ruby/tool/downloader.rb

393 строки
11 KiB
Ruby
Исходник Обычный вид История

2019-05-28 08:20:42 +03:00
# Used by configure and make to download or update mirrored Ruby and GCC
# files. This will use HTTPS if possible, falling back to HTTP.
require 'fileutils'
require 'open-uri'
require 'pathname'
begin
require 'net/https'
rescue LoadError
https = 'http'
else
https = 'https'
# open-uri of ruby 2.2.0 accepts an array of PEMs as ssl_ca_cert, but old
# versions do not. so, patching OpenSSL::X509::Store#add_file instead.
class OpenSSL::X509::Store
alias orig_add_file add_file
def add_file(pems)
Array(pems).each do |pem|
if File.directory?(pem)
add_path pem
else
orig_add_file pem
end
end
end
end
# since open-uri internally checks ssl_ca_cert using File.directory?,
# allow to accept an array.
class <<File
alias orig_directory? directory?
def File.directory? files
files.is_a?(Array) ? false : orig_directory?(files)
end
end
end
class Downloader
def self.https=(https)
@@https = https
end
def self.https?
@@https == 'https'
end
def self.https
@@https
end
class GNU < self
def self.download(name, *rest)
if https?
super("https://raw.githubusercontent.com/gcc-mirror/gcc/master/#{name}", name, *rest)
else
super("https://repo.or.cz/official-gcc.git/blob_plain/HEAD:/#{name}", name, *rest)
end
end
end
class RubyGems < self
def self.download(name, dir = nil, since = true, options = {})
require 'rubygems'
options = options.dup
options[:ssl_ca_cert] = Dir.glob(File.expand_path("../lib/rubygems/ssl_certs/**/*.pem", File.dirname(__FILE__)))
super("https://rubygems.org/downloads/#{name}", name, dir, since, options)
end
end
Gems = RubyGems
class Unicode < self
INDEX = {} # cache index file information across files in the same directory
UNICODE_PUBLIC = "http://www.unicode.org/Public/"
def self.download(name, dir = nil, since = true, options = {})
options = options.dup
unicode_beta = options.delete(:unicode_beta)
name_dir_part = name.sub(/[^\/]+$/, '')
if unicode_beta == 'YES'
if INDEX.size == 0
index_options = options.dup
index_options[:cache_save] = false # TODO: make sure caching really doesn't work for index file
index_data = File.read(under(dir, "index.html")) rescue nil
index_file = super(UNICODE_PUBLIC+name_dir_part, "#{name_dir_part}index.html", dir, true, index_options)
INDEX[:index] = File.read(index_file)
since = true unless INDEX[:index] == index_data
end
file_base = File.basename(name, '.txt')
return if file_base == '.' # Use pre-generated headers and tables
beta_name = INDEX[:index][/#{Regexp.quote(file_base)}(-[0-9.]+d\d+)?\.txt/]
# make sure we always check for new versions of files,
# because they can easily change in the beta period
super(UNICODE_PUBLIC+name_dir_part+beta_name, name, dir, since, options)
else
index_file = Pathname.new(under(dir, name_dir_part+'index.html'))
2019-05-24 13:01:39 +03:00
if index_file.exist? and name_dir_part !~ /^(12\.1\.0|emoji\/12\.0)/
raise "Although Unicode is not in beta, file #{index_file} exists. " +
"Remove all files in this directory and in .downloaded-cache/ " +
"because they may be leftovers from the beta period."
end
super(UNICODE_PUBLIC+name, name, dir, since, options)
end
end
end
def self.mode_for(data)
/\A#!/ =~ data ? 0755 : 0644
end
def self.http_options(file, since)
options = {}
if since
case since
when true
since = (File.mtime(file).httpdate rescue nil)
when Time
since = since.httpdate
end
if since
options['If-Modified-Since'] = since
end
end
options['Accept-Encoding'] = 'identity' # to disable Net::HTTP::GenericRequest#decode_content
options
end
# Downloader.download(url, name, [dir, [since]])
#
# Update a file from url if newer version is available.
# Creates the file if the file doesn't yet exist; however, the
# directory where the file is being created has to exist already.
# The +since+ parameter can take the following values, with associated meanings:
# true ::
# Take the last-modified time of the current file on disk, and only download
# if the server has a file that was modified later. Download unconditionally
# if we don't have the file yet. Default.
# +some time value+ ::
# Use this time value instead of the time of modification of the file on disk.
# nil ::
# Only download the file if it doesn't exist yet.
# false ::
# always download url regardless of whether we already have a file,
# and regardless of modification times. (This is essentially just a waste of
# network resources, except in the case that the file we have is somehow damaged.
# Please note that using this recurringly might create or be seen as a
# denial of service attack.)
#
# Example usage:
# download 'http://www.unicode.org/Public/UCD/latest/ucd/UnicodeData.txt',
# 'UnicodeData.txt', 'enc/unicode/data'
def self.download(url, name, dir = nil, since = true, options = {})
options = options.dup
url = URI(url)
dryrun = options.delete(:dryrun)
options.delete(:unicode_beta) # just to be on the safe side for gems and gcc
if name
file = Pathname.new(under(dir, name))
else
name = File.basename(url.path)
end
cache_save = options.delete(:cache_save) {
ENV["CACHE_SAVE"] != "no"
}
cache = cache_file(url, name, options.delete(:cache_dir))
file ||= cache
if since.nil? and file.exist?
