In the crowded space of Common Lisp HTML generators, SPINNERET occupies the following coordinates:

  • Modern. Targets HTML5. Does not treat XML and HTML as the same problem. Assumes you will be serving your documents as UTF-8.

  • Composable. Makes it easy to refactor HTML generation into separate functions and macros.

  • Pretty. Treats HTML as a document format, not a serialization. Output is idiomatic and readable, following the coding style of the HTML5 specification.

  • Aggressive. If something can be interpreted as HTML, then it will be, meaning that some Lisp forms can't be mixed with HTML syntax. In the trade-off between 90% convenience and 10% correctness SPINNERET is on the side of convenience.

  • Bilingual. Spinneret has the same semantics in Lisp and Parenscript.

HTML generation with SPINNERET looks like this:

 (in-package #:spinneret)

 (defparameter *shopping-list*
   '("Atmospheric ponds"
     "Electric gumption socks"
     "Mrs. Leland's embyronic television combustion"
     "Savage gymnatic aggressors"
     "Pharmaceutical pianos"
     "Intravenous retribution champions"))

 (defparameter *user-name* "John Q. Lisper")

 (defparameter *last-login* "12th Never")

 (defmacro with-page ((&key title) &body body)
         (:title ,title))
        (:body ,@body))))

 (defun shopping-list ()
   (with-page (:title "Home page")
      (:h1 "Home page"))
      ("~A, here is *your* shopping list: " *user-name*)
      (:ol (dolist (item *shopping-list*)
             (:li (1+ (random 10)) item))))
     (:footer ("Last login: ~A" *last-login*))))

Which produces:

 <!DOCTYPE html>
 <html lang=en>
   <meta charset=UTF-8>
   <title>Home page</title>
    <h1>Home page</h1>
    John Q. Lisper, here is <em>your</em> shopping list:
     <li>10 Atmospheric ponds
     <li>6 Electric gumption socks
     <li>4 Mrs. Leland's embyronic television combustion
     <li>9 Savage gymnatic aggressors
     <li>6 Pharmaceutical pianos
     <li>9 Intravenous retribution champions
    Last login: 12th Never

(Pretty-printing is pretty fast, but SPINNERET obeys *print-pretty* should you want to turn it off.)


The rules for WITH-HTML are these:

  • All generated forms write to *html*.

  • A keyword in function position is interpreted as a tag name. If the name is not valid as a tag, it is ignored.

Certain keywords are recognized as pseudo-tags and given special treatment:


The value of the LANG attribute of HTML is controlled by *html-lang*; the value of the meta charset attribute is controlled by *html-charset*.

Constant classes and ids can be specified with a selector-like syntax. E.g.:

 (:div#wrapper (:div.section ...))
 ? (:div :id "wrapper" (:div :class "section" ...))
  • Keyword-value pairs following a tag are interpreted as attributes. HTML syntax may not be used in attribute values. Attributes with nil values are omitted from the output. Boolean attributes with non-nil values are minimized.

Duplicate attributes are handled like duplicate keyword arguments: all values are evaluated, but only the leftmost value is used. The exception is the handling of tokenized attributes, such as :CLASS or :REL. The class of a tag is the union of all its :CLASS arguments.

The argument :DATASET introduces a list of :DATA-FOO arguments:

 (:p :dataset (:duck (dolomphious) :fish 'fizzgigious
                     :spoon "runcible"))
 ? (:p :data-duck (dolomphious) :data-fish 'fizzgigious
       :data-spoon "runcible")

For flexibility, even at the cost of efficiency, the argument :ATTRS introduces a form to evaluate at run time for a plist of extra attributes and values.

  • Forms after the attributes are treated as arguments. Each non-nil (primary) value returned by an argument to a tag is written to the stream by HTML, a generic function on which you can define your own methods. By default only literal arguments are printed. Literal arguments are strings, characters, numbers and symbols beside NIL.

WITH-HTML-STRING is like WITH-HTML, but intercepts the generated HTML at run time and returns a string.


