Pure is a fast radix-tree based HTTP router that sticks to the native implimentations of Go's "net/http" package; in essence, keeping the handler implimentations 'pure' by using Go 1.7's "context" package.
Why Another HTTP Router?
I initially created lars, which I still maintain, that wraps the native implimentation, think of this package as a Go pure implimentation of lars
Key & Unique Features
- It sticks to Go's native implimentations while providing helper functions for convenience
- Fast & Efficient - pure uses a custom version of httprouter's radix tree, so incredibly fast and efficient.
Installation
Use go get
go get -u github.com/go-playground/pureUsage
package main
import (
"net/http"
"github.com/go-playground/pure"
mw "github.com/go-playground/pure/examples/middleware/logging-recovery"
)
func main() {
p := pure.New()
p.Use(mw.LoggingAndRecovery(true))
p.Get("/", helloWorld)
http.ListenAndServe(":3007", p.Serve())
}
func helloWorld(w http.ResponseWriter, r *http.Request) {
w.Write([]byte("Hello World"))
}RequestVars
This is an interface that is used to pass request scoped variables and functions using context.Context.
It is implimented in this way because retrieving values from context isn't the fastest, and so using this
the router can store multiple pieces of information while reducing lookup time to a single stored RequestVars.
Currently only the URL/SEO params are stored on the RequestVars but if/when more is added they can merely be added
to the RequestVars and there will be no additional lookup time.
URL Params
p := p.New()
// the matching param will be stored in the context's params with name "id"
l.Get("/user/:id", UserHandler)
// extract params like so
rv := pure.ReqestVars(r) // done this way so only have to extract from context once, read above
rv.URLParam(paramname)
// serve css, js etc.. pure.RequestVars(r).URLParam(pure.WildcardParam) will return the remaining path if
// you need to use it in a custom handler...
l.Get("/static/*", http.FileServer(http.Dir("static/")))
...Note: Since this router has only explicit matches, you can not register static routes and parameters for the same path segment. For example you can not register the patterns /user/new and /user/:user for the same request method at the same time. The routing of different request methods is independent from each other. I was initially against this, however it nearly cost me in a large web application where the dynamic param value say :type actually could have matched another static route and that's just too dangerous and so it is not allowed.
Groups
p.Use(LoggingAndRecovery, Gzip...)
...
p.Post("/users/add", ...)
// creates a group for user + inherits all middleware registered using p.Use()
user := p.Group("/user/:userid")
user.Get("", ...)
user.Post("", ...)
user.Delete("/delete", ...)
contactInfo := user.Group("/contact-info/:cid")
contactinfo.Delete("/delete", ...)
// creates a group for others + inherits all middleware registered using p.Use() + adds
// OtherHandler to middleware
others := p.Group("/others", OtherHandler)
// creates a group for admin WITH NO MIDDLEWARE... more can be added using admin.Use()
admin := p.Group("/admin", nil)
admin.Use(SomeAdminSecurityMiddleware)
...Decoding Body
currently JSON, XML, FORM, Multipart Form and url.Values are support out of the box.
// second argument denotes yes or no I would like URL query parameter fields
// to be included. i.e. 'id' and 'id2' in route '/user/:id?id2=val' should it be included.
if err := pure.Decode(r, true, maxBytes, &user); err != nil {
log.Println(err)
}Misc
// set custom 404 ( not Found ) handler
l.Register404(404Handler, middleware_like_logging)
// Redirect to or from ending slash if route not found, default is true
l.SetRedirectTrailingSlash(true)
// Handle 405 ( Method Not allowed ), default is false
l.RegisterMethodNotAllowed(middleware)
// automatically handle OPTION requests; manually configured
// OPTION handlers take precedence. default false
l.RegisterAutomaticOPTIONS(middleware)Middleware
There are some pre-defined middlewares within the middleware folder; NOTE: that the middleware inside will comply with the following rule(s):
- Are completely reusable by the community without modification
Other middleware will be listed under the examples/middleware/... folder for a quick copy/paste modify. As an example a LoddingAndRecovery middleware is very application dependent and therefore will be listed under the examples/middleware/...
Benchmarks
Run on MacBook Pro (Retina, 15-inch, Late 2013) 2.6 GHz Intel Core i7 16 GB 1600 MHz DDR3 using Go version go1.7.3 darwin/amd64
NOTICE: pure uses a custom version of httprouter's radix tree, benchmarks can be found here the slowdown is with the use of the context package, as you can see when no SEO params are defined, and therefore no need to store anything in the context, it is faster than even lars.
go test -bench=. -benchmem=true
#GithubAPI Routes: 203
Pure: 37560 Bytes
#GPlusAPI Routes: 13
Pure: 2808 Bytes
#ParseAPI Routes: 26
Pure: 5072 Bytes
#Static Routes: 157
Pure: 21224 Bytes
BenchmarkPure_Param 10000000 156 ns/op 240 B/op 1 allocs/op
BenchmarkPure_Param5 10000000 199 ns/op 240 B/op 1 allocs/op
BenchmarkPure_Param20 5000000 349 ns/op 240 B/op 1 allocs/op
BenchmarkPure_ParamWrite 10000000 209 ns/op 240 B/op 1 allocs/op
BenchmarkPure_GithubStatic 30000000 45.4 ns/op 0 B/op 0 allocs/op
BenchmarkPure_GithubParam 10000000 219 ns/op 240 B/op 1 allocs/op
BenchmarkPure_GithubAll 30000 40244 ns/op 40082 B/op 167 allocs/op
BenchmarkPure_GPlusStatic 50000000 30.4 ns/op 0 B/op 0 allocs/op
BenchmarkPure_GPlusParam 10000000 173 ns/op 240 B/op 1 allocs/op
BenchmarkPure_GPlus2Params 10000000 190 ns/op 240 B/op 1 allocs/op
BenchmarkPure_GPlusAll 1000000 2108 ns/op 2640 B/op 11 allocs/op
BenchmarkPure_ParseStatic 50000000 29.9 ns/op 0 B/op 0 allocs/op
BenchmarkPure_ParseParam 10000000 154 ns/op 240 B/op 1 allocs/op
BenchmarkPure_Parse2Params 10000000 167 ns/op 240 B/op 1 allocs/op
BenchmarkPure_ParseAll 500000 3209 ns/op 3840 B/op 16 allocs/op
BenchmarkPure_StaticAll 200000 9881 ns/op 0 B/op 0 allocs/opPackage Versioning
I'm jumping on the vendoring bandwagon, you should vendor this package as I will not be creating different version with gopkg.in like allot of my other libraries.
Why? because my time is spread pretty thin maintaining all of the libraries I have + LIFE, it is so freeing not to worry about it and will help me keep pouring out bigger and better things for you the community.
I am open versioning with gopkg.in should anyone request it, but this should be stable going forward.
Licenses
- MIT License (MIT), Copyright (c) 2016 Dean Karn
- BSD License, Copyright (c) 2013 Julien Schmidt. All rights reserved.
