr[T] => Expr[U])(implicit t: Type[T]): Expr[T => U] =
'{ (x: ~t) => ~f('(x)) }
In this version of reflect
, the type of x
is now the result of
splicing the Type
value t
. This operation is splice correct  there
is one quote and one splice between the use of t
and its definition.
To avoid clutter, the Scala implementation tries to convert any phaseincorrect
reference to a type T
to a typesplice, by rewriting T
to ~implicitly[Type[T]]
.
For instance, the userlevel definition of reflect
:
def reflect[T: Type, U](f: Expr[T] => Expr[U]): Expr[T => U] =
'{ (x: T) => ~f('(x)) }
would be rewritten to
def reflect[T: Type, U](f: Expr[T] => Expr[U]): Expr[T => U] =
'{ (x: ~implicitly[Type[T]]) => ~f('(x)) }
The implicitly
query succeeds because there is an implicit value of
type Type[T]
available (namely the evidence parameter corresponding
to the context bound : Type
), and the reference to that value is
phasecorrect. If that was not the case, the phase inconsistency for
T
would be reported as an error.
Lifting Types
The previous section has shown that the metaprogramming framework has
to be able to take a type T
and convert it to a type tree of type
Type[T]
that can be reified. This means that all free variables of
the type tree refer to types and values defined in the current stage.
For a reference to a global class, this is easy: Just issue the fully
qualified name of the class. Members of reifiable types are handled by
just reifying the containing type together with the member name. But
what to do for references to type parameters or local type definitions
that are not defined in the current stage? Here, we cannot construct
the Type[T]
tree directly, so we need to get it from a recursive
implicit search. For instance, to implemenent
implicitly[Type[List[T]]]
where T
is not defined in the current stage, we construct the type constructor
of List
applied to the splice of the result of searching for an implicit Type[T]
:
'[List[~implicitly[Type[T]]]]
This is in exactly the algorithm that Scala 2 uses to search for type tags.
In fact Scala 2's type tag feature can be understood as a more adhoc version of
quoted.Type
. As was the case for type tags, the implicit search for a quoted.Type
is handled by the compiler, using the algorithm sketched above.
Example Expansion
Assume an Array
class with an inline map
method that forwards to a macro implementation.
class Array[T] {
inline def map[U](f: T => U): Array[U] = ~ Macros.mapImpl[T, U]('[U], '(this), '(f))
}
Here’s the definition of the mapImpl
macro, which takes quoted types and expressions to a quoted expression:
object Macros {
def mapImpl[T, U](u: Type[U], arr: Expr[Array[T]], op: Expr[T => U]): Expr[Array[U]] = '{
var i = 0
val xs = ~arr
var len = xs.length
val ys = new Array[~u](len)
while (i < len) {
ys(i) = ~op('(xs(i)))
i += 1
}
ys
}
}
Here’s an application of map
and how it rewrites to optimized code:
genSeq[Int]().map(x => x + 1)
==> (inline)
val $this: Seq[Int] = genSeq[Int]()
val f: Int => Int = x => x + 1
~ _root_.Macros.mapImpl[Int, Int]('[Int], '($this), '(f))
==> (splice)
val $this: Seq[Int] = genSeq[Int]()
val f: Int => Int = x => x + 1
{
var i = 0
val xs = ~'($this)
var len = xs.length
val ys = new Array[~'[Int]](len)
while (i < len) {
ys(i) = ~('(f)('(xs(i))))
i += 1
}
ys
}
==> (expand and splice inside quotes)
val $this: Seq[Int] = genSeq[Int]()
val f: Int => Int = x => x + 1
{
var i = 0
val xs = $this
var len = xs.length
val ys = new Array[Int](len)
while (i < len) {
ys(i) = xs(i) + 1
i += 1
}
ys
}
==> (elim dead code)
val $this: Seq[Int] = genSeq[Int]()
{
var i = 0
val xs = $this
var len = xs.length
val ys = new Array[Int](len)
while (i < len) {
ys(i) = xs(i) + 1
i += 1
}
ys
}
Relationship with Inline and Macros
Seen by itself, principled metaprogramming looks more like a
framework for staging than one for compiletime meta programming with
macros. But combined with Dotty’s inline
feature it can be turned into a
