HTTP router based on a strict path tree structure
var HttpHash = require('http-hash');
// Create a new http hash
var hash = HttpHash();
// Create a route mapping to /test/<anything but "/">
hash.set('/test/:foo/', function (req, res) {
res.end();
});
// Get a valid route
var route = hash.get('/test/var');
console.log(route);
/*
-> {
handler: function (req, res) {},
params: {
foo: 'var'
},
splat: null
}
*/
// Get an invalid route (returns null)
var missing = hash.get('/missing');
console.log(missing);
/*
-> {
handler: null,
params: {},
splat: null
}
*/
var HttpHash = require('http-hash');
// Create a new http hash
var hash = HttpHash();
// Create a route mapping to /foo/<anything but "/">/<anything>
hash.set('/foo/:test/*', function (req, res) {
res.end();
});
var route = hash.get('/foo/val/one/two/three');
console.log(route);
/*
-> {
handler: function (req, res) { ... },
params: {
test: 'val'
},
splat: 'one/two/three'
}
*/
The most popular node routers are based on regular expression matching. This means that the order in which the routes are defined affects the resolution of a route to handler. Sometimes this is desirable, but it would often be better to have a resolution scheme that is easier to reason about.
http-hash
solves the routing problem by making route resolution
independent of the order in which routes are defined. It does so
by breaking a path into a tree of nodes, based on a simple split
on /
. For example, the route /foo/bar/baz
is treated as tree
nodes foo > bar > baz
. We call foo
, bar
and baz
path
segments.
Theses path segments are arranged into a tree of nodes, where each segment defines a node in the tree. Each node can point to:
-
a fixed handler
node.handler
, otherwise known as the node value -
a set of static paths indexed by path name
node.staticPaths
-
a variable subtree
node.variablePaths
, that can match a single named parameter OR the remainder of a route (splat).
If the last character of a defined route is *
, a variable path
(or splat) will be inserted, consuming the rest of the path.
This allows for subrouting, i.e. if you want to mount a static
filesystem on /fs
you would set the path as /fs/*
where
the nodes are broken down into the tree fs > *
. The remainder
of the route will be returned as a "splat" value, allowing for
further routing.
In the simple case, the route tree is based on exact matches on
the name of the segment. That is to say, for the case where we
want to match /foo/bar/baz
, the tree looks like
{
staticPaths: {
foo: {
staticPaths: {
bar: {
staticPaths: {
baz: {
handler: function (req, res) {}
}
}
}
}
}
}
}
When defining routes, variable paths may be specified. This is
where path segments are prefixed with :
i.e. /foo/:bar/baz
.
For the :bar
segment, the route consumes the single variable
route slot for that node in the tree. So for example, the route
`/foo/:bar/baz looks like
{
staticPaths: {
foo: {
variablePaths: {
staticPaths: {
'baz': function (req, res) {}
}
}
}
}
}
Since a node can have both static and dynamic paths associated with it, the static path will win over the variable path when we resolve the path.
In most cases the trailing / does not matter. Variables cannot be the empty string, and neither can splats. A splat value will not contain the leading slash as it is consumed by the parent node.
If a path conflict occurs, an exception will be thrown. Conflicts occur when:
- A route is defined twice, resolving to two handlers
This is the simplest case where /foo
has been defined twice.
- Variable names in the path are different
Note that /foo/:vara/
and /foo/:varb/
conflict, since they
both resolve to foo.variablePaths
, but have different param
names.
- A variable route is defined for a splat node
In the case of splats being defined at a level, no other
other variables may be specified, as we cannot distinguish
between /foo/:var
and /foo/*
. It is however ok to put static
paths on the same level, i.e. /foo/bar
and /foo/*
. In this
case, the static paths will be tried first before yielding the
splat.
http-hash := () => HttpHash
type HttpHash := {
get: (String: url) => RouteResult,
set: (String: path, Any: handler) => void,
_hash: RouteNode
}
type RouteNode := {
handler: Any,
fixedPaths: Object<String, RouteNode>,
variablePaths: RouteNode | null
}
type RouteResult := {
handler: Any | null,
params: Object<String, String>,
splat: String | null
}
http-hash
exports a safe constructor function that when called
returns a new HttpHash
. get
and set
methods are exposed for
public consumption and the underlying data structure _hash
is
exposed for private inspection/internal use.
hash.set := (String: path, Any: handler) => void
Puts a path in the route table. If the path conflicts with an existing path, an exception will be thrown.
Routes containing a *
that are not part of a /*
prefix will
also throw an exception.
A path should look like /
or /foo
or /:foo
or a union of
theses things, or optionally end with /*
-
param names should not be repeated as they will conflict but there is no strong assertion for this. The last param name wins.
-
specifying a variable twice for a node will cause an exception
-
repeated and trailing '/' are ignored
-
paths are case sensitive
-
variables and splats are not matched by the empty string.
hash.get := (String: path) => RouteResult
Gets a route from the route table. If there is no viable route,
the handler will be returned as null
in the RouteResult
object.
The route result contains a params hash
, containing a key for
each named variable in the path. Additionally, if a splat route
was defined, the splat
property will contain the tail portion
of the route matched.
npm install http-hash
npm test
- Matt Esch