In this chapter and the next you'll take a look under the hood of the FOO HTML generator that you've been using in the past few chapters. FOO is an example of a kind of programming that's quite common in Common Lisp and relatively uncommon in non-Lisp languages, namely, language-oriented programming. Rather than provide an API built primarily out of functions, classes, and macros, FOO provides language processors for a domain-specific language that you can embed in your Common Lisp programs.
FOO provides two language processors for the same s-expression
language. One is an interpreter that takes a FOO "program" as data
and interprets it to generate HTML. The other is a compiler that
compiles FOO expressions, possibly with embedded Common Lisp code,
into Common Lisp that generates HTML and runs the embedded code. The
interpreter is exposed as the function
emit-html and the
compiler as the macro
html, which you used in previous
In this chapter you'll look at some of the infrastructure shared between the interpreter and the compiler and then at the implementation of the interpreter. In the next chapter, I'll show you how the compiler works.
Designing an embedded language requires two steps: first, design the language that'll allow you to express the things you want to express, and second, implement a processor, or processors, that accepts a "program" in that language and either performs the actions indicated by the program or translates the program into Common Lisp code that'll perform equivalent behaviors.
So, step one is to design the HTML-generating language. The key to designing a good domain-specific language is to strike the right balance between expressiveness and concision. For instance, a highly expressive but not very concise "language" for generating HTML is the language of literal HTML strings. The legal "forms" of this language are strings containing literal HTML. Language processors for this "language" could process such forms by simply emitting them as-is.
(defvar *html-output* *standard-output*) (defun emit-html (html) "An interpreter for the literal HTML language." (write-sequence html *html-output*)) (defmacro html (html) "A compiler for the literal HTML language." `(write-sequence ,html *html-output*))
This "language" is highly expressive since it can express any HTML you could possibly want to generate.1 On the other hand, this language doesn't win a lot of points for its concision because it gives you zero compression--its input is its output.
To design a language that gives you some useful compression without sacrificing too much expressiveness, you need to identify the details of the output that are either redundant or uninteresting. You can then make those aspects of the output implicit in the semantics of the language.
For instance, because of the structure of HTML, every opening tag is paired with a matching closing tag.2 When you write HTML by hand, you have to write those closing tags, but you can improve the concision of your HTML-generating language by making the closing tags implicit.
Another way you can gain concision at a slight cost in expressiveness is to make the language processors responsible for adding appropriate whitespace between elements--blank lines and indentation. When you're generating HTML programmatically, you typically don't care much about which elements have line breaks before or after them or about whether different elements are indented relative to their parent elements. Letting the language processor insert whitespace according to some rule means you don't have to worry about it. As it turns out, FOO actually supports two modes--one that uses the minimum amount of whitespace, which allows it to generate extremely efficient code and compact HTML, and another that generates nicely formatted HTML with different elements indented and separated from other elements according to their role.
Another detail that's best moved into the language processor is the
escaping of certain characters that have a special meaning in HTML
&. Obviously, if you generate
HTML by just printing strings to a stream, then it's up to you to
replace any occurrences of those characters in the string with the
appropriate escape sequences,
&. But if the language processor can know which strings
are to be emitted as element data, then it can take care of
automatically escaping those characters for you.
So, enough theory. I'll give you a quick overview of the language implemented by FOO, and then you'll look at the implementation of the two FOO language processors--the interpreter, in this chapter, and the compiler, in the next.
Like Lisp itself, the basic syntax of the FOO language is defined in terms of forms made up of Lisp objects. The language defines how each legal FOO form is translated into HTML.
The simplest FOO forms are self-evaluating Lisp objects such as
strings, numbers, and keyword symbols.3 You'll need a function
tests whether a given object is self-evaluating for FOO's purposes.
(defun self-evaluating-p (form) (and (atom form) (if (symbolp form) (keywordp form) t)))
Objects that satisfy this predicate will be emitted by converting
them to strings with
PRINC-TO-STRING and then escaping any
reserved characters, such as
the value is being emitted as an attribute, the characters
' are also escaped. Thus, you can invoke the
macro on a self-evaluating object to emit it to
(which is initially bound to
*STANDARD-OUTPUT*). Table 30-1
shows how a few different self-evaluating values will be output.
|FOO Form||Generated HTML|
Of course, most HTML consists of tagged elements. The three pieces of
information that describe each element are the tag, a set of
attributes, and a body containing text and/or more HTML elements.
