Racket – macros – 7

ror

Continuo, sono qui: [doc]/guide/macros.html, in particolare qui: [doc]/guide/proc-macros.html.

Livelli di fase

A phase can be thought of as a way to separate computations in a pipeline of processes where one produces code that is used by the next. (E.g., a pipeline that consists of a preprocessor process, a compiler, and an assembler.)

Imagine starting two Racket processes for this purpose. If you ignore inter-process communication channels like sockets and files, the processes will have no way to share anything other than the text that is piped from the standard output of one process into the standard input of the other. Similarly, Racket effectively allows multiple invocations of a module to exist in the same process but separated by phase. Racket enforces separation of such phases, where different phases cannot communicate in any way other than via the protocol of macro expansion, where the output of one phases is the code used in the next.

Fasi e collegamenti

Every binding of an identifier exists in a particular phase. The link between a binding and its phase is represented by an integer phase level. Phase level 0 is the phase used for “plain” (or “runtime”) definitions, so

(define age 5)

adds a binding for age into phase level 0. The identifier age can be defined at a higher phase level using begin-for-syntax:

(begin-for-syntax
  (define age 5))

With a single begin-for-syntax wrapper, age is defined at phase level 1. We can easily mix these two definitions in the same module or in a top-level namespace, and there is no clash between the two ages that are defined at different phase levels:

m40

The age binding at phase level 0 has a value of 3, and the age binding at phase level 1 has a value of 9.

Syntax objects capture binding information as a first-class value. Thus,

#'age

is a syntax object that represents the age binding—but since there are two ages (one at phase level 0 and one at phase level 1), which one does it capture? In fact, Racket imbues #'age with lexical information for all phase levels, so the answer is that #'age captures both.

The relevant binding of age captured by #'age is determined when #'age is eventually used. As an example, we bind #'age to a pattern variable so we can use it in a template, and then we evaluate the template:

m41

Note: We use eval here to demonstrate phases, but see Reflection and Dynamic Evaluation [reflections e valutazione dinamica] for caveats about eval.

The result is 3 because age is used at phase 0 level. We can try again with the use of age inside begin-for-syntax:

m42

In this case, the answer is 9, because we are using age at phase level 1 instead of 0 (i.e., begin-for-syntax evaluates its expressions at phase level 1). So, you can see that we started with the same syntax object, #'age, and we were able to use it in two different ways: at phase level 0 and at phase level 1.

A syntax object has a lexical context from the moment it first exists. A syntax object that is provided from a module retains its lexical context, and so it references bindings in the context of its source module, not the context of its use. The following example defines button at phase level 0 and binds it to 0, while see-button binds the syntax object for button in module a:

m43

The result of the m macro is the value of see-button, which is #'button with the lexical context of the a module. Even though there is another button in b, the second button will not confuse Racket, because the lexical context of #'button (the value bound to see-button) is a.

Note that see-button is bound at phase level 1 by virtue of defining it with define-for-syntax. Phase level 1 is needed because m is a macro, so its body executes at one phase higher than the context of its definition. Since m is defined at phase level 0, its body is at phase level 1, so any bindings referenced by the body must be at phase level 1.

Fasi e moduli

A phase level is a module-relative concept. When importing from another module via require, Racket lets us shift imported bindings to a phase level that is different from the original one:

(require "a.rkt")                ; import with no phase shift
(require (for-syntax "a.rkt"))   ; shift phase by +1
(require (for-template "a.rkt")) ; shift phase by -1
(require (for-meta 5 "a.rkt"))   ; shift phase by +5

