gzip过滤器源码分析

本文参考gzip过滤器的源码
https://github.com/phyber/negroni-gzip/blob/master/gzip/gzip.go

该过滤器是Negroni（https://github.com/urfave/negroni） 的一个中间件，gzip是用来压缩的。

我们关注所有首字母大写的函数，一共有五个。

func (grw *gzipResponseWriter) WriteHeader(code int)
func (grw *gzipResponseWriter) Write(b []byte) (int, error)
func (rw *gzipResponseWriterCloseNotifier) CloseNotify() <-chan bool 
func Gzip(level int) *handle
func (h *handler) ServeHTTP(w http.ResponseWriter, r *http.Request, next http.HandlerFunc)

这主要是因为我们在外部只能调用首字母大写的函数，那我们至少要知道这份代码能帮我们干些什么。
接下来我们依次自上到下的阅读理解这些函数。

WriteHeader函数

// Check whether underlying response is already pre-encoded and disable
// gzipWriter before the body gets written, otherwise encoding headers
func (grw *gzipResponseWriter) WriteHeader(code int) {
	headers := grw.ResponseWriter.Header()
	if headers.Get(headerContentEncoding) == "" {
		headers.Set(headerContentEncoding, encodingGzip)
		headers.Add(headerVary, headerAcceptEncoding)
	} else {
		grw.w.Reset(ioutil.Discard)
		grw.w = nil
	}

	// Avoid sending Content-Length header before compression. The length would
	// be invalid, and some browsers like Safari will report
	// "The network connection was lost." errors
	grw.Header().Del(headerContentLength)

	grw.ResponseWriter.WriteHeader(code)
	grw.wroteHeader = true
}

首先开始的if else语句是用来判断http相应的头部“Content-Encoding”一栏是否为空，该栏目代表着http响应的压缩格式，为空就是还没压缩过，那就对它头部进行设置，设置压缩方式为“zip”。
若“Content-Encoding”不为空，说明该已经压缩过了，将grw的Writer置为nil，这也容易理解，既然已经压缩过了，那么就不能再压缩一次了。
然后避免把头部的“Content-Length”这一栏，把头部的“Content-Length”这一栏删掉。

Write函数

func (grw *gzipResponseWriter) Write(b []byte) (int, error) {
	if !grw.wroteHeader {
		grw.WriteHeader(http.StatusOK)
	}
	if grw.w == nil {
		return grw.ResponseWriter.Write(b)
	}
	if len(grw.Header().Get(headerContentType)) == 0 {
		grw.Header().Set(headerContentType, http.DetectContentType(b))
	}
	return grw.w.Write(b)
}

当没写头部的时候，写头部。
若grw的writer为nil，那么从上一函数可以得知该http响应已经压缩了，那么就调用ResponseWriter（negroni）进行写。
若grw的头部没写，那么就写上。
最后用gzip的Writer进行写。
总而言之，这一段函数就两个返回值，要么grw的writer为nil，已经压缩了，那就调用ResponseWriter（negroni）进行写，要么就没有压缩过，那么就用gzip的writer写。

CloseNotify函数

func (rw *gzipResponseWriterCloseNotifier) CloseNotify() <-chan bool {
	return rw.ResponseWriter.(http.CloseNotifier).CloseNotify()
}

当客户端关闭时，服务端可以通过该函数返回的一个通道得知连接中断。

Gzip函数

func Gzip(level int) *handler {
	h := &handler{}
	h.pool.New = func() interface{} {
		gz, err := gzip.NewWriterLevel(ioutil.Discard, level)
		if err != nil {
			panic(err)
		}
		return gz
	}
	return h
}

返回一个handler，这个handler主要在下一个函数用到

ServeHTTP函数

func (h *handler) ServeHTTP(w http.ResponseWriter, r *http.Request, next http.HandlerFunc) {
	// Skip compression if the client doesn't accept gzip encoding.
	if !strings.Contains(r.Header.Get(headerAcceptEncoding), encodingGzip) {
		next(w, r)
		return
	}

	// Skip compression if client attempt WebSocket connection
	if len(r.Header.Get(headerSecWebSocketKey)) > 0 {
		next(w, r)
		return
	}

	// Retrieve gzip writer from the pool. Reset it to use the ResponseWriter.
	// This allows us to re-use an already allocated buffer rather than
	// allocating a new buffer for every request.
	// We defer g.pool.Put here so that the gz writer is returned to the
	// pool if any thing after here fails for some reason (functions in
	// next could potentially panic, etc)
	gz := h.pool.Get().(*gzip.Writer)
	defer h.pool.Put(gz)
	gz.Reset(w)

	// Wrap the original http.ResponseWriter with negroni.ResponseWriter
	// and create the gzipResponseWriter.
	nrw := negroni.NewResponseWriter(w)
	grw := newGzipResponseWriter(nrw, gz)

	// Call the next handler supplying the gzipResponseWriter instead of
	// the original.
	next(grw, r)

	gz.Close()
}

首先是判断请求的头部是否有“Accept-Encoding”一栏，若没有，则说明客户端不接受压缩的响应，那我们就不能压缩响应的报文，直接调用下一个中间件处理该请求。
其次是判断请求的头部是否有“Sec-WebSocket-Key”一栏，如果有，代表客户端想要进行长连接，那也不能压缩，调用下一个中间件处理该请求。
最后就是创建gzipResponseWriter，进行压缩处理，关闭gzipwriter。