If the additional system spinneret/cl-markdown is loaded, then a string in function position is first compiled as Markdown (using CL-MARKDOWN), then passed to format as a control string and applied to its arguments.

This is useful for inline formatting, like links, where sexps would be clumsy:

 ("Here is some copy, with [a link](~a)" link))

  (:span "Here is some copy, with "
    (:a :href link "a link.")))


Sometimes it is useful for a piece of HTML-generating code to know where in the document it appears. You might, for example, want to define a tabulate that prints list-of-lists as rows of cells, but only prints the surrounding <table></table> if it is not already within a table. The function get-html-path returns a list of open tags, from latest to earliest. Usually it will look something like

  (get-html-path) ;-> '(:table :section :body :html)

Thus `tabulate' could be written

 (defun tabulate (&rest rows)
     (flet ((tabulate ()
              (loop for row in rows do
                (:tr (loop for cell in row do
                  (:td cell))))))
       (if (find :table (get-html-path))
           (:table (:tbody (tabulate)))))))

Note that get-html-path returns a freshly-consed list each time it is called.


The stumbling block for all sexp-based HTML generators is order of evaluation. It's tempting to write something like this:

 ;; Doesn't work
 (defun field (control)
   (with-html (:p control)))

 (defun input (default &key name label (type "text"))
     (:label :for name label)
     (:input :name name :id name :type type :value default)))

But it won't work: in (field (input "Default" :name "why" :label "Reason")), (input) gets evaluated before (field), and the HTML is printed inside-out.

Macros do work:

 (defmacro field (control)
   `(with-html (:p ,control)))

 (defmacro input (name label &key (type "text"))
      (:label :for ,name ,label)
      (:input :name ,name :id ,name :type ,type)))

But we can do better than this. Spinneret provides a macro-writing macro, deftag, which lets you refactor HTML without hiding it.

 (deftag field (control attrs)
  `(:p ,@attrs ,@control))

 (deftag input (default attrs &key name label (type "text"))
   (once-only (name)
        (:label :for ,name ,label)
        (:input :name ,name :id ,name :type ,type
          :value (progn ,@default)))))

A macro defined using deftag takes its arguments just like an HTML element. Instead of

(input "Default" :name "why" :label "Reason") ; defmacro

You write

(input :name "why" :label "Reason" "Default") ; deftag

The macro re-arranges the arguments so they can be bound to an ordinary lambda list, like the one above: the body of the tag is bound to the first argument, and matching attributes are bound to keywords. Multiple :class arguments, :dataset, and other shorthands are handled exactly as in the usual HTML syntax.

But the great advantage of deftag is how it handles attributes which are not bound to keywords. In the definition of input using deftag, you see that the attrs catch-all argument is spliced into the call to :input. This means that any unhandled attributes pass through to the actual input element.

(input :name "why" :label "Reason" :required t :class "special" "Default")
=> <label for=why>Reason</label>
   <input class=special name=why id=why type=text required value=Default>

In effect, input extends the :input tag, almost like a subclass. This is a very idiomatic and expressive way of building abstractions over HTML.

(SPINNERET used to provide a more elaborate way of building HTML abstractions, deftemplate, but deftag is simpler and more useful.)


The semantics of SPINNERET in Parenscript are almost the same. There is no with-html-string, and with-html returns a DocumentFragment.

If Markdown support is enabled, strings in function position are still parsed as Markdown, but supplying arguments triggers an error (since Parenscript does not have format).

get-html-path is not implemented for Parenscript.


SPINNERET does not do document validation, but it does warn, at compile time, about invalid tags and attributes.

Although HTML5 does include a mechanism for application-specific attributes (the data- prefix), some client-side frameworks choose to employ their own prefixes instead. You can disable validation for a given prefix by adding it to *unvalidated-attribute-prefixes*.

(pushnew "ng-" *unvalidated-attribute-prefixes* :test #?equal)
Paul M. Rodriguez <pmr@ruricolist.com>