if $VERBOSE
$stdout.puts "#{file} already exists"
$stdout.flush
end
if cache_save
save_cache(cache, file, name)
end
2019-05-28 06:54:20 +03:00
return file.respond_to?(:to_path) ? file.to_path : file.to_s
end
if dryrun
puts "Download #{url} into #{file}"
return
end
if link_cache(cache, file, name, $VERBOSE)
return file.to_path
end
if !https? and URI::HTTPS === url
warn "*** using http instead of https ***"
url.scheme = 'http'
url = URI(url.to_s)
end
if $VERBOSE
$stdout.print "downloading #{name} ... "
$stdout.flush
end
begin
data = with_retry(9) do
url.read(options.merge(http_options(file, since.nil? ? true : since)))
end
rescue OpenURI::HTTPError => http_error
if http_error.message =~ /^304 / # 304 Not Modified
if $VERBOSE
$stdout.puts "#{name} not modified"
$stdout.flush
end
return file.to_path
end
raise
rescue Timeout::Error
if since.nil? and file.exist?
puts "Request for #{url} timed out, using old version."
return file.to_path
end
raise
rescue SocketError
if since.nil? and file.exist?
puts "No network connection, unable to download #{url}, using old version."
return file.to_path
end
raise
end
mtime = nil
dest = (cache_save && cache && !cache.exist? ? cache : file)
dest.parent.mkpath
dest.open("wb", 0600) do |f|
f.write(data)
f.chmod(mode_for(data))
mtime = data.meta["last-modified"]
end
if mtime
mtime = Time.httpdate(mtime)
dest.utime(mtime, mtime)
end
if $VERBOSE
$stdout.puts "done"
$stdout.flush
end
if dest.eql?(cache)
link_cache(cache, file, name)
elsif cache_save
save_cache(cache, file, name)
end
return file.to_path
rescue => e
raise "failed to download #{name}\n#{e.class}: #{e.message}: #{url}"
end
def self.under(dir, name)
dir ? File.join(dir, File.basename(name)) : name
end
def self.cache_file(url, name, cache_dir = nil)
case cache_dir
when false
return nil
when nil
cache_dir = ENV['CACHE_DIR']
if !cache_dir or cache_dir.empty?
cache_dir = ".downloaded-cache"
end
end
Pathname.new(cache_dir) + (name || File.basename(URI(url).path))
end
def self.link_cache(cache, file, name, verbose = false)
return false unless cache and cache.exist?
return true if cache.eql?(file)
if /cygwin/ !~ RUBY_PLATFORM or /winsymlink:nativestrict/ =~ ENV['CYGWIN']
begin
file.make_symlink(cache.relative_path_from(file.parent))
rescue SystemCallError
else
if verbose
$stdout.puts "made symlink #{name} to #{cache}"
$stdout.flush
end
return true
end
end
begin
file.make_link(cache)
rescue SystemCallError
else
if verbose
$stdout.puts "made link #{name} to #{cache}"
$stdout.flush
end
return true
end
end
def self.save_cache(cache, file, name)
return unless cache or cache.eql?(file)
begin
st = cache.stat
rescue
begin
file.rename(cache)
rescue
return
end
else
return unless st.mtime > file.lstat.mtime
file.unlink
end
link_cache(cache, file, name)
end
def self.with_retry(max_times, &block)
times = 0
begin
block.call
rescue Errno::ETIMEDOUT, SocketError, OpenURI::HTTPError, Net::ReadTimeout, Net::OpenTimeout => e
raise if e.is_a?(OpenURI::HTTPError) && e.message !~ /^50[023] / # retry only 500, 502, 503 for http error
times += 1
if times <= max_times
$stderr.puts "retrying #{e.class} (#{e.message}) after #{times ** 2} seconds..."
sleep(times ** 2)
retry
else
raise
end
end
end
private_class_method :with_retry
end
Downloader.https = https.freeze
if $0 == __FILE__
since = true
options = {}
until ARGV.empty?
case ARGV[0]
when '-d'
destdir = ARGV[1]
ARGV.shift
when '-p'
# strip directory names from the name to download, and add the
# prefix instead.
prefix = ARGV[1]
ARGV.shift
when '-e'
since = nil
when '-a'
since = false
when '-n', '--dryrun'
options[:dryrun] = true
when '--cache-dir'
options[:cache_dir] = ARGV[1]
ARGV.shift
when '--unicode-beta'
options[:unicode_beta] = ARGV[1]
ARGV.shift
when /\A--cache-dir=(.*)/m
options[:cache_dir] = $1
when /\A-/
abort "#{$0}: unknown option #{ARGV[0]}"
else
break
end
ARGV.shift
end
dl = Downloader.constants.find do |name|
ARGV[0].casecmp(name.to_s) == 0
end unless ARGV.empty?
$VERBOSE = true
if dl
dl = Downloader.const_get(dl)
ARGV.shift
ARGV.each do |name|
dir = destdir
if prefix
name = name.sub(/\A\.\//, '')
destdir2 = destdir.sub(/\A\.\//, '')
if name.start_with?(destdir2+"/")
name = name[(destdir2.size+1)..-1]
if (dir = File.dirname(name)) == '.'
dir = destdir
else
dir = File.join(destdir, dir)
end
else
name = File.basename(name)
end
name = "#{prefix}/#{name}"
end
dl.download(name, dir, since, options)
end
else
abort "usage: #{$0} url name" unless ARGV.size == 2
Downloader.download(ARGV[0], ARGV[1], destdir, since, options)
end
end