compiletime system. The idea is that macro elaboration can be
understood as a combination of a macro library and a quoted
program. For instance, here’s the assert
macro again together with a
program that calls assert
.
object Macros {
inline def assert(expr: => Boolean): Unit =
~ assertImpl('(expr))
def assertImpl(expr: Expr[Boolean]) =
'{ if !(~expr) then throw new AssertionError(s"failed assertion: ${~expr}") }
}
// has to be in a different compilation unit that depends on Macros
object App {
val program = {
val x = 1
Macros.assert(x != 0)
}
}
Inlining the assert
function would give the following program:
val program = {
val x = 1
~Macros.assertImpl('(x != 0))
}
The example is only phase correct because Macros is a global value and
as such not subject to phase consistency checking. Conceptually that’s
a bit unsatisfactory. If the PCP is so fundamental, it should be
applicable without the global value exception. But in the example as
given this does not hold since both assert
and program
call
assertImpl
with a splice but no quote.
However, one could argue that the example is really missing an important aspect: The macro library has to be compiled in a phase prior to the program using it, but in the code above, macro and program are defined together. A more accurate view of macros would be to have the user program be in a phase after the macro definitions, reflecting the fact that macros have to be defined and compiled before they are used. Hence, conceptually the program part should be treated by the compiler as if it was quoted:
val program = '{
val x = 1
~Macros.assertImpl('(x != 0))
}
If program
is treated as a quoted expression, the call to
Macro.assertImpl
becomes phase correct even if macro library and
program are conceptualized as local definitions.
But what about the call from assert
to assertImpl
? Here, we need a
tweak of the typing rules. An inline function such as assert
that
contains a splice operation outside an enclosing quote is called a
macro. Macros are supposed to be expanded in a subsequent phase,
i.e. in a quoted context. Therefore, they are also type checked as if
they were in a quoted context. For instance, the definition of
assert
is typechecked as if it appeared inside quotes. This makes
the call from assert
to assertImpl
phasecorrect, even if we
assume that both definitions are local.
The inline
modifier is used to declare a val
that is
either a constant or is a parameter that will be a constant when instantiated. This
aspect is also important for macro expansion. To illustrate this,
consider an implementation of the power
function that makes use of a
statically known exponent:
inline def power(inline n: Int, x: Double) = ~powerCode(n, '(x))
private def powerCode(n: Int, x: Expr[Double]): Expr[Double] =
if (n == 0) '(1.0)
else if (n == 1) x
else if (n % 2 == 0) '{ val y = ~x * ~x; ~powerCode(n / 2, '(y)) }
else '{ ~x * ~powerCode(n  1, x) }
The reference to n
as an argument in ~powerCode(n, '(x))
is not
phaseconsistent, since n
appears in a splice without an enclosing
quote. Normally that would be a problem because it means that we need
the value of n
at compile time, which is not available for general
parameters. But since n
is a inline parameter of a macro, we know
that at the macro’s expansion point n
will be instantiated to a
constant, so the value of n
will in fact be known at this
point. To reflect this, we loosen the phase consistency requirements
as follows:
 If
x
is a inline value (or a inline parameter of an inline function) of type Boolean, Byte, Short, Int, Long, Float, Double, Char or String, it can be accessed in all contexts where the number of splices minus the number of quotes between use and definition is either 0 or 1.
Relationship with Staging
The framework expresses at the same time compiletime metaprogramming and staging. The phase in which code is run is determined by the difference between the number of splice scopes and quote scopes in which it is embedded.

