Reincarnation

Load Rails Environment with Ruby 1.9 Nearly as Fast as Ruby 1.8

2011-09-17T00:44:00+08:00

Use Xavier Shay’s patched Ruby 1.9.3dev

There were two patches for Ruby 1.9 to resolve the load performance issue. Xavier Shay’s approach, which use a hash data structure to store loaded files, absolutely can result much better performance.

Use Xavier Shay’s patch may not easy, here is a small trick to install his fork of Ruby with rvm:

Shell commands

$ cd ~/.rvm/repos
$ git clone git://github.com/xaviershay/ruby.git ruby-head
$ rvm install ruby-head --branch require-performance-fix

$ rvm use ruby-head
$ ruby --version
ruby 1.9.3dev (2011-05-31 trunk 31827) [x86_64-linux]

Yeah, it will be installed as ruby-head.

Note: Absolutely this should only be used in development environment, since it’s a development version of Ruby 1.9.3.

Updated on Sep 28, 2011:

The official Ruby 1.9.3 rc1 was released recently, it has almost same load performance compared to Xavier Shay’s patched Ruby 1.9.3dev. Which make this trick no longer necessary.

Tweak Gemfile to not require unnecessary gems immediately

Most gems in development and test group are unnecessary to be required immediately, gems for test can be required in spec_helper.rb. Here is part of my Gemfile for a small project:

Gemfile

group :development do
  gem 'bond', require: nil
  gem 'irbtools', require: nil
  gem 'irb_rocket', require: nil
  gem 'capistrano', require: nil
  gem 'rails3-generators', require: nil
  # Only used for mo/po file generation in development, see rake -T gettext.
  gem 'gettext', '>= 1.9.3', require: nil
end

group :development, :test do
  gem 'awesome_print', require: 'ap'
  gem 'factory_girl_rails'
  gem 'rspec-rails', '>= 2.5.0', require: nil
  gem 'capybara', '>= 0.3.6', require: nil
  gem 'capybara-webkit', '>= 1.0.0.beta4', require: nil
  gem 'launchy', require: nil
  gem 'cucumber-rails', require: nil
end

group :test do
  gem 'spork', '>= 0.9.0.rc5', require: nil
  gem 'rspec', '>= 2.5.0', require: nil
  gem 'remarkable', '>= 4.0.0.alpha4', require: nil
  gem 'remarkable_activemodel', '>= 4.0.0.alpha4', require: nil
  gem 'remarkable_mongoid', require: nil
  gem 'database_cleaner', '>= 0.5.0', require: nil

  gem 'guard', require: nil
  gem 'guard-bundler', require: nil
  gem 'guard-spork', require: nil
  gem 'guard-rspec', require: nil
  gem 'guard-cucumber', require: nil
end

group :linux do
  gem 'rb-inotify', require: nil
  gem 'libnotify', require: nil
  gem 'therubyracer', require: nil
end

How about the result?

My recent project using Xavier Shay’s Ruby 1.9.3dev:

$ time rake about
/home/rainux/.rvm/gems/ruby-head/gems/activesupport-3.0.10/lib/active_support/dependencies.rb:239:in `block in require': iconv will be deprecated in the future, use String#encode instead.
About your application's environment
Ruby version              1.9.3 (x86_64-linux)
RubyGems version          1.8.10
Rack version              1.2
Rails version             3.0.10
Action Pack version       3.0.10
Active Resource version   3.0.10
Action Mailer version     3.0.10
Active Support version    3.0.10
Middleware                ActionDispatch::Static, Rack::Lock, ActiveSupport::Cache::Strategy::LocalCache, Rack::Runtime, Rails::Rack::Logger, ActionDispatch::ShowExceptions, ActionDispatch::RemoteIp, Rack::Sendfile, ActionDispatch::Callbacks, ActionDispatch::Cookies, ActionDispatch::Session::CookieStore, ActionDispatch::Flash, ActionDispatch::ParamsParser, Rack::MethodOverride, ActionDispatch::Head, ActionDispatch::BestStandardsSupport, Warden::Manager, Sass::Plugin::Rack, Rack::Mongoid::Middleware::IdentityMap, Barista::Filter, Barista::Server::Proxy
Application root          /home/rainux/devel/a_rails_app
Environment               development
rake about  5.15s user 0.35s system 99% cpu 5.501 total

Compare to another project using Ruby 1.8.7 (actually, ree-1.8.7-2011.03), without tweak Gemfile:

$ time rake about
About your application's environment
Ruby version              1.8.7 (x86_64-linux)
RubyGems version          1.6.2
Rack version              1.2
Rails version             3.0.9
Active Record version     3.0.9
Action Pack version       3.0.9
Active Resource version   3.0.9
Action Mailer version     3.0.9
Active Support version    3.0.9
Application root          /home/rainux/devel/yet_another_rails_app
Environment               development
rake about  3.65s user 0.92s system 99% cpu 4.585 total

Configure capybara-webkit to Run Acceptance Specs With Javascript/AJAX

2011-07-23T20:09:00+08:00

Add capybara-webkit to your Gemfile and let Guard::Bundler install it automatically (or manually via bundle install if you don’t use Guard).

