Copious Free Time

Packaging an Application With Crate (2008-11-30)

Crate is way to package up your ruby applications as statically compiled binaries. I was lucky enough to talk about Crate at RubyConf '08.

This is a small tutorial that expands upon my RubyConf talk and demonstrates how to package gem application as a standalone statically compiled executable.

Step 1 — Install Crate

To get started, you'll need to install crate. It is distributed mainly as a gem from rubyforge and installs in the standard way.

gem install crate

Crate is part of my copiousfreetime rubyforge project and its git repository is available on github. Patches are always welcome.

• crate on github
• public clone url - <git://github.com/copiousfreetime/crate.git>

Step 2 — Create a new Crate Application

In this tutorial we are going to package up the htpassword-ruby commandline application in my htauth gem as a standalone executable.

% crate -v
Crate 0.2.1

% crate htpassword-ruby
[16:39:00]  INFO: creating htpasswd-ruby
[16:39:00]  INFO: creating htpasswd-ruby/Rakefile
[16:39:00]  INFO: creating htpasswd-ruby/crate_boot.c
[16:39:00]  INFO: creating htpasswd-ruby/recipes/amalgalite
[16:39:00]  INFO: creating htpasswd-ruby/recipes/amalgalite/amalgalite.rake
[16:39:00]  INFO: creating htpasswd-ruby/recipes/arrayfields
[16:39:00]  INFO: creating htpasswd-ruby/recipes/arrayfields/arrayfields.rake
[16:39:00]  INFO: creating htpasswd-ruby/recipes/configuration
[16:39:00]  INFO: creating htpasswd-ruby/recipes/configuration/configuration.rake
[16:39:00]  INFO: creating htpasswd-ruby/recipes/openssl
[16:39:00]  INFO: creating htpasswd-ruby/recipes/openssl/openssl.rake
[16:39:00]  INFO: creating htpasswd-ruby/recipes/ruby
[16:39:00]  INFO: creating htpasswd-ruby/recipes/ruby/ext-extmk.rb.patch
[16:39:00]  INFO: creating htpasswd-ruby/recipes/ruby/ruby.rake
[16:39:00]  INFO: creating htpasswd-ruby/recipes/rubygems
[16:39:00]  INFO: creating htpasswd-ruby/recipes/rubygems/rubygems.rake
[16:39:00]  INFO: creating htpasswd-ruby/recipes/zlib
[16:39:00]  INFO: creating htpasswd-ruby/recipes/zlib/zlib.rake

% cd htpassword-ruby

This created a new directory structure to do the building of our standalone htpassword-ruby application. Change into the new htpassword-ruby directory now and run rake -T to see the full set of tasks that are available. Most of them you will not use. The two you want to pay attention to are default and ruby.

Step 3 — Integrate HTAuth Dependency Targets

The way all of this is integrated together is a final single binary with the name htpassword-ruby which is an embedded ruby interpreter wrapped up in a thin C application. The C application is in the crate_boot.c file you'll see in the top level of the project directory.

The pure ruby code involved in the whole system is stored in SQLite databases. This will included the ruby standard library and your application code. As part of a Crate application, the require statement is overwritten to load from rows in an SQLite database instead of the file system. This feature is all from the amalgalite gem and anyone case use it for other projects if they so choose.

In the mean time, we now have a build system setup to build ruby, but not our application. There is no facility as of yet in Crate to automatically add build targets for a gem, but they are in the works for a later version. In the mean time we'll need to roll our own, its not that hard. Currently I'm not completely satisfied with the way this integrates and will most likely change it in a future release.

The HTAuth gem is pretty self contained, it only has one dependency, highline. We're going to add two new recipes, one for highline and one for htauth.

% mkdir recipes/highline
% vi recipes/highline/highline.rake

We edit it to be the following:

#
# The recipe for integrating highline into the ruby build
#
Crate::GemIntegration.new("highline", "1.5.0") do |t|
  t.upstream_source = "http://rubyforge.org/frs/download.php/46328/highline-1.5.0.gem"
end

And we do the same for htauth:

% mkdir recipes/htauth
% vi recipes/htauth/htauth.rake

#
# The recipe for integrating htauth into the ruby build
#
Crate::GemIntegration.new("htauth", "1.0.2") do |t|
  t.upstream_source = "http://rubyforge.org/frs/download.php/47663/htauth-1.0.2.gem"
end

And finally we we integrate these two new build targets into the final ruby build. Edit the recipes/ruby/ruby.rake file and add the two noted lines.

