2
0
mirror of https://github.com/boostorg/website.git synced 2026-01-19 04:42:17 +00:00
Files
website/common/code/boost_archive.php
Daniel James c465e8d550 Merge from beta.
Release 1.46.1 and some documentation tweaks.


[SVN r69893]
2011-03-12 15:59:45 +00:00

452 lines
13 KiB
PHP

<?php
/*
Copyright 2005-2008 Redshift Software, Inc.
Distributed under the Boost Software License, Version 1.0.
(See accompanying file LICENSE_1_0.txt or http://www.boost.org/LICENSE_1_0.txt)
*/
require_once(dirname(__FILE__) . '/boost.php');
define('BOOST_DOCS_MODIFIED_DATE', 'Tue, 21 Sep 2010 22:19:30 +0100');
function get_archive_location(
$pattern,
$vpath,
$archive_subdir = true,
$zipfile = true,
$archive_dir = ARCHIVE_DIR,
$archive_file_prefix = ARCHIVE_FILE_PREFIX)
{
$path_parts = array();
preg_match($pattern, $vpath, $path_parts);
$version = $path_parts[1];
$key = $path_parts[2];
$file = ($zipfile ? '' : $archive_dir . '/');
if ($archive_subdir)
{
$file = $file . $archive_file_prefix . $version . '/' . $key;
}
else
{
$file = $file . $archive_file_prefix . $key;
}
$archive = $zipfile ? str_replace('\\','/', $archive_dir . '/' . $version . '.zip') : Null;
return array(
'version' => $version,
'key' => $key,
'file' => $file,
'archive' => $archive,
'zipfile' => $zipfile
);
}
function display_from_archive(
$params,
$content_map = array(),
$override_extractor = null,
$expires = null)
{
$params['template'] = dirname(__FILE__)."/template.php";
$params['title'] = NULL;
$params['charset'] = NULL;
$params['content'] = NULL;
// Check file exists.
if ($params['zipfile'])
{
$check_file = $params['archive'];
if (!is_file($check_file)) {
file_not_found($params, 'Unable to find zipfile.');
return;
}
}
else
{
$check_file = $params['file'];
if (is_dir($check_file))
{
if(substr($check_file, -1) != '/') {
$redirect = resolve_url(basename($check_file).'/');
header("Location: $redirect", TRUE, 301);
return;
}
$found_file = NULL;
if (is_file("$check_file/index.html")) $found_file = 'index.html';
else if (is_file("$check_file/index.htm")) $found_file = 'index.htm';
if ($found_file) {
$params['file'] = $check_file = $check_file.$found_file;
$params['key'] = $params['key'].$found_file;
}
else {
if (!http_headers('text/html', filemtime($check_file), $expires))
return;
return display_dir($params, $check_file);
}
}
else if (!is_file($check_file)) {
file_not_found($params, 'Unable to find file.');
return;
}
}
// Choose filter to use
$info_map = array_merge($content_map, array(
array('@.*@','@[.](txt|py|rst|jam|v2|bat|sh|xml|toyxml|qbk)$@i','text','text/plain'),
array('@.*@','@[.](c|h|cpp|hpp)$@i','cpp','text/plain'),
array('@.*@','@[.]png$@i','raw','image/png'),
array('@.*@','@[.]gif$@i','raw','image/gif'),
array('@.*@','@[.](jpg|jpeg|jpe)$@i','raw','image/jpeg'),
array('@.*@','@[.]css$@i','raw','text/css'),
array('@.*@','@[.]js$@i','raw','application/x-javascript'),
array('@.*@','@[.]pdf$@i','raw','application/pdf'),
array('@.*@','@[.](html|htm)$@i','raw','text/html'),
array('@.*@','@(/|^)(Jamroot|Jamfile|ChangeLog|configure)$@i','text','text/plain'),
array('@.*@','@[.]dtd$@i','raw','application/xml-dtd'),
));
$preprocess = null;
$extractor = null;
$type = null;
foreach ($info_map as $i)
{
if (preg_match($i[1],$params['key']))
{
$extractor = $i[2];
$type = $i[3];
$preprocess = isset($i[4]) ? $i[4] : NULL;
break;
}
}
if ($override_extractor) $extractor = $override_extractor;
if (!$extractor) {
file_not_found($params);
return;
}
// Handle ETags and Last-Modified HTTP headers.