Thus, you need a way to represent these three pieces of information
as Lisp objects, preferably ones that the Lisp reader already knows
how to read.4 If you forget about attributes for a moment, there's an
obvious mapping between Lisp lists and HTML elements: any HTML
element can be represented by a list whose
FIRST is a symbol
where the name is the name of the element's tag and whose
is a list of self-evaluating objects or lists representing other HTML
<p>Foo</p> <==> (:p "Foo") <p><i>Now</i> is the time</p> <==> (:p (:i "Now") " is the time")
Now the only problem is where to squeeze in the attributes. Since most elements have no attributes, it'd be nice if you could use the preceding syntax for elements without attributes. FOO provides two ways to notate elements with attributes. The first is to simply include the attributes in the list immediately following the symbol, alternating keyword symbols naming the attributes and objects representing the attribute value forms. The body of the element starts with the first item in the list that's in a position to be an attribute name and isn't a keyword symbol. Thus:
HTML> (html (:p "foo")) <p>foo</p> NIL HTML> (html (:p "foo " (:i "bar") " baz")) <p>foo <i>bar</i> baz</p> NIL HTML> (html (:p :style "foo" "Foo")) <p style='foo'>Foo</p> NIL HTML> (html (:p :id "x" :style "foo" "Foo")) <p id='x' style='foo'>Foo</p> NIL
For folks who prefer a bit more obvious delineation between the
element's attributes and its body, FOO supports an alternative
syntax: if the first element of a list is itself a list with a
keyword as its first element, then the outer list represents an
HTML element with that keyword indicating the tag, with the
of the nested list as the attributes, and with the
REST of the
outer list as the body. Thus, you could write the previous two
expressions like this:
HTML> (html ((:p :style "foo") "Foo")) <p style='foo'>Foo</p> NIL HTML> (html ((:p :id "x" :style "foo") "Foo")) <p id='x' style='foo'>Foo</p> NIL
The following function tests whether a given object matches either of these syntaxes:
(defun cons-form-p (form &optional (test #'keywordp)) (and (consp form) (or (funcall test (car form)) (and (consp (car form)) (funcall test (caar form))))))
You should parameterize the
test function because later you'll
need to test the same two syntaxes with a slightly different predicate
on the name.
To completely abstract the differences between the two syntax
variants, you can define a function,
takes a form and parses it into three elements, the tag, the
attributes plist, and the body list, returning them as multiple
values. The code that actually evaluates cons forms will use this
function and not have to worry about which syntax was used.
(defun parse-cons-form (sexp) (if (consp (first sexp)) (parse-explicit-attributes-sexp sexp) (parse-implicit-attributes-sexp sexp))) (defun parse-explicit-attributes-sexp (sexp) (destructuring-bind ((tag &rest attributes) &body body) sexp (values tag attributes body))) (defun parse-implicit-attributes-sexp (sexp) (loop with tag = (first sexp) for rest on (rest sexp) by #'cddr while (and (keywordp (first rest)) (second rest)) when (second rest) collect (first rest) into attributes and collect (second rest) into attributes end finally (return (values tag attributes rest))))
Now that you have the basic language specified, you can think about how you're actually going to implement the language processors. How do you get from a series of FOO forms to the desired HTML? As I mentioned previously, you'll be implementing two language processors for FOO: an interpreter that walks a tree of FOO forms and emits the corresponding HTML directly and a compiler that walks a tree and translates it into Common Lisp code that'll emit the same HTML. Both the interpreter and compiler will be built on top of a common foundation of code, which provides support for things such as escaping reserved characters and generating nicely indented output, so it makes sense to start there.