Crate::Ruby.new( "ruby", "1.8.6-p287") do |t| 
  t.depends_on( "openssl" )
  t.depends_on( "zlib" )

  t.integrates( "amalgalite" )
  t.integrates( "arrayfields" )
  t.integrates( "configuration" )

  t.integrates( "highline" )    # Add this line
  t.integrates( "htauth" )      # and this line

  t.upstream_source  = "ftp://ftp.ruby-lang.org/pub/ruby/1.8/ruby-1.8.6-p287.tar.gz"
  t.upstream_md5     = "f6cd51001534ced5375339707a757556"

  ENV["CPPFLAGS"]= "-I#{File.join( t.install_dir, 'usr', 'include')}"
  ENV["LDFLAGS"] = "-L#{File.join( t.install_dir, 'usr', 'lib' )}" 

  def t.build
    # put the .a files from the fakeroot/usr/lib directory into the package
    # directory so the compilation can use them
    %w[ libz.a libcrypto.a libssl.a ].each do |f| 
      FileUtils.cp File.join( install_dir, "usr", "lib", f ), pkg_dir
    end 
    sh "./configure --disable-shared --prefix=#{File.join( '/', 'usr' )}" 
    sh "make"
  end

  t.install_commands << "make install DESTDIR=#{t.install_dir}"

end

Now we are setup to do the two big build steps.

Step 4 — Building Ruby as a Static Library

In order to build a static ruby based application we first need to build ruby itself as a static library. In doing so we also need to build all the binary dependencies of ruby statically. In this case, that includes zlib, openssl and amalgalite.

Fortunately, all that is taken care of for you in the rake tasks. All you need to do at this point is run the ruby rake task. It will download, unpack and build all the various components. This may take a while, so take a break and get something to drink.

% rake ruby
# snip lots of output, look in project.log for the full output.  
....
23:54:58  INFO: Bulding zlib 1.2.3
....
23:55:30  INFO: Bulding openssl 0.9.8i
...
00:01:22  INFO: Integrating highline into ruby's tree
...
00:01:24  INFO: Integrating htauth into ruby's tree
...
00:01:24  INFO: Bulding ruby 1.8.6-p287
...
00:02:59  INFO: ruby 1.8.6-p287 is installed

You should now have a ruby executable in fakeroot/usr/bin/ruby. You can run this script to prove that it is statically compiled. Notice that there are no require statements. This script will not run on a normal ruby, it will only work with one that is statically compiled.

% cat versions.rb
puts "zlib    : #{Zlib::zlib_version}"
puts "OpenSSL : #{OpenSSL::OPENSSL_VERSION}"
puts "SQLite  : #{Amalgalite::SQLite3::Version}"

% /opt/local/bin/ruby versions.rb
ruby versions.rb
versions.rb:1: uninitialized constant Zlib (NameError)

% fakeroot/usr/bin/ruby versions.rb
zlib    : 1.2.3
OpenSSL : OpenSSL 0.9.8j 07 Jan 2009
SQLite  : 3.6.10

The really important products of this build process are all the .a files in the ruby build directory that will be used to build the final httpassword-ruby executable.

% ls -1 build/ruby/ruby-1.8.6-p287/*.a
build/ruby/ruby-1.8.6-p287/libcrypto.a
build/ruby/ruby-1.8.6-p287/libruby-static.a
build/ruby/ruby-1.8.6-p287/libssl.a
build/ruby/ruby-1.8.6-p287/libz.a

Additionally the highline and htauth ruby code was integrated into the ruby stdlib directory structure.

Step 5 — Building the final executable

The final step is to create out final htpasswd-ruby executable. The output of this step is:

• a single executable
• multiple SQLite databases holding all the ruby source code

At this point we need to update the top level Rakefile in your build system. We need to set the executable name and tell the crate build system how to launch the application.

In our case, we are going to wrap only the htpasswd-ruby application that is part of htauth. For this we look at that script which comes with the htauth gem. This script boils down to the single line:

HTAuth::Passwd.new.run(ARGV, ENV)

Which, it turns out, is exactly the way that crate likes to have applications launched. At this point I should sidestep and say how a crate based application is launched.