// Output raw files.
if($extractor == 'raw') {
if (!http_headers($type, filemtime($check_file), $expires))
return;
if($_SERVER['REQUEST_METHOD'] != 'HEAD')
display_raw_file($params, $type);
}
else {
if (!http_headers('text/html', filemtime($check_file), $expires))
return;
// Read file from hard drive or zipfile
// Note: this sets $params['content'] with either the content or an error
// message:
if(!extract_file($params, $params['content'])) {
file_not_found($params, $params['content']);
return;
}
// Check if the file contains a redirect.
if($type == 'text/html') {
if($redirect = detect_redirect($params['content'])) {
header("Location: $redirect", TRUE, 301);
if($_SERVER['REQUEST_METHOD'] != 'HEAD') echo $params['content'];
return;
}
}
// Finally process the file and display it.
if($_SERVER['REQUEST_METHOD'] != 'HEAD') {
if ($preprocess) {
$params['content'] = call_user_func($preprocess, $params['content']);
}
echo_filtered($extractor, $params);
}
}
}
// HTTP header handling
function http_headers($type, $last_modified, $expires = null)
{
header('Content-type: '.$type);
switch($type) {
case 'image/png':
case 'image/gif':
case 'image/jpeg':
case 'text/css':
case 'application/x-javascript':
case 'application/pdf':
case 'application/xml-dtd':
header('Expires: '.date(DATE_RFC2822, strtotime("+1 year")));
header('Cache-Control: max-age=31556926'); // A year, give or take a day.
break;
default:
if($expires) {
header('Expires: '.date(DATE_RFC2822, strtotime($expires)));
header('Cache-Control: max-age='.strtotime($expires, 0));
}
break;
}
return conditional_get(max(strtotime(BOOST_DOCS_MODIFIED_DATE), $last_modified));
}
function conditional_get($last_modified)
{
if(!$last_modified) return true;
$last_modified_text = date(DATE_RFC2822, $last_modified);
$etag = '"'.md5($last_modified).'"';
header("Last-Modified: $last_modified_text");
header("ETag: $etag");
$checked = false;
if(isset($_SERVER['HTTP_IF_MODIFIED_SINCE'])) {
$checked = true;
$if_modified_since = strtotime(stripslashes($_SERVER['HTTP_IF_MODIFIED_SINCE']));
if(!$if_modified_since || $if_modified_since < $last_modified)
return true;
}
if(isset($_SERVER['HTTP_IF_NONE_MATCH'])) {
$checked = true;
if(stripslashes($_SERVER['HTTP_IF_NONE_MATCH'] != $etag))
return true;
}
if(!$checked) return true;
header($_SERVER["SERVER_PROTOCOL"].' 304 Not Modified');
return false;
}
// General purpose render callbacks.
function boost_archive_render_callbacks($content, $params) {
$version_title =
str_replace('_', ' ',
preg_replace('@(?<=\d)_(?=\d)@', '.',
ucwords($params['version'])));
$charset = $params['charset'] ? $params['charset'] : 'us-ascii';
$title = $params['title'] ? "$params[title] - $version_title" : 'Boost C++ Libraries';
$head = <<<HTML
<meta http-equiv="Content-Type" content="text/html; charset=${charset}" />
<title>${title}</title>
HTML;
return Array(
'head' => $head,
'content' => $content
);
}
function display_dir($params, $dir)
{
$handle = opendir($dir);
$title = htmlentities("Index listing for $params[key]");
$params['title'] = $title;
$content = "<h3>$title</h3>\n<ul>\n";
while (($file = readdir($handle)) !== false)
{
if (substr($file, 0, 1) == '.') continue;
if (is_dir("$dir$file")) $file .= '/';
$file = htmlentities($file);
$content .= "<li><a href='$file'>$file</a></li>\n";
}
$content .= "</ul>\n";
$params['content'] = $content;
display_template($params['template'], boost_archive_render_callbacks($content, $params));
}
function display_raw_file($params, $type)
{
## header('Content-Disposition: attachment; filename="downloaded.pdf"');
if($params['zipfile']) {
$file_handle = popen(unzip_command($params), 'rb');
fpassthru($file_handle);
$exit_status = pclose($file_handle);
// Don't display errors for a corrupt zip file, as we seemd to
// be getting them for legitimate files.
