cgrand / Enliven
Programming Languages
Enliven 
Enliven: a (not yet) continuous templating system.
WARNING very wet paint
Enliven is the successor to Enlive
Even dictionaries say so:
Enlive: To Enliven (Obs.)
Enliven also stands for Enlive N(ext).
Not tied to HTML
Currently Enliven can template plain text (enliven.text) and html (enliven.html).
The HTML "domain" is even composite since it uses the text "domain" to template text nodes.
Simpler selectors
Selectors are now functions from loc to locs. Most domains should expose a sel function to coerce
a domain specific selector (eg a CSS-selector in string or a regex) to a selector fn.
Parallel execution of transformations
All transformations occur at once for maximum declarativeness (and it enables good performance). It follows that two transformations can't work on the same node or on a node and one of its ancestors.
This constraint is heavily mitigated by infinite-resolution selectors and transformation-refined selectors.
Infinite-resolution selectors
Selectors don't stop at nodes as specified in the DOM. Any node can be subdivided at will!
For example classes in a class attribute can be targeted independently. Each character of a text node can be transformed independently.
You can append/content/prepend on an element without conflicts!
However this happens under the hood, see:
(at
"div" (class "important" :important)
"div" (class "footnote" :foot-note))
Those two transformations won't conflict because they refine their selector.
Point-free
Templates take a single argument which is the data model to render.
Transformations don't take as arguments the actual values but keys or paths into the model.
=> ((static-template
(enliven.html.jsoup/parse "<div>")
"div" (content :sentence))
{:sentence "Hello world"})
"<html><head></head><body><div>Hello world</div></body></html>"
Some transformations like dup introduces a new scope so that sub-transformation can only see
the current item.
=> ((static-template
(enliven.html.jsoup/parse "<ul><li>")
"li" (dup :todos
(content []))) ; [] is the empty path, so points to the whole value
{:todos ["Laundry" "Walk the dog"]})
"<html><head></head><body><ul><li>Laundry</li><li>Walk the dog</li></ul></body></html>"
It's fast
Generating a 5x5 table. (pr-str just dumps the 5*5 vector):
- pr-str (24.5µs),
- enliven (24.8µs),
- hiccup (44.8µs),
- enlive (130µs),
- laser (1280µs).
https://gist.github.com/cgrand/8471718
Dev details
Processing model
- Nodes are selected
- For each selected node, associated transformations are grounded -- this is where the nodes are refined to avoid conflicts. The result is a set of rules.
- A hierarchical plan is created from the rules.
- The plan can either be executed as is or "compiled".
Paths and segments
Rules (as produced by the grounding phase) are pairs of a path and an action.
As a first approximation, you can imagine that they are going to be used in conjunction with assoc-in to updates parts of the structure.
If you have one rule on [:a :b] and another one [:a :c], they can be executed in any order (they may even be performed concurrently as long as you coordinate on the longest common prefix).
However rules on [:a :b] and [:a :b :c] can't be freely reordered and as such are forbidden.
To alleviate this constraint, segments are a bit more expressive than simple keys in associative data structures.
Segments are used in conjunction with fetch and putback which are akin to get and assoc.
Numbers, strings and keywords are segments and Enliven knows that if they are different they don't impede.
Slices (ranges) are also segments and non-overlapping ranges don't impede. (Slices and numbers should not be mixed and that's the main purpose of path canonicalization).
Other segments are opaque and thus will conflict with any other segment. A good practice is to use non-opaque segments as much as possible.
For example the html/classes segment can't be used on an HTML element: otherwise it would clash with any other transformation on the element.
So the html/classes segment is meant to be used after [:attrs :class] thus giving to the planner the information that html/classes owns only the class attribute.
Furthermore the html/classes segment returns a map so individual classes can be addressed individually. For example [:attrs :class html/classes "class1"] and [:attrs :class html/classes "class2"] don't clash.
A custom segment thus creates a sort of contention point but by having it to return an associative structure it becomes a coordination point.
Locs
Locs are a kind of zippers for associative structures. They support only two moves: up and down (which takes a segment as additional argument).
License
Copyright © 2014 Christophe Grand
Distributed under the Eclipse Public License, the same as Clojure.