If there are more splices than quotes, the code is run at "compiletime" i.e. as a macro. In the general case, this means running an interpreter that evaluates the code, which is represented as a typed abstract syntax tree. The interpreter can fall back to reflective calls when evaluating an application of a previously compiled method. If the splice excess is more than one, it would mean that a macro’s implementation code (as opposed to the code it expands to) invokes other macros. If macros are realized by interpretation, this would lead to towers of interpreters, where the first interpreter would itself interpret an interpreter code that possibly interprets another interpreter and so on.

If the number of splices equals the number of quotes, the code is compiled and run as usual.
 If the number of quotes exceeds the number of splices, the code is staged. That is, it produces a typed abstract syntax tree or type structure at runtime. A quote excess of more than one corresponds to multistaged programming.
Providing an interpreter for the full language is quite difficult, and it is even more difficult to make that interpreter run efficiently. So we currently impose the following restrictions on the use of splices.

A toplevel splice must appear in an inline method (turning that method into a macro)

The splice must call a previously compiled method passing quoted arguments, constant arguments or inline arguments.

Splices inside splices (but no intervening quotes) are not allowed.
 A macro method is effectively final and it may override no other method.
The framework as discussed so far allows code to be staged, i.e. be prepared
to be executed at a later stage. To run that code, there is another method
in class Expr
called run
. Note that ~
and run
both map from Expr[T]
to T
but only ~
is subject to the PCP, whereas run
is just a normal method.
sealed abstract class Expr[T] {
def unary_~: T
def run(implicit runner: Runner[T]): T // run staged code
def show(implicit runner: Runner[T]): String // show staged code
}
Limitations to Splicing
Quotes and splices are duals as far as the PCP is concerned. But there is an additional restriction that needs to be imposed on splices to guarantee soundness: code in splices must be free of side effects. The restriction prevents code like this:
var x: Expr[T]
'{ (y: T) => ~{ x = '(y); 1 } }
This code, if it was accepted, would "extrude" a reference to a quoted variable y
from its scope.
This means we an subsequently access a variable outside the scope where it is defined, which is
likely problematic. The code is clearly phase consistent, so we cannot use PCP to
rule it out. Instead we postulate a future effect system that can guarantee that splices
are pure. In the absence of such a system we simply demand that spliced expressions are
pure by convention, and allow for undefined compiler behavior if they are not. This is analogous
to the status of pattern guards in Scala, which are also required, but not verified, to be pure.
There is also a problem with run
in splices. Consider the following expression:
'{ (x: Int) => ~{ {'(x)}.run; 1 } }
This is again phase correct, but will lead us into trouble. Indeed, evaluating the splice will reduce the
expression {'(x)}.run
to x
. But then the result
'{ (x: Int) => ~{ x; 1 } }
is no longer phase correct. To prevent this soundness hole it seems easiest to classify run
as a sideeffecting
operation. It would thus be prevented from appearing in splices. In a base language with sideeffects we'd have to
do this anyway: Since run
runs arbitrary code it can always produce a side effect if the code it runs produces one.
The Liftable
typeclass
Consider the following implementation of a staged interpreter that implements a compiler through staging.
import scala.quoted._
enum Exp {
case Num(n: Int)
case Plus(e1: Exp, e2: Exp)
case Var(x: String)
case Let(x: String, e: Exp, in: Exp)
}
The interpreted language consists of numbers Num
, addition Plus
, and variables
Var
which are bound by Let
. Here are two sample expressions in the language:
val exp = Plus(Plus(Num(2), Var("x")), Num(4))
val letExp = Let("x", Num(3), exp)
Here’s a compiler that maps an expression given in the interpreted
language to quoted Scala code of type Expr[Int]
.
The compiler takes an environment that maps variable names to Scala Expr
s.
def compile(e: Exp, env: Map[String, Expr[Int]]): Expr[Int] = e match {
case Num(n) =>
n.toExpr
case Plus(e1, e2) =>
'(~compile(e1, env) + ~compile(e2, env))
case Var(x) =>
env(x)
case Let(x, e, body) =>
'{ val y = ~compile(e, env); ~compile(body, env + (x > '(y))) }
}
Running compile(letExp, Map())
would yield the following Scala code:
'{ val y = 3; (2 + y) + 4 }
The body of the first clause, case Num(n) => n.toExpr
, looks suspicious. n
is declared as an Int
, yet it is conveted to an Expr[Int]
with toExpr
.
Shouldn’t n
be quoted? In fact this would not
work since replacing n
by 'n
in the clause would not be phase
correct.
What happens instead "under the hood" is an extension method toExpr
is added: n.toExpr
is expanded to new scala.quoted.LiftExprOps(n).toExpr
. The toExpr
extension
is defined in the companion object of class Liftable
as follows:
package object quoted {
implicit class LiftExprOps[T](val x: T) extends AnyVal {