if($exit_status > 3)
echo 'Error extracting file: '.unzip_error($exit_status);
}
else {
readfile($params['file']);
}
}
function extract_file($params, &$content) {
if($params['zipfile']) {
$file_handle = popen(unzip_command($params),'r');
$text = '';
while ($file_handle && !feof($file_handle)) {
$text .= fread($file_handle,8*1024);
}
$exit_status = pclose($file_handle);
if($exit_status == 0) {
$content = $text;
return true;
}
else {
$content = strstr($_SERVER['HTTP_HOST'], 'beta') ? unzip_error($exit_status) : null;
return false;
}
}
else {
$content = file_get_contents($params['file']);
return true;
}
}
function unzip_command($params) {
return
UNZIP
.' -p '.escapeshellarg($params['archive'])
.' '.escapeshellarg($params['file']);
}
//
// Filters
//
function echo_filtered($extractor, $params) {
require_once(dirname(__FILE__)."/boost_filter_$extractor.php");
$extractor_name = $extractor.'_filter';
call_user_func($extractor_name, $params);
}
/* File Not Found */
function file_not_found($params, $message = null)
{
if(is_string($params)) {
$params = Array(
'file' => $params,
'template' => dirname(__FILE__)."/template.php"
);
}
header($_SERVER["SERVER_PROTOCOL"]." 404 Not Found");
$head = <<<HTML
<meta http-equiv="Content-Type" content="text/html; charset=us-ascii" />
<title>Boost C++ Libraries - 404 Not Found</title>
HTML;
$content = '<h1>404 Not Found</h1><p>File "' . $params['file'] . '" not found.</p><p>';
if($params['zipfile']) $content .= "Unzip error: ";
$content .= htmlentities($message);
$content .= '</p>';
display_template($params['template'], Array('head' => $head, 'content' => $content));
}
/*
* HTML processing functions
*/
function detect_redirect($content)
{
// Only check small files, since larger files are never redirects, and are
// expensive to search.
if(strlen($content) <= 2000 &&
preg_match(
'@<meta\s+http-equiv\s*=\s*["\']?refresh["\']?\s+content\s*=\s*["\']0;\s*URL=([^"\']*)["\']\s*/?>@i',
$content, $redirect))
{
return resolve_url($redirect[1]);
}
return false;
}
// Not a full implementation. Just good enough for redirecting.
function resolve_url($url) {
$url = parse_url($url);
if(isset($url['schme'])) return $url;
$url['scheme'] = 'http'; # Detect other schemes?
if(!isset($url['host'])) {
$url['host'] = $_SERVER['HTTP_HOST'];
if($url['path'][0] != '/') {
$path = explode('/', $_SERVER['REQUEST_URI']);
array_pop($path);
$rel_path = explode('/', $url['path']);
while(isset($rel_path[0]) && $rel_path[0] == '..') {
array_pop($path);
array_shift($rel_path);
}
$url['path'] = implode('/', $path).'/'.implode('/', $rel_path);
}
}
return $url['scheme'].'://'.$url['host'] . $url['path'];
}
// Display the content in the standard boost template
function display_template($template, $_file) {
include($template);
}
// Return a readable error message for unzip exit state.
function unzip_error($exit_status) {
switch($exit_status) {
case 0: return 'No error.';
case 1: return 'One or more warning errors were encountered.';
case 2: return 'A generic error in the zipfile format was detected.';
case 3: return 'A severe error in the zipfile format was detected.';
case 4: return 'Unzip was unable to allocate memory for one or more buffers during program initialization.';
case 5: return 'Unzip was unable to allocate memory or unable to obtain a tty to read the decryption password(s).';
case 6: return 'Unzip was unable to allocate memory during decompression to disk.';
case 7: return 'Unzip was unable to allocate memory during in-memory decompression.';
case 9: return 'The specified zipfile was not found.';
case 10: return 'Invalid options were specified on the command line.';
case 11: return 'No matching files were found.';
case 50: return 'The disk is (or was) full during extraction.';
case 51: return 'The end of the ZIP archive was encountered prematurely.';
case 80: return 'The user aborted unzip prematurely with control-C (or similar).';
case 81: return 'Testing or extraction of one or more files failed due to unsupported compression methods or unsupported decryption.';
case 82: return 'No files were found due to bad decryption password(s).';
default: return 'Unknown unzip error code: ' + $exit_status;
}
}