Gemfile

gem 'capybara-webkit', '>= 1.0.0.beta4'

Set Javascript driver to :webkit for Capybara in spec_helper.rb.

spec_helper.rb

Capybara.javascript_driver = :webkit

Configure RSpec use non-transactional fixtures, configure Database Cleaner in spec_helper.rb.

Notice with this setup, we’ll only use truncation strategy when driver is not :rack_test. this will make normal specs run faster.

spec_helper.rb

config.use_transactional_fixtures = false

config.before :each do
  if Capybara.current_driver == :rack_test
    DatabaseCleaner.strategy = :transaction
    DatabaseCleaner.start
  else
    DatabaseCleaner.strategy = :truncation
  end
end

config.after :each do
  DatabaseCleaner.clean
end

Tag your scenarios in spec/acceptance/*_spec.rb to use Javascript driver if necessary.

some_spec.rb

scenario 'Create a lolita via AJAX', :js => true do
end

Wait for any AJAX call to be completed in your specs. This is very important, or you will get many strange issues like no database record found, AJAX call get empty response with 0 status code, etc.

For example if you have a simple AJAX form, the success callback will simply redirect browser to another page via location.href = '/yet_another_page';. You can use the following code to wait for it done.

another_spec.rb

scenario 'Create a lolita via AJAX', :js => true do
  visit new_lolita_path

  click_on 'Submit'

  wait_until { page.current_path == lolita_path(Lolita.last) }

  # Your expections for the new page
end

Google 拼音输入法 Android 版 MOD v3

2011-02-03T14:44:00+08:00

前段时间入手 T-Mobile G2 之后把 Google 拼音输入法升级到了在 G1 上跑起来慢吞吞的最新版，结果遇到各种小问题用得很不爽，一怒之下抄起家伙把它狠狠改了一通。虽然还有些小问题，但影响没那么严重就暂时懒得管了。

懒得另外写说明了，直接引用 GitHub repo 里的 README 吧。

目的

使用 apktool 对 Google 拼音输入法 Android 版进行反向工程[注1]，在 Dalvik JVM 汇编层级微调修改它，使其更适合日常使用。

注1：实际上不是真正完整的反向工程，后来有大量借助 Android Open Source Project 中 Google 拼音输入法 1.0.0 的 Java 源代码理解反汇编得到的 smali 代码。

问题修正

中文模式下硬件键盘上某些标点符号没有被正确映射为中文（全角）形态，尤其是在 T-Mobile G2 上。
中文模式下在软键盘未初始化时使用硬件键盘 Enter 键会导致 Google 拼音崩溃。
中文模式下软键盘上的圆括号不是中文（全角）形态。
软键盘上的 & 和 < 符号被不正确地转义成了 & 和 <。（MOD 版引入的问题。）
中文模式下软键盘在空闲状态时 Del 不能删除字符。（MOD 版引入的问题。）

细节改进

中文模式空闲和联想状态下 Alt + Del 像系统默认行为一样删除当前行。
中文模式选字状态下 Alt + Del 删除所有拼音字符并回到空闲状态。
中文模式下回到空闲状态以及空闲状态输入字符后重置 Alt 和 Shift 状态，避免使用 Del 修正输入内容时误删当前行。（MOD 版引入的不便。）
中文模式下硬件键盘 _ （下划线）映射为—（半个破折号），`（反单引号）映射为·（英文人名分隔符）。

如何使用

下载 Google 拼音输入法 1.3.4 MOD v3，使用 adb 工具或 Android 的 Package Manager 安装。或者 git clone 此 smali 源代码仓库自己用 apktool 编译生成 .apk 文件。

注意：安装 MOD 版本之前必须先卸载官方版本，这是因为 MOD 版本签名所使用的证书不可能与官方版本一致。

感谢

@pipitu: 在我头昏眼花犯下低级错误的时候帮我查阅 smali 资料，让我得以“拨云见日醍醐灌顶茅塞顿开遍体舒畅的神一般的展开”。