源代码

以下是gzip源码

// Package gzip implements a gzip compression handler middleware for Negroni.
package gzip

import (
	"compress/gzip"
	"io/ioutil"
	"net/http"
	"strings"
	"sync"

	"github.com/urfave/negroni"
)

// These compression constants are copied from the compress/gzip package.
const (
	encodingGzip = "gzip"

	headerAcceptEncoding  = "Accept-Encoding"
	headerContentEncoding = "Content-Encoding"
	headerContentLength   = "Content-Length"
	headerContentType     = "Content-Type"
	headerVary            = "Vary"
	headerSecWebSocketKey = "Sec-WebSocket-Key"

	BestCompression    = gzip.BestCompression
	BestSpeed          = gzip.BestSpeed
	DefaultCompression = gzip.DefaultCompression
	NoCompression      = gzip.NoCompression
)

// gzipResponseWriter is the ResponseWriter that negroni.ResponseWriter is
// wrapped in.
type gzipResponseWriter struct {
	w *gzip.Writer
	negroni.ResponseWriter
	wroteHeader bool
}

// Check whether underlying response is already pre-encoded and disable
// gzipWriter before the body gets written, otherwise encoding headers
func (grw *gzipResponseWriter) WriteHeader(code int) {
	headers := grw.ResponseWriter.Header()
	if headers.Get(headerContentEncoding) == "" {
		headers.Set(headerContentEncoding, encodingGzip)
		headers.Add(headerVary, headerAcceptEncoding)
	} else {
		grw.w.Reset(ioutil.Discard)
		grw.w = nil
	}

	// Avoid sending Content-Length header before compression. The length would
	// be invalid, and some browsers like Safari will report
	// "The network connection was lost." errors
	grw.Header().Del(headerContentLength)

	grw.ResponseWriter.WriteHeader(code)
	grw.wroteHeader = true
}

// Write writes bytes to the gzip.Writer. It will also set the Content-Type
// header using the net/http library content type detection if the Content-Type
// header was not set yet.
func (grw *gzipResponseWriter) Write(b []byte) (int, error) {
	if !grw.wroteHeader {
		grw.WriteHeader(http.StatusOK)
	}
	if grw.w == nil {
		return grw.ResponseWriter.Write(b)
	}
	if len(grw.Header().Get(headerContentType)) == 0 {
		grw.Header().Set(headerContentType, http.DetectContentType(b))
	}
	return grw.w.Write(b)
}

type gzipResponseWriterCloseNotifier struct {
	*gzipResponseWriter
}

func (rw *gzipResponseWriterCloseNotifier) CloseNotify() <-chan bool {
	return rw.ResponseWriter.(http.CloseNotifier).CloseNotify()
}

func newGzipResponseWriter(rw negroni.ResponseWriter, w *gzip.Writer) negroni.ResponseWriter {
	wr := &gzipResponseWriter{w: w, ResponseWriter: rw}

	if _, ok := rw.(http.CloseNotifier); ok {
		return &gzipResponseWriterCloseNotifier{gzipResponseWriter: wr}
	}

	return wr
}

// handler struct contains the ServeHTTP method
type handler struct {
	pool sync.Pool
}

// Gzip returns a handler which will handle the Gzip compression in ServeHTTP.
// Valid values for level are identical to those in the compress/gzip package.
func Gzip(level int) *handler {
	h := &handler{}
	h.pool.New = func() interface{} {
		gz, err := gzip.NewWriterLevel(ioutil.Discard, level)
		if err != nil {
			panic(err)
		}
		return gz
	}
	return h
}

// ServeHTTP wraps the http.ResponseWriter with a gzip.Writer.
func (h *handler) ServeHTTP(w http.ResponseWriter, r *http.Request, next http.HandlerFunc) {
	// Skip compression if the client doesn't accept gzip encoding.
	if !strings.Contains(r.Header.Get(headerAcceptEncoding), encodingGzip) {
		next(w, r)
		return
	}

	// Skip compression if client attempt WebSocket connection
	if len(r.Header.Get(headerSecWebSocketKey)) > 0 {
		next(w, r)
		return
	}

	// Retrieve gzip writer from the pool. Reset it to use the ResponseWriter.
	// This allows us to re-use an already allocated buffer rather than
	// allocating a new buffer for every request.
	// We defer g.pool.Put here so that the gz writer is returned to the
	// pool if any thing after here fails for some reason (functions in
	// next could potentially panic, etc)
	gz := h.pool.Get().(*gzip.Writer)
	defer h.pool.Put(gz)
	gz.Reset(w)

	// Wrap the original http.ResponseWriter with negroni.ResponseWriter
	// and create the gzipResponseWriter.
	nrw := negroni.NewResponseWriter(w)
	grw := newGzipResponseWriter(nrw, gz)

	// Call the next handler supplying the gzipResponseWriter instead of
	// the original.
	next(grw, r)

	gz.Close()
}