armor/encoder.go

   1 package armor
   2
   3 import (
   4         "encoding/base64"
   5         "io"
   6 )
   7
   8 // https://amp.dev/boilerplate/
   9 // https://amp.dev/documentation/guides-and-tutorials/learn/spec/amp-boilerplate/?format=websites
  10 // https://amp.dev/documentation/guides-and-tutorials/learn/spec/amphtml/?format=websites#the-amp-html-format
  11 const (
  12         boilerplateStart = `<!doctype html>
  13 <html amp>
  14 <head>
  15 <meta charset="utf-8">
  16 <script async src="https://cdn.ampproject.org/v0.js"></script>
  17 <link rel="canonical" href="#">
  18 <meta name="viewport" content="width=device-width">
  19 <style amp-boilerplate>body{-webkit-animation:-amp-start 8s steps(1,end) 0s 1 normal both;-moz-animation:-amp-start 8s steps(1,end) 0s 1 normal both;-ms-animation:-amp-start 8s steps(1,end) 0s 1 normal both;animation:-amp-start 8s steps(1,end) 0s 1 normal both}@-webkit-keyframes -amp-start{from{visibility:hidden}to{visibility:visible}}@-moz-keyframes -amp-start{from{visibility:hidden}to{visibility:visible}}@-ms-keyframes -amp-start{from{visibility:hidden}to{visibility:visible}}@-o-keyframes -amp-start{from{visibility:hidden}to{visibility:visible}}@keyframes -amp-start{from{visibility:hidden}to{visibility:visible}}</style><noscript><style amp-boilerplate>body{-webkit-animation:none;-moz-animation:none;-ms-animation:none;animation:none}</style></noscript>
  20 </head>
  21 <body>
  22 `
  23         boilerplateEnd = `</body>
  24 </html>`
  25 )
  26
  27 const (
  28         // We restrict the amount of text may go inside an HTML element, in
  29         // order to limit the amount a decoder may have to buffer.
  30         elementSizeLimit = 32 * 1024
  31
  32         // The payload is conceptually a long base64-encoded string, but we
  33         // break the string into short chunks separated by whitespace. This is
  34         // to protect against modification by AMP caches, which reportedly may
  35         // truncate long words in text:
  36         // https://bugs.torproject.org/tpo/anti-censorship/pluggable-transports/snowflake/25985#note_2592348
  37         bytesPerChunk = 32
  38
  39         // We set the number of chunks per element so as to stay under
  40         // elementSizeLimit. Here, we assume that there is 1 byte of whitespace
  41         // after each chunk (with an additional whitespace byte at the beginning
  42         // of the element).
  43         chunksPerElement = (elementSizeLimit - 1) / (bytesPerChunk + 1)
  44 )
  45
  46 // The AMP armor encoder is a chain of a base64 encoder (base64.NewEncoder) and
  47 // an HTML element encoder (elementEncoder). A top-level encoder (armorEncoder)
  48 // coordinates these two, and handles prepending and appending the AMP
  49 // boilerplate. armorEncoder's Write method writes data into the base64 encoder,
  50 // where it makes its way through the chain.
  51
  52 // NewEncoder returns a new AMP armor encoder. Anything written to the returned
  53 // io.WriteCloser will be encoded and written to w. The caller must call Close
  54 // to flush any partially written data and output the AMP boilerplate trailer.
  55 func NewEncoder(w io.Writer) (io.WriteCloser, error) {
  56         // Immediately write the AMP boilerplate header.
  57         _, err := w.Write([]byte(boilerplateStart))
  58         if err != nil {
  59                 return nil, err
  60         }
  61
  62         element := &elementEncoder{w: w}
  63         // Write a server–client protocol version indicator, outside the base64
  64         // layer.
  65         _, err = element.Write([]byte("0"))
  66         if err != nil {
  67                 return nil, err
  68         }
  69
  70         base64 := base64.NewEncoder(base64.StdEncoding, element)
  71         return &armorEncoder{
  72                 w:       w,
  73                 element: element,
  74                 base64:  base64,
  75         }, nil
  76 }
  77
  78 type armorEncoder struct {
  79         base64  io.WriteCloser
  80         element *elementEncoder
  81         w       io.Writer
  82 }
  83
  84 func (enc *armorEncoder) Write(p []byte) (int, error) {
  85         // Write into the chain base64 | element | w.
  86         return enc.base64.Write(p)
  87 }
  88
  89 func (enc *armorEncoder) Close() error {
  90         // Close the base64 encoder first, to flush out any buffered data and
  91         // the final padding.
  92         err := enc.base64.Close()
  93         if err != nil {
  94                 return err
  95         }
  96
  97         // Next, close the element encoder, to close any open elements.
  98         err = enc.element.Close()
  99         if err != nil {
 100                 return err
 101         }
 102
 103         // Finally, output the AMP boilerplate trailer.
 104         _, err = enc.w.Write([]byte(boilerplateEnd))
 105         if err != nil {
 106                 return err
 107         }
 108
 109         return nil
 110 }
 111
 112 // elementEncoder arranges written data into pre elements, with the text within
 113 // separated into chunks. It does no HTML encoding, so data written must not
 114 // contain any bytes that are meaningful in HTML.
 115 type elementEncoder struct {
 116         w              io.Writer
 117         chunkCounter   int
 118         elementCounter int
 119 }
 120
 121 func (enc *elementEncoder) Write(p []byte) (n int, err error) {
 122         total := 0
 123         for len(p) > 0 {
 124                 if enc.elementCounter == 0 && enc.chunkCounter == 0 {
 125                         _, err := enc.w.Write([]byte("<pre>\n"))
 126                         if err != nil {
 127                                 return total, err
 128                         }
 129                 }
 130
 131                 n := bytesPerChunk - enc.chunkCounter
 132                 if n > len(p) {
 133                         n = len(p)
 134                 }
 135                 nn, err := enc.w.Write(p[:n])
 136                 if err != nil {
 137                         return total, err
 138                 }
 139                 total += nn
 140                 p = p[n:]
 141
 142                 enc.chunkCounter += n
 143                 if enc.chunkCounter >= bytesPerChunk {
 144                         enc.chunkCounter = 0
 145                         enc.elementCounter += 1
 146                         nn, err = enc.w.Write([]byte("\n"))
 147                         if err != nil {
 148                                 return total, err
 149                         }
 150                         total += nn
 151                 }
 152
 153                 if enc.elementCounter >= chunksPerElement {
 154                         enc.elementCounter = 0
 155                         nn, err = enc.w.Write([]byte("</pre>\n"))
 156                         if err != nil {
 157                                 return total, err
 158                         }
 159                         total += nn
 160                 }
 161         }
 162         return total, nil
 163 }
 164
 165 func (enc *elementEncoder) Close() error {
 166         var err error
 167         if !(enc.elementCounter == 0 && enc.chunkCounter == 0) {
 168                 if enc.chunkCounter == 0 {
 169                         _, err = enc.w.Write([]byte("</pre>\n"))
 170                 } else {
 171                         _, err = enc.w.Write([]byte("\n</pre>\n"))
 172                 }
 173         }
 174         _, err2 := enc.w.Write([]byte(boilerplateEnd))
 175         if err == nil {
 176                 err = err2
 177         }
 178         return err
 179 }