The first bit of the foundation you'll need to lay is the code that
knows how to escape characters with a special meaning in HTML. There
are three such characters, and they must not appear in the text of an
element or in an attribute value; they are
&. In element text or attribute values, these characters must
be replaced with the character reference entities
&. Similarly, in attribute values, the
quotation marks used to delimit the value must be escaped,
". Additionally, any
character can be represented by a numeric character reference entity
consisting of an ampersand, followed by a sharp sign, followed by the
numeric code as a base 10 integer, and followed by a semicolon. These
numeric escapes are sometimes used to embed non-ASCII characters in
The following function accepts a single character and returns a string containing a character reference entity for that character:
(defun escape-char (char) (case char (#\& "&") (#\< "<") (#\> ">") (#\' "'") (#\" """) (t (format nil "&#~d;" (char-code char)))))
You can use this function as the basis for a function,
that takes a string and a sequence of characters and returns a copy of
the first argument with all occurrences of the characters in the
second argument replaced with the corresponding character entity
(defun escape (in to-escape) (flet ((needs-escape-p (char) (find char to-escape))) (with-output-to-string (out) (loop for start = 0 then (1+ pos) for pos = (position-if #'needs-escape-p in :start start) do (write-sequence in out :start start :end pos) when pos do (write-sequence (escape-char (char in pos)) out) while pos))))
You can also define two parameters:
contains the characters you need to escape in normal element data,
*attribute-escapes*, which contains the set of characters
to be escaped in attribute values.
(defparameter *element-escapes* "<>&") (defparameter *attribute-escapes* "<>&\"'")
Here are some examples:
HTML> (escape "foo & bar" *element-escapes*) "foo & bar" HTML> (escape "foo & 'bar'" *element-escapes*) "foo & 'bar'" HTML> (escape "foo & 'bar'" *attribute-escapes*) "foo & 'bar'"
Finally, you'll need a variable,
*escapes*, that will be bound
to the set of characters that need to be escaped. It's initially set
to the value of
*element-escapes*, but when generating
attributes, it will, as you'll see, be rebound to the value of
(defvar *escapes* *element-escapes*)
To handle generating nicely indented output, you can define a class
indenting-printer, which wraps around an output stream, and
functions that use an instance of that class to emit strings to the
stream while keeping track of when it's at the beginning of the line.
The class looks like this:
(defclass indenting-printer () ((out :accessor out :initarg :out) (beginning-of-line-p :accessor beginning-of-line-p :initform t) (indentation :accessor indentation :initform 0) (indenting-p :accessor indenting-p :initform t)))
The main function that operates on
emit, which takes the printer and a string and emits the
string to the printer's output stream, keeping track of when it emits
a newline so it can reset the
(defun emit (ip string) (loop for start = 0 then (1+ pos) for pos = (position #\Newline string :start start) do (emit/no-newlines ip string :start start :end pos) when pos do (emit-newline ip) while pos))
To actually emit the string, it uses the function
emit/no-newlines, which emits any needed indentation, via the
indent-if-necessary, and then writes the string to the
stream. This function can also be called directly by other code to
emit a string that's known not to contain any newlines.
(defun emit/no-newlines (ip string &key (start 0) end) (indent-if-necessary ip) (write-sequence string (out ip) :start start :end end) (unless (zerop (- (or end (length string)) start)) (setf (beginning-of-line-p ip) nil)))
indenting-p to determine
whether it needs to emit indentation and, if they're both true, emits
as many spaces as indicated by the value of
that uses the
indenting-printer can control the indentation by
Incrementing and decrementing
indentation changes the number
of leading spaces, while setting
temporarily turn off indentation.
(defun indent-if-necessary (ip) (when (and (beginning-of-line-p ip) (indenting-p ip)) (loop repeat (indentation ip) do (write-char #\Space (out ip))) (setf (beginning-of-line-p ip) nil)))
The last two functions in the
indenting-printer API are
emit-freshline, which are both used to
emit a newline character, similar to the
FORMAT directives. That is, the only difference is that
emit-newline always emits a newline, while
emit-freshline does so only if
false. Thus, multiple calls to
emit-freshline without any
emits won't result in a blank line. This is handy
when one piece of code wants to generate some output that should end
with a newline while another piece of code wants to generate some
output that should start on a newline but you don't want a blank line
between the two bits of output.
(defun emit-newline (ip) (write-char #\Newline (out ip)) (setf (beginning-of-line-p ip) t)) (defun emit-freshline (ip) (unless (beginning-of-line-p ip) (emit-newline ip)))
With those preliminaries out of the way, you're ready to get to the guts of the FOO processor.