def toExpr(implicit ev: Liftable[T]): Expr[T] = ev.toExpr(x)
}
}
The extension says that values of types implementing the Liftable
type class can be
converted ("lifted") to Expr
values using toExpr
when scala.quoted._
is imported.
Dotty comes with instance definitions of Liftable
for
several types including Boolean
, String
, and all primitive number
types. For example, Int
values can be converted to Expr[Int]
values by wrapping the value in a Literal
tree node. This makes use
of the underlying tree representation in the compiler for
efficiency. But the Liftable
instances are nevertheless not "magic"
in the sense that they could all be defined in a user program without
knowing anything about the representation of Expr
trees. For
instance, here is a possible instance of Liftable[Boolean]
:
implicit def BooleanIsLiftable: Liftable[Boolean] = new {
implicit def toExpr(b: Boolean) = if (b) '(true) else '(false)
}
Once we can lift bits, we can work our way up. For instance, here is a
possible implementation of Liftable[Int]
that does not use the underlying
tree machinery:
implicit def IntIsLiftable: Liftable[Int] = new {
def toExpr(n: Int): Expr[Int] = n match {
case Int.MinValue => '(Int.MinValue)
case _ if n < 0 => '((~toExpr(n)))
case 0 => '(0)
case _ if n % 2 == 0 => '(~toExpr(n / 2) * 2)
case _ => '(~toExpr(n / 2) * 2 + 1)
}
}
Since Liftable
is a type class, its instances can be conditional. For example,
a List
is liftable if its element type is:
implicit def ListIsLiftable[T: Liftable]: Liftable[List[T]] = new {
def toExpr(xs: List[T]): Expr[List[T]] = xs match {
case x :: xs1 => '(~x.toExpr :: ~toExpr(xs1))
case Nil => '(Nil: List[T])
}
}
In the end, Liftable
resembles very much a serialization
framework. Like the latter it can be derived systematically for all
collections, case classes and enums. Note also that the implicit synthesis
of "typetag" values of type Type[T]
is essentially the typelevel
analogue of lifting.
Using lifting, we can now give the missing definition of showExpr
in the introductory example:
def showExpr[T](expr: Expr[T]): Expr[String] = {
val code = expr.show
code.toExpr
}
That is, the showExpr
method converts its Expr
argument to a string (code
), and lifts
the result back to an Expr[String]
using the implicit toExpr
conversion.
Implementation
Syntax changes
A splice ~e
on an expression of type Expr[T]
is a normal prefix
operator. To make it work as a type operator on Type[T]
as well, we
need a syntax change that introduces prefix operators as types.
SimpleType ::= ...
[‘’  ‘+’  ‘~’  ‘!’] StableId
Analogously to the situation with expressions, a prefix type operator
such as ~ e
is treated as a shorthand for the type e.unary_~
.
Quotes are supported by introducing new tokens '(
, '{
, and '[
and adding quoted variants '(...)
, '{...}
and '[...]
to the
SimpleExpr
productions.
SimpleExpr ::= ...
 ‘'{’ BlockExprContents ‘}’
 ‘'’ ‘(’ ExprsInParens ‘)’
 ‘'’ ‘[’ Type ‘]’
Syntax changes are given relative to the Dotty reference grammar.
An alternative syntax would treat '
as a separate operator. This
would be attractive since it enables quoting single identifiers as
e.g. 'x
instead of '(x)
. But it would clash with symbol
literals. So it could be done only if symbol literals were abolished.
Implementation in dotc
Quotes and splices are primitive forms in the generated abstract
syntax trees. They are eliminated in an expansion phase
Staging
. This phase runs after typing and pickling.
Macroexpansion works outsidein. If the outermost scope is a splice, the spliced AST will be evaluated in an interpreter. A call to a previously compiled method can be implemented as a reflective call to that method. With the restrictions on splices that are currently in place that’s all that’s needed. We might allow more interpretation in splices in the future, which would allow us to loosen the restriction. Quotes in spliced, interpreted code are kept as they are, after splices nested in the quotes are expanded.
If the outermost scope is a quote, we need to generate code that constructs the quoted tree at runtime. We implement this by serializing the tree as a Tasty structure, which is stored in a string literal. At runtime, an unpickler method is called to deserialize the string into a tree.
Splices inside quoted code insert the spliced tree as is, after expanding any quotes in the spliced code recursively.
Formalization
The phase consistency principle can be formalized in a calculus that extends simplytyped lambda calculus with quotes and splices.
Syntax
The syntax of terms, values, and types is given as follows:
Terms t ::= x variable
(x: T) => t lambda
t t application
't quote
~t splice
Values v ::= (x: T) => t lambda
'u quote
Simple terms u ::= x  (x: T) => u  u u  't
Types T ::= A base type
T > T function type
expr T quoted
Typing rules are formulated using a stack of environments
Es
. Individual environments E
consist as usual of variable
bindings x: T
. Environments can be combined using the two
combinators '
and ~
.
Environment E ::= () empty
E, x: T
Env. stack Es ::= () empty
E simple
Es * Es combined
Separator * ::= '
~
The two environment combinators are both associative with left and
right identity ()
.
Operational semantics:
We define a small step reduction relation >
with the following rules:
((x: T) => t) v > [x := v]t
~('u) > u
t1 > t2

