1
0
mirror of https://github.com/golang/go synced 2024-11-19 13:54:56 -07:00
go/doc/htmlgen.go
Rob Pike 9cf37c3723 docs: fold the prog.sh scripting from makehtml into htmlgen itself.
This allows us to drop some crufty scripting and provides a firmer
footing for building better tools for preparing documents with source
code inside.

Also eliminate line numbers from the examples and text.

R=golang-dev, adg
CC=golang-dev
https://golang.org/cl/4650069
2011-07-04 16:15:14 +10:00

313 lines
7.1 KiB
Go

// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// If --html is set, process plain text into HTML.
// - h2's are made from lines followed by a line "----\n"
// - tab-indented blocks become <pre> blocks with the first tab deleted
// - blank lines become <p> marks (except inside <pre> tags)
// - "quoted strings" become <code>quoted strings</code>
// Lines beginning !src define pieces of program source to be
// extracted from other files and injected as <pre> blocks.
// The syntax is simple: 1, 2, or 3 space-separated arguments:
//
// Whole file:
// !src foo.go
// One line (here the signature of main):
// !src foo.go /^func.main/
// Block of text, determined by start and end (here the body of main):
// !src foo.go /^func.main/ /^}/
//
// Patterns can be /regular.expression/, a decimal number, or $
// to signify the end of the file.
// TODO: the regular expression cannot contain spaces; does this matter?
package main
import (
"bufio"
"bytes"
"flag"
"fmt"
"io/ioutil"
"log"
"os"
"regexp"
"strconv"
"strings"
"template"
)
var (
html = flag.Bool("html", true, "process text into HTML")
)
var (
// lines holds the input and is reworked in place during processing.
lines = make([][]byte, 0, 20000)
empty = []byte("")
newline = []byte("\n")
tab = []byte("\t")
quote = []byte(`"`)
indent = []byte(" ")
sectionMarker = []byte("----\n")
preStart = []byte("<pre>")
preEnd = []byte("</pre>\n")
pp = []byte("<p>\n")
srcPrefix = []byte("!src")
)
func main() {
flag.Parse()
read()
programs()
if *html {
headings()
coalesce(preStart, foldPre)
coalesce(tab, foldTabs)
paragraphs()
quotes()
}
write()
}
// read turns standard input into a slice of lines.
func read() {
b := bufio.NewReader(os.Stdin)
for {
line, err := b.ReadBytes('\n')
if err == os.EOF {
break
}
if err != nil {
log.Fatal(err)
}
lines = append(lines, line)
}
}
// write puts the result on standard output.
func write() {
b := bufio.NewWriter(os.Stdout)
for _, line := range lines {
b.Write(expandTabs(line))
}
b.Flush()
}
// programs injects source code from !src invocations.
func programs() {
nlines := make([][]byte, 0, len(lines)*3/2)
for _, line := range lines {
if bytes.HasPrefix(line, srcPrefix) {
line = trim(line)[len(srcPrefix):]
prog := srcCommand(string(line))
if *html {
nlines = append(nlines, []byte(fmt.Sprintf("<pre><!--%s\n-->", line)))
}
for _, l := range prog {
nlines = append(nlines, htmlEscape(l))
}
if *html {
nlines = append(nlines, preEnd)
}
} else {
nlines = append(nlines, line)
}
}
lines = nlines
}
// srcCommand processes one !src invocation.
func srcCommand(command string) [][]byte {
// TODO: quoted args so we can have 'a b'?
args := strings.Fields(command)
if len(args) == 0 || len(args) > 3 {
log.Fatal("bad syntax for src command: %s", command)
}
file := args[0]
lines := bytes.SplitAfter(readFile(file), newline)
// File plus zero args: whole file:
// !src file.go
if len(args) == 1 {
return lines
}
start := match(file, 0, lines, string(args[1]))
// File plus one arg: one line:
// !src file.go /foo/
if len(args) == 2 {
return [][]byte{lines[start]}
}
// File plus two args: range:
// !src file.go /foo/ /^}/