Now you're ready to define the interface that'll be used by the FOO
language processor to emit HTML. You can define this interface as a
set of generic functions because you'll need two implementations--one
that actually emits HTML and another that the
html macro can
use to collect a list of actions that need to be performed, which can
then be optimized and compiled into code that emits the same output
in a more efficient way. I'll call this set of generic functions the
backend interface. It consists of the following eight generic
(defgeneric raw-string (processor string &optional newlines-p)) (defgeneric newline (processor)) (defgeneric freshline (processor)) (defgeneric indent (processor)) (defgeneric unindent (processor)) (defgeneric toggle-indenting (processor)) (defgeneric embed-value (processor value)) (defgeneric embed-code (processor code))
While several of these functions have obvious correspondence to
indenting-printer functions, it's important to understand that
these generic functions define the abstract operations that are used by
the FOO language processors and won't always be implemented in terms
of calls to the
That said, perhaps the easiest way to understand the semantics of
these abstract operations is to look at the concrete implementations
of the methods specialized on
html-pretty-printer, the class
used to generate human-readable HTML.
You can start by defining a class with two slots--one to hold an
indenting-printer and one to hold the tab
width--the number of spaces you want to increase the indentation for
each level of nesting of HTML elements.
(defclass html-pretty-printer () ((printer :accessor printer :initarg :printer) (tab-width :accessor tab-width :initarg :tab-width :initform 2)))
Now you can implement methods specialized on
html-pretty-printer on the eight generic functions that make up
the backend interface.
The FOO processors use the
raw-string function to emit strings
that don't need character escaping, either because you actually want
to emit normally reserved characters or because all reserved
characters have already been escaped. Usually
invoked with strings that don't contain newlines, so the default
behavior is to use
emit/no-newlines unless the caller
specifies a non-
(defmethod raw-string ((pp html-pretty-printer) string &optional newlines-p) (if newlines-p (emit (printer pp) string) (emit/no-newlines (printer pp) string)))
toggle-indenting implement fairly
straightforward manipulations of the underlying
indenting-printer. The only wrinkle is that the HTML pretty
printer generates pretty output only when the dynamic variable
*pretty* is true. When it's
NIL, you should generate
compact HTML with no unnecessary whitespace. So, these methods, with
the exception of
newline, all check
(defmethod newline ((pp html-pretty-printer)) (emit-newline (printer pp))) (defmethod freshline ((pp html-pretty-printer)) (when *pretty* (emit-freshline (printer pp)))) (defmethod indent ((pp html-pretty-printer)) (when *pretty* (incf (indentation (printer pp)) (tab-width pp)))) (defmethod unindent ((pp html-pretty-printer)) (when *pretty* (decf (indentation (printer pp)) (tab-width pp)))) (defmethod toggle-indenting ((pp html-pretty-printer)) (when *pretty* (with-slots (indenting-p) (printer pp) (setf indenting-p (not indenting-p)))))
Finally, the functions
used only by the FOO compiler--
embed-value is used to
generate code that'll emit the value of a Common Lisp expression,
embed-code is used to embed a bit of code to be run and
its result discarded. In the interpreter, you can't meaningfully
evaluate embedded Lisp code, so the methods on these functions always
signal an error.
(defmethod embed-value ((pp html-pretty-printer) value) (error "Can't embed values when interpreting. Value: ~s" value)) (defmethod embed-code ((pp html-pretty-printer) code) (error "Can't embed code when interpreting. Code: ~s" code))
Now to connect the FOO language to the processor interface, all you need is a function that takes an object and processes it, invoking the appropriate processor functions to generate HTML. For instance, when given a simple form like this:
this function might execute this sequence of calls on the processor:
(freshline processor) (raw-string processor "<p" nil) (raw-string processor ">" nil) (raw-string processor "Foo" nil) (raw-string processor "</p>" nil) (freshline processor)
For now you can define a simple function that just checks whether a
form is, in fact, a legal FOO form and, if it is, hands it off to the
process-sexp-html for processing. In the next
chapter, you'll add some bells and whistles to this function to allow
it to handle macros and special operators. But for now it looks like
(defun process (processor form) (if (sexp-html-p form) (process-sexp-html processor form) (error "Malformed FOO form: ~s" form)))
sexp-html-p determines whether the given object
is a legal FOO expression, either a self-evaluating form or a
properly formatted cons.
(defun sexp-html-p (form) (or (self-evaluating-p form) (cons-form-p form)))
Self-evaluating forms are easily handled: just convert to a string
PRINC-TO-STRING and escape the characters in the variable
*escapes*, which, as you'll recall, is initially bound to the
*element-escapes*. Cons forms you pass off to
(defun process-sexp-html (processor form) (if (self-evaluating-p form) (raw-string processor (escape (princ-to-string form) *escapes*) t) (process-cons-sexp-html processor form)))
process-cons-sexp-html is then responsible for
emitting the opening tag, any attributes, the body, and the closing
tag. The main complication here is that to generate pretty HTML, you
need to emit fresh lines and adjust the indentation according to the
type of the element being emitted. You can categorize all the
elements defined in HTML into one of three categories: block,
paragraph, and inline. Block elements--such as
ul--are emitted with fresh lines before and after both their
opening and closing tags and with their contents indented one level.
Paragraph elements--such as
blockquote--are emitted with a fresh line before the opening
tag and after the closing tag. Inline elements are simply emitted in
line. The following three parameters list the elements of each type:
(defparameter *block-elements* '(:body :colgroup :dl :fieldset :form :head :html :map :noscript :object :ol :optgroup :pre :script :select :style :table :tbody :tfoot :thead :tr :ul)) (defparameter *paragraph-elements* '(:area :base :blockquote :br :button :caption :col :dd :div :dt :h1 :h2 :h3 :h4 :h5 :h6 :hr :input :li :link :meta :option :p :param :td :textarea :th :title)) (defparameter *inline-elements* '(:a :abbr :acronym :address :b :bdo :big :cite :code :del :dfn :em :i :img :ins :kbd :label :legend :q :samp :small :span :strong :sub :sup :tt :var))
test whether a given tag is a member of the corresponding
(defun block-element-p (tag) (find tag *block-elements*)) (defun paragraph-element-p (tag) (find tag *paragraph-elements*))
Two other categorizations with their own predicates are the elements
that are always empty, such as
hr, and the three
script, in which
whitespace is supposed to be preserved. The former are handled
specially when generating regular HTML (in other words, not XHTML)
since they're not supposed to have a closing tag. And when emitting
the three tags in which whitespace is preserved, you can temporarily
turn off indentation so the pretty printer doesn't add any spaces
that aren't part of the element's actual contents.
(defparameter *empty-elements* '(:area :base :br :col :hr :img :input :link :meta :param)) (defparameter *preserve-whitespace-elements* '(:pre :script :style)) (defun empty-element-p (tag) (find tag *empty-elements*)) (defun preserve-whitespace-p (tag) (find tag *preserve-whitespace-elements*))
The last piece of information you need when generating HTML is whether you're generating XHTML since that affects how you emit empty elements.
(defparameter *xhtml* nil)
With all that information, you're ready to process a cons FOO form.
parse-cons-form to parse the list into three parts, the
tag symbol, a possibly empty plist of attribute key/value pairs, and a
possibly empty list of body forms. You then emit the opening tag, the
body, and the closing tag with the helper functions
(defun process-cons-sexp-html (processor form) (when (string= *escapes* *attribute-escapes*) (error "Can't use cons forms in attributes: ~a" form)) (multiple-value-bind (tag attributes body) (parse-cons-form form) (emit-open-tag processor tag body attributes) (emit-element-body processor tag body) (emit-close-tag processor tag body)))
emit-open-tag you have to call
appropriate and then emit the attributes with
You need to pass the element's body to
emit-open-tag so when
it's emitting XHTML, it knows whether to finish the tag with
(defun emit-open-tag (processor tag body-p attributes) (when (or (paragraph-element-p tag) (block-element-p tag)) (freshline processor)) (raw-string processor (format nil "<~(~a~)" tag)) (emit-attributes processor attributes) (raw-string processor (if (and *xhtml* (not body-p)) "/>" ">")))
emit-attributes the attribute names aren't evaluated since
they must be keyword symbols, but you should invoke the top-level
process function to evaluate the attribute values, binding
*attribute-escapes*. As a convenience for
specifying boolean attributes, whose value should be the name of the
attribute, if the value is
T--not just any true value but
T--then you replace the value with the name of the
(defun emit-attributes (processor attributes) (loop for (k v) on attributes by #'cddr do (raw-string processor (format nil " ~(~a~)='" k)) (let ((*escapes* *attribute-escapes*)) (process processor (if (eql v t) (string-downcase k) v))) (raw-string processor "'")))
Emitting the element's body is similar to emitting the attribute