e[t1] > e[t2]
The first rule is standard callbyvalue betareduction. The second
rule says that splice and quotes cancel each other out. The third rule
is a context rule; it says that reduction is allowed in the hole [ ]
position of an evaluation context. Evaluation contexts e
and
splice evaluation context e_s
are defined syntactically as follows:
Eval context e ::= [ ]  e t  v e  'e_s[~e]
Splice context e_s ::= [ ]  (x: T) => e_s  e_s t  u e_s
Typing rules
Typing judgments are of the form Es  t: T
. There are two
substructural rules which express the fact that quotes and splices
cancel each other out:
Es1 * Es2  t: T

Es1 ~ E1 ' E2 * Es2  t: T
Es1 * Es2  t: T

Es1 ' E1 ~ E2 * Es2  t: T
The lambda calculus fragment of the rules is standard, except that we use a stack of environments. The rules only interact with the topmost environment of the stack.
x: T in E

Es * E  x: T
Es * E, x: T1  t: T2

Es * E  (x: T1) => t: T > T2
Es  t1: T2 > T Es  t2: T2

Es  t1 t2: T
The rules for quotes and splices map between expr T
and T
by trading '
and ~
between
environments and terms.
Es ~ ()  t: expr T

Es  ~t: T
Es ' ()  t: T

Es  't: expr T
The meta theory of a slightly simplified variant 2stage variant of this calculus is studied separately.
Going Further
The metaprogramming framework as presented and currently implemented is quite restrictive
in that it does not allow for the inspection of quoted expressions and
types. It’s possible to work around this by providing all necessary
information as normal, unquoted inline parameters. But we would gain
more flexibility by allowing for the inspection of quoted code with
pattern matching. This opens new possibilities. For instance, here is a
version of power
that generates the multiplications directly if the
exponent is statically known and falls back to the dynamic
implementation of power otherwise.
inline def power(n: Int, x: Double): Double = ~{
'(n) match {
case Constant(n1) => powerCode(n1, '(x))
case _ => '{ dynamicPower(n, x) }
}
}
private def dynamicPower(n: Int, x: Double): Double =
if (n == 0) 1.0
else if (n % 2 == 0) dynamicPower(n / 2, x * x)
else x * dynamicPower(n  1, x)
This assumes a Constant
extractor that maps tree nodes representing
constants to their values.
With the right extractors the "AsFunction" operation that maps expressions over functions to functions over expressions can be implemented in user code:
implicit class AsFunction[T, U](f: Expr[T => U]) extends AnyVal {
def apply(x: Expr[T]): Expr[U] =
f match {
case Lambda(g) => g(x)
case _ => '((~f)(~x))
}
This assumes an extractor
object Lambda {
def unapply[T, U](x: Expr[T => U]): Option[Expr[T] => Expr[U]]
}
Once we allow inspection of code via extractors, it’s tempting to also
add constructors that create typed trees directly without going
through quotes. Most likely, those constructors would work over Expr
types which lack a known type argument. For instance, an Apply
constructor could be typed as follows:
def Apply(fn: Expr[_], args: List[Expr[_]]): Expr[_]
This would allow constructing applications from lists of arguments
without having to match the arguments onebyone with the
corresponding formal parameter types of the function. We then need "at
the end" a method to convert an Expr[_]
to an Expr[T]
where T
is
given from the outside. E.g. if code
yields a Expr[_]
, then
code.atType[T]
yields an Expr[T]
. The atType
method has to be
implemented as a primitive; it would check that the computed type
structure of Expr
is a subtype of the type structure representing
T
.
Before going down that route, we should evaluate in detail the tradeoffs it presents. Constructing trees that are only verified a posteriori to be type correct loses a lot of guidance for constructing the right trees. So we should wait with this addition until we have more usecases that help us decide whether the loss in typesafety is worth the gain in flexibility. In this context, it seems that deconstructing types is less errorprone than deconstructing terms, so one might also envisage a solution that allows the former but not the latter.
Conclusion
Metaprogramming has a reputation of being difficult and confusing.
But with explicit Expr/Type
types and quotes and splices it can become
downright pleasant. A simple strategy first defines the underlying quoted or unquoted
values using Expr
and Type
and then inserts quotes and splices to make the types
line up. Phase consistency is at the same time a great guideline
where to insert a splice or a quote and a vital sanity check that
the result makes sense.