end := match(file, start, lines, string(args[2]))
return lines[start : end+1] // +1 to include matched line.
}
// htmlEscape makes sure input is HTML clean, if necessary.
func htmlEscape(input []byte) []byte {
if !*html || bytes.IndexAny(input, `&"<>`) < 0 {
return input
}
var b bytes.Buffer
template.HTMLEscape(&b, input)
return b.Bytes()
}
// readFile reads and returns a file as part of !src processing.
func readFile(name string) []byte {
file, err := ioutil.ReadFile(name)
if err != nil {
log.Fatal(err)
}
return file
}
// match identifies the input line that matches the pattern in a !src invocation.
// If start>0, match lines starting there rather than at the beginning.
func match(file string, start int, lines [][]byte, pattern string) int {
// $ matches the end of the file.
if pattern == "$" {
return len(lines) - 1
}
// Number matches the line.
if i, err := strconv.Atoi(pattern); err == nil {
return i - 1 // Lines are 1-indexed.
}
// /regexp/ matches the line that matches the regexp.
if len(pattern) > 2 && pattern[0] == '/' && pattern[len(pattern)-1] == '/' {
re, err := regexp.Compile(pattern[1 : len(pattern)-1])
if err != nil {
log.Fatal(err)
}
for i := start; i < len(lines); i++ {
if re.Match(lines[i]) {
return i
}
}
log.Fatalf("%s: no match for %s", file, pattern)
}
log.Fatalf("unrecognized pattern: %s", pattern)
return 0
}
// coalesce combines lines. Each time prefix is found on a line,
// it calls fold and replaces the line with return value from fold.
func coalesce(prefix []byte, fold func(i int) (n int, line []byte)) {
j := 0 // output line number goes up by one each loop
for i := 0; i < len(lines); {
if bytes.HasPrefix(lines[i], prefix) {
nlines, block := fold(i)
lines[j] = block
i += nlines
} else {
lines[j] = lines[i]
i++
}
j++
}
lines = lines[0:j]
}
// foldPre returns the <pre> block as a single slice.
func foldPre(i int) (n int, line []byte) {
buf := new(bytes.Buffer)
for i < len(lines) {
buf.Write(lines[i])
n++
if bytes.Equal(lines[i], preEnd) {
break
}
i++
}
return n, buf.Bytes()
}
// foldTabs returns the tab-indented block as a single <pre>-bounded slice.
func foldTabs(i int) (n int, line []byte) {
buf := new(bytes.Buffer)
buf.WriteString("<pre>\n")
for i < len(lines) {
if !bytes.HasPrefix(lines[i], tab) {
break
}
buf.Write(lines[i][1:]) // delete leading tab.
n++
i++
}
buf.WriteString("</pre>\n")
return n, buf.Bytes()
}
// headings turns sections into HTML sections.
func headings() {
b := bufio.NewWriter(os.Stdout)
for i, l := range lines {
if i > 0 && bytes.Equal(l, sectionMarker) {
lines[i-1] = []byte("<h2>" + string(trim(lines[i-1])) + "</h2>\n")
lines[i] = empty
}
}
b.Flush()
}
// paragraphs turns blank lines into paragraph marks.
func paragraphs() {
for i, l := range lines {
if bytes.Equal(l, newline) {
lines[i] = pp
}
}
}
// quotes turns "x" in the file into <code>x</code>.
func quotes() {
for i, l := range lines {
lines[i] = codeQuotes(l)
}
}
// quotes turns "x" in the line into <code>x</code>.
func codeQuotes(l []byte) []byte {
if bytes.HasPrefix(l, preStart) {
return l
}
n := bytes.Index(l, quote)
if n < 0 {
return l
}
buf := new(bytes.Buffer)
inQuote := false
for _, c := range l {
if c == '"' {
if inQuote {
buf.WriteString("</code>")
} else {
buf.WriteString("<code>")
}
inQuote = !inQuote
} else {
buf.WriteByte(c)
}
}
return buf.Bytes()
}
// trim drops the trailing newline, if present.
func trim(l []byte) []byte {
n := len(l)
if n > 0 && l[n-1] == '\n' {
return l[0 : n-1]
}
return l
}
// expandTabs expands tabs to spaces. It doesn't worry about columns.
func expandTabs(l []byte) []byte {
return bytes.Replace(l, tab, indent, -1)
}