values: you can loop through the body calling
evaluate each form. The rest of the code is dedicated to emitting
fresh lines and adjusting the indentation as appropriate for the type
(defun emit-element-body (processor tag body) (when (block-element-p tag) (freshline processor) (indent processor)) (when (preserve-whitespace-p tag) (toggle-indenting processor)) (dolist (item body) (process processor item)) (when (preserve-whitespace-p tag) (toggle-indenting processor)) (when (block-element-p tag) (unindent processor) (freshline processor)))
emit-close-tag, as you'd probably expect, emits the
closing tag (unless no closing tag is necessary, such as when the
body is empty and you're either emitting XHTML or the element is one
of the special empty elements). Regardless of whether you actually
emit a close tag, you need to emit a final fresh line for block and
(defun emit-close-tag (processor tag body-p) (unless (and (or *xhtml* (empty-element-p tag)) (not body-p)) (raw-string processor (format nil "</~(~a~)>" tag))) (when (or (paragraph-element-p tag) (block-element-p tag)) (freshline processor)))
process is the basic FOO interpreter. To make it
a bit easier to use, you can define a function,
process, passing it an
and a form to evaluate. You can define and use a helper function,
get-pretty-printer, to get the pretty printer, which returns
the current value of
*html-pretty-printer* if it's bound;
otherwise, it makes a new instance of
*html-output* as its output stream.
(defun emit-html (sexp) (process (get-pretty-printer) sexp)) (defun get-pretty-printer () (or *html-pretty-printer* (make-instance 'html-pretty-printer :printer (make-instance 'indenting-printer :out *html-output*))))
With this function, you can emit HTML to
than expose the variable
*html-output* as part of FOO's public
API, you should define a macro,
with-html-output, that takes
care of binding the stream for you. It also lets you specify whether
you want pretty HTML output, defaulting to the value of the variable
(defmacro with-html-output ((stream &key (pretty *pretty*)) &body body) `(let* ((*html-output* ,stream) (*pretty* ,pretty)) ,@body))
So, if you wanted to use
emit-html to generate HTML to a file,
you could write the following:
(with-open-file (out "foo.html" :direction output) (with-html-output (out :pretty t) (emit-html *some-foo-expression*)))
In the next chapter, you'll look at how to implement a macro that compiles FOO expressions into Common Lisp so you can embed HTML generation code directly into your Lisp programs. You'll also extend the FOO language to make it a bit more expressive by adding its own flavor of special operators and macros.
1In fact, it's probably too expressive since it can also generate all sorts of output that's not even vaguely legal HTML. Of course, that might be a feature if you need to generate HTML that's not strictly correct to compensate for buggy Web browsers. Also, it's common for language processors to accept programs that are syntactically correct and otherwise well formed that'll nonetheless provoke undefined behavior when run.
2Well, almost every tag.
Certain tags such as
BR don't. You'll deal with
those in the section "The Basic Evaluation Rule."
3In the strict language of the Common Lisp standard, keyword symbols aren't self-evaluating, though they do, in fact, evaluate to themselves. See section 184.108.40.206.3 of the language standard or HyperSpec for a brief discussion.
4The requirement to use objects that the Lisp reader knows how to read isn't a hard-and-fast one. Since the Lisp reader is itself customizable, you could also define a new reader-level syntax for a new kind of object. But that tends to be more trouble than it's worth.
5Another, more purely object-oriented, approach
would be to define two classes, perhaps
html-raw-printer, and then define no-op methods
html-raw-printer for the methods that should do
stuff only when
*pretty* is true. However, in this case, after
defining all the no-op methods, you'd end up with more code, and then
you'd have the hassle of making sure you created an instance of the
right class at the right time. But in general, using polymorphism to
replace conditionals is a good strategy.
6You don't need a predicate for
since you only ever test for block and paragraph elements. I include
the parameter here for completeness.
7While XHTML requires boolean attributes to be notated
with their name as the value to indicate a true value, in HTML it's
also legal to simply include the name of the attribute with no value,
<option selected> rather than
selected='selected'>. All HTML 4.0-compatible browsers should
understand both forms, but some buggy browsers understand only the
no-value form for certain attributes. If you need to generate HTML
for such browsers, you'll need to hack
emit-attributes to emit
those attributes a bit differently.