6.2 A more involved example

Let’s illustrate some more involved uses of defsystem via a slightly convoluted example:

(in-package :asdf-user)

(defsystem "foo"
  :version (:read-file-form "variables" :at (3 2))
  :components
  ((:file "package")
   (:file "variables" :depends-on ("package"))
   (:module "mod"
     :depends-on ("package")
     :serial t
     :components ((:file "utils")
                  (:file "reader")
                  (:file "cooker")
                  (:static-file "data.raw"))
     :output-files (compile-op (o c) (list "data.cooked"))
     :perform (compile-op :after (o c)
        (cook-data
         :in (component-pathname (find-component c "data.raw"))
         :out (first (output-files o c)))))
   (:file "foo" :depends-on ("mod"))))

(defmethod action-description
    ((o compile-op) (c (eql (find-component "foo" "mod"))))
  "cooking data")

Here are some notes about this example:

• The main thing this file does is define a system foo. It also contains other Lisp forms, which we’ll examine below.
• Besides Lisp source files, this system contains a :module component named "mod", which is a collection of three Lisp source files utils.lisp, reader.lisp, cooker.lisp and data.raw
• Note that the :static-file does not have an implicit file type, unlike the Lisp source files.
• This files will be located in a subdirectory of the main code directory named mod/ (this location could have been overridden to be in the same directory, or in a different subdirectory; see the discussion of the :pathname option in The defsystem grammar).
• The :serial t says that each sub-component of mod depends on the previous components, so that cooker.lisp depends-on utils.lisp, which depends-on reader.lisp. Also data.raw depends on all of them, but that doesn’t matter since it’s a static file; on the other hand, if it appeared first, then all the Lisp files would be recompiled when the data is modified, which is probably not what is desired in this case.
• The method-form tokens provide a shorthand for defining methods on particular components. This part

       :output-files (compile-op (o c) (list "data.cooked"))
       :perform (compile-op :after (o c)
          (cook-data
           :in (component-pathname (find-component c "data.raw"))
           :out (first (output-files o c))))
  

  has the effect of

  (defmethod output-files ((o compile-op) (c (eql ...)))
    (list "data.cooked"))
  (defmethod perform :after ((o compile-op) (c (eql ...)))
    (cook-data
     :in (component-pathname (find-component c "data.raw"))
     :out (first (output-files o c))))
  

  where ... is the component in question. In this case ... would expand to something like

  (find-component "foo" "mod")
  

  For more details on the syntax of such forms, see The defsystem grammar. For more details on what these methods do, see Operations in The object model of ASDF.

• There is an additional defmethod with a similar effect, because ASDF (as of ASDF 3.1.5) fails to accept inline-methods as above for action-description, instead only supporting the deprecated explain interface.
• In this case, these methods describe how this module defines code that it then uses to cook some data.
• Importantly, ASDF is told about the input and output files used by the data cooker, and to make sure everyone agrees, the cooking function explicitly uses ASDF to access pathnames to the input and output data.
• The file starts with a form (in-package :asdf-user), but it is actually redundant, not necessary and not recommended. But yet more complex cases (also not recommended) may usefully use an in-package form.
• Indeed, ASDF does not load .asd files simply with cl:load, and neither should you. You should let ASDF find and load them when you operate on systems. If you somehow must load a .asd file, use the same function asdf:load-asd that ASDF uses. Among other things, it already binds the *package* to asdf-user. Recent versions of SLIME (2013-02 and later) know to do that when you C-c C-k when you use the slime-asdf contrib.
• You shouldn’t use an in-package form if you’re keeping things simple. You should only use in-package (and before it, a defpackage) when you’re going to define new classes, functions, variables, macros, etc., in the .asd file, and want to thereby avoid name clashes. Manuals for old versions of ASDF recommended use of such an idiom in .asd files, but as of ASDF 3, we recommend that you don’t do that anymore, and instead define any ASDF extensions in their own system, on which you can then declare a dependency using :defsystem-depends-on. See The defsystem grammar.
• More generally, you can always rely on symbols from packages asdf, common-lisp and uiop being available in .asd files — most importantly including defsystem. It is therefore redundant and in bad taste to use a package-prefixed asdf:defsystem symbol in a .asd file. Just use (defsystem ...). Only package-prefix it when somehow dynamically generating system definitions from a package that doesn’t already use the ASDF package.
• asdf-user is actually only available starting since ASDF 3, but then again, ASDF 1 and 2 did crazy things with packages that ASDF 3 has stopped doing⁹, and since all implementations provide ASDF 3, you shouldn’t care about compatibility with ASDF 2. We do not support ASDF 2 anymore, and we recommend that neither should you.
• Starting with ASDF 3.1, asdf-user uses uiop, whereas in earlier variants of ASDF 3 it only used uiop/package. We recommend you either prefix use of UIOP functions with the package prefix uiop:, or make sure your system :depends-on ((:version "asdf" "3.1.2")) or has a #-asdf3.1 (error "MY-SYSTEM requires ASDF 3.1.2").
• Finally, we elided most metadata, but showed how you can have ASDF automatically extract the system’s version from a source file. In this case, the 3rd subform of the 4th form (note that Lisp uses 0-based indexing, English uses 1-based indexing). Presumably, the 4th form looks like (defparameter *foo-version* "5.6.7").