Nota perso: qui farò pasticci, lo sento 😦

That is, using for-syntax in require means that all of the bindings from that module will have their phase levels increased by one. A binding that is defined at phase level 0 and imported with for-syntax becomes a phase-level 1 binding:

m44

Let’s see what happens if we try to create a binding for the #'button syntax object at phase level 0:

m45

Now both button and see-button are defined at phase 0. The lexical context of #'button will know that there is a binding for button at phase 0. In fact, it seems like things are working just fine if we try to eval see-button:

m46

Now, let’s use see-button in a macro:

m47

Clearly, see-button is not defined at phase level 1, so we cannot refer to it inside the macro body. Let’s try to use see-button in another module by putting the button definitions in a module and importing it at phase level 1. Then, we will get see-button at phase level 1:

m48

Racket says that button is unbound now! When a is imported at phase level 1, we have the following bindings:

button     at phase level 1
see-button at phase level 1

So the macro m can see a binding for see-button at phase level 1 and will return the #'button syntax object, which refers to button binding at phase level 1. But the use of m is at phase level 0, and there is no button at phase level 0 in b. That is why see-button needs to be bound at phase level 1, as in the original a. In the original b, then, we have the following bindings:

button     at phase level 0
see-button at phase level 1

In this scenario, we can use see-button in the macro, since see-button is bound at phase level 1. When the macro expands, it will refer to a button binding at phase level 0.

Defining see-button with (define see-button #'button) isn’t inherently wrong; it depends on how we intend to use see-button. For example, we can arrange for m to sensibly use see-button because it puts it in a phase level 1 context using begin-for-syntax:

m49

In this case, module b has both button and see-button bound at phase level 1. The expansion of the macro is

(begin-for-syntax
  (displayln button))

which works, because button is bound at phase level 1.

Now, you might try to cheat the phase system by importing a at both phase level 0 and phase level 1. Then you would have the following bindings

button     at phase level 0
see-button at phase level 0
button     at phase level 1
see-button at phase level 1

You might expect now that see-button in a macro would work, but it doesn’t:

m50

The see-button inside the m macro comes from the (for-syntax 'a) import. For the macro to work, there must be a button at phase 0 bound, and there is such a binding implied by (require 'a). However, (require 'a) and (require (for-syntax 'a)) are different instantiations of the same module. The see-button at phase 1 only refers to the button at phase level 1, not the button bound at phase 0 from a different instantiation—even from the same source module.

Mismatches like the one above can show up when a macro tries to match literal bindings—using syntax-case or syntax-parse.

m51

In this example, make is being used in y at phase level 2, and it returns the #'button syntax object—which refers to button bound at phase level 0 inside x and at phase level 2 in y from (for-meta 2 'x). The process macro is imported at phase level 1 from (for-meta 1 'x), and it knows that button should be bound at phase level 1. When the syntax-parse is executed inside process, it is looking for button bound at phase level 1 but it sees only a phase level 2 binding and doesn’t match.

To fix the example, we can provide make at phase level 1 relative to x, and then we import it at phase level 1 in y:

m52

Continua 😀 però chissà perché mi salta in mente questo:

geb

:mrgreen:

Posta un commento o usa questo indirizzo per il trackback.

Trackback

  • Racket – macros – 8 | Ok, panico su 1 ottobre 2015 alle 09:46

    […] Continuo ma oggi finisco il capitolo, questo [doc]/guide/macros.html, dove sono qui [doc]/guide/proc-macros.html. […]

Rispondi

Inserisci i tuoi dati qui sotto o clicca su un'icona per effettuare l'accesso:

Logo di WordPress.com

Stai commentando usando il tuo account WordPress.com. Chiudi sessione /  Modifica )

Google photo

Stai commentando usando il tuo account Google. Chiudi sessione /  Modifica )

Foto Twitter

Stai commentando usando il tuo account Twitter. Chiudi sessione /  Modifica )

Foto di Facebook

Stai commentando usando il tuo account Facebook. Chiudi sessione /  Modifica )

Connessione a %s...

Questo sito utilizza Akismet per ridurre lo spam. Scopri come vengono elaborati i dati derivati dai commenti.

%d blogger hanno fatto clic su Mi Piace per questo: