Computer Science/164/HW3

Choose the problem that your language will solve

Navigation bars are a standard feature of many web pages. In recent years, a somewhat standard paradigm has developed of using unordered lists to create them. The lists are then styled with CSS to give them a more traditional appearance.

Accomplishing this in a good-looking and cross-browser-compatible manner is not trivial and requires a fair amount of CSS fiddling. Furthermore, the HTML code which must be written to set up the navigational elements contains a fair amount of extraneous markup which detracts from getting a clear picture of the navigational elements and their actions. Yet another problem is that the CSS and HTML code live in separate files, requiring the user to have two separate documents open to design their navigational bars.

In addition, navigational menus can often contain submenus for a structured hierarchy. This requires additional javascript coding to work properly and further detracts the web page designer from concentrating on accomplishing his task; namely, working on the actual design of the menu.

The problem that we seek to address is that designing navigational elements in HTML is needlessly complicated and requires too much plumbing. Therefore, we propose a language with a special syntax designed specifically to make this process easy and fun.

For comparison, here's an example of a very simple group of navigational links:

HTML

<div id="social">
<ul id="socialinks"><li><a href="/one.htm">One</a></li><li><a href="/two.htm">Two</a></li><li><a href="/fun/">Fun</a></li>
</ul>
</div>

Corresponding CSS

#social {
  background: #6d6d6d;
  color: white;
  padding: 1em 0px;
  margin: 0px auto;
  clear: both;
  bottom: 0px;
  position: absolute;
  width: 100%;
}

ul {
  margin: 0px auto;
  padding: 0px;
  list-style: none;
  min-width: 15em;
  text-align: center;
}

li {
  display: inline;
  margin: 0px;
  padding: 0px;
}

li a {
  padding: 0.25em 0.5em;
  margin: 0px;
  text-decoration: none;
}

li a { color: #82c753; }

li a:hover {
  background: white;
  text-decoration: underline;
}

To produce the following

Note that there is a lot of information here which is rather superfluous. If we want to add an element here which is not linked, the CSS becomes more involved. Furthermore, adding things like submenus is far from trivial and requires knowledge of JavaScript and avoiding a number of pitfalls to deal with different browser vendors.

Our solution seeks to abstract away all of these problems with a simple syntax which makes the important things clear immediately. The example above simply becomes:

{ color: #82c753; background: #6d6d6d; text-decoration: none; padding: 0.25em 0.5em;
	hover: { background: white; text-decoration: underline; };
layout: horizontal; }
*One | /one.htm
*Two | /two.htm
*Fun | /fun/

The formatting definition immediately preceding the list of items contains CSS properties in property: value format separated by semicolons. hover is a special keyword whose value is another set of definitions applied to hovered navbar elements. layout is another special keyword whose value is either horizontal or vertical and controls whether the navbar elements are shown side-by-side or one above the other, respectively. The navbar elements are delimited by a newline and an asterisk and consist of a string of text (for the title of the navbar item) followed by a pipe and then the URL for the item. If the developer wants to include a pipe as part of the item title, the title simply needs to be enclosed in quotes. The URL is optional and leaving it out results in an item which does not link to anything. This can be useful for drop-down menus.

Creating hierarchical menus is simple. For example:

*Item one
{ direction: below }
**Submenu item one | /submenu/one.htm
**Submenu item two | /submenu/two.htm
**Submenu item three | /submenu/three.htm
{ direction: left }
***Subsubmenu item a | /submenu/sub/a.htm
*Item two | /two.htm

By default, all the submenus inherit the styles of their parents. If this is not desired, properties to override can be specified in the curly braces. The direction keyword shown above simply controls where the submenu appears relative to its parent item. In the example above, the "Submenu item one/two/three" elements appear below the navbar containing "Item one/two", and the "Subsubmenu item a" appears to the left of that menu.

This problem is worth solving because it allows web developers to save time and not have to deal with debugging browser issues. Our library will allow developers to make beautiful-looking text-based navigation bars in just a few seconds and the syntax is designed to be intuitive such that new features take very little time to learn (the content description element) while existing features will be familiar to those who have worked with this before (the formatting elements). It also improves the user experience, as the code we generate will be accessible to users with disabilities and will be designed in such a way that it degrades nicely across browser versions and vendors. Thus we are actually solving two problems at once: a development problem, in which creating well-functioning and nice-looking navigational elements is difficult, and a user interface problem, in which users are often presented with poorly-designed navigational tools.

Part 2: Study a language that solves a similar problem and describe it in this homework

1. One simple code example. Your comments must sufficiently explain the meaning of the code.
2. Describe the implementation. (Two paragraphs and pseudocode of the implementation.)

Part 3: What features specifically will your language support

1. The domain. Give a list of a few small programs that you hope to be able to write in your language. These will guide your language design.
2. Give an outline of the programming model of your language. What are the "objects" in your language? What operations on these objects will you allow? How will you combine small programs into bigger ones, hiding the implementation details of the smaller ones (what abstraction mechanism will you support, if any?).

Part 4: Decide on the implementation

Choose two alternative implementations, and compare them along these aspects. For each aspect, write a paragraph. Your two approaches may be identical in some aspects.

• frontend: How will you translate the input (source) program into the program's internal presentation (often, this is an AST)? You know that this can be done with a parser or with an embedded DSL, as in rake or protovis. Some languages may not need an internal representation.

The decision to go with a parsed input rather than an embedded DSL was not one we made lightly. The primary advantage of the embedded DSL is simplicity; it doesn't require us to go through the trouble of writing a grammar for the input. However, that simplicity comes at the price of flexibility; the parsed input can be specified in any format we choose, whereas the input of an embedded DSL must fit into the syntax of the parent language. Considering that the main goal of this language is to provide a clean and simple syntax, restricting ourselves to the syntax of the parent language is clearly counterproductive.

• the core language: Are some features built as sugar on top of a simpler-to-implement language? If yes, what is that core language, and how will you desugar those features?

The purpose of this language is to provide a convenient and concise way for users to create CSS navigation bars. As such, it must be implemented entirely as sugar on top of CSS and HTML. The input will take the form of a list with associated attributes, and the program will translate the input into the corresponding CSS code along with all the plumbing needed to arrange the bar properly.

• internal representation. What are the alternative ways of representing the program for the interpreter or compiler?

We considered using both an AST and objects for the interpreter's internal representation of the input. In the end, we decided that objects are better suited to the data. Our input is not going to come in the form of a deeply-nested hierarchy, so the tree structure is unnecessary; the input is a collection of data points, each of which has a set of associated attributes. This lends itself very well to object representation.

• interpreter/compiler: how will you interpret or compile the language? Is there a prepossessing step, eg translation from AST to bytecode? If you want to generate code, what is the target language? Why did you choose that language?

The interpretation will be fairly straightforward; the interpreter will go through the object representation of the input, generating the appropriate CSS and HTML as it goes.

• debugging: How will you debug the implementation of the language?

We will provide an HTML page with a text box. You can type an input program into the text box, and the output navigation bar will be displayed above. If there is an error, the error message will be displayed instead of the navigation bar.