Now we dive into the Rakefile and put the final touches on our build to make everything come out the way we want. Currently crate only supports launching an application from a top level class. That means no Module::Class to use for the main class.

require 'crate'

PROJ_NAME = "htpasswd-ruby"
Crate::Project.new( PROJ_NAME ) do |crate|

  # setting these will set the appropriate C constants
  crate.main_file  = "application"
  crate.main_class = "App"
  crate.run_method = "run"

  # make sure 'application.rb' is packed into the databases
  crate.packing_lists << Crate::PackingList.new( Dir.glob("*.rb") )

end

We then have the application.rb file at the top of our build system directory. It is very simple and gives you an idea of how to put a thin ruby wrapper around any existing ruby application.

class App
  def run( argv, env )
    require 'htauth'
    HTAuth::Passwd.new.run( argv, env )
  end
end

Run the default rake task and take a look at the output.

% rake default
(in /Users/jeremy/tmp/htpasswd-ruby)
mkdir -p /Users/jeremy/tmp/htpasswd-ruby/dist
16:59:27  INFO: Packing amalgalite into /Users/jeremy/tmp/htpasswd-ruby/dist/lib.db
16:59:28  INFO: Packing ruby standard lib into /Users/jeremy/tmp/htpasswd-ruby/dist/lib.db
16:59:29  INFO: Packing ruby extension libs into /Users/jeremy/tmp/htpasswd-ruby/dist/lib.db
16:59:31  INFO: Packing project packing lists lists into /Users/jeremy/tmp/htpasswd-ruby/dist/app.db
16:59:32  INFO: Build htpasswd-ruby

% file dist/*
dist/app.db:        SQLite database (Version 3)
dist/htpasswd-ruby: Mach-O executable i386
dist/lib.db:        SQLite database (Version 3)

Our final result, a self-contained set of 3 files containing a statically compiled ruby interpreter and and app to run in it. In this case, we can deploy these 3 files to any i386 Mac OS X machine and run it, it will not use the system ruby. All the ruby libs are in the SQLite3 databases. And just to show that it works.

% cd dist/
% ./htpasswd-ruby -c test.db jjh
Adding password for jjh.
        New password: *****************
Re-type new password: *****************
% cat test.db 
jjh:GtPOQTplES6BE

Closing Thoughts

Well, there you have it; self-contained, statically-built, shippable ruby applications. There are definite improvements that can be made, and I would love to hear what people would like to do with Crate.

Some specific features that I will be adding in no particular order and as my copious free time allows:

• cratify an existing gem for one-stop packaging
• cross compilation targets so you can build, from one library, any executable for any target you have a cross compiler
• installer wrappers, so you can ship
  • .dmg for Mac OS X
  • setup.exe for Windows
  • rpm for CentOS/RHEL/Fedora/etc
  • shar for general UNIX
• smart dependency tracking so only those portions of the ruby stdlib that your application needs is packaged

Comments (View)

Building Binary Ruby Gems for Windows (2008-10-12)

I have two C extension based gems — Hitimes and Amalgalite — that I think would be useful to Windows users. This weekend I sat down, did the research and figured out a way to integrate building binary gems for Windows into my current setup.

Probably the biggest issue for developing binary gems for Windows is that the official Windows on Ruby builds are built built using VC6. This compiler is basically impossible to track down, and Luis has a great explanation of the issues developing Ruby for Windows.

I could have used the new One Click Dev-Kit, and it does look tempting. I already have rake scripts setup for all my building, testing, and distributing and I want to be able to say rake dist:rubyforge and all versions of the gem in question are pushed up to rubyforge in one fell swoop.

Fortunately Mauricio Fernandez had a great blog post about 2 years ago on Cross-compiling Ruby extensions for win32: rcovrt. His approach integrates better with my current setup and I was able to learn about cross-compilers in the process, which are something I've been meaning to experiment with for a while. Although the rcovrt blog post is 2 years old, it is still relevant and was able to point me in the right direction.

Install a Cross-Compiler

The first item we need is a cross compiler.

Right now I'm developing on a Mac so I need the mingw32 cross compiler so I can build binaries that will be compatible with the official VC6 built Ruby. Luckily the i386-mingw cross compiler is available in MacPorts.

sudo port install i386-mingw32-gcc

Unfortunately there is a problem with the i386-mingw32-binutils port. If you see an error relating to makeinfo in your build, use the patch I submitted to the ticket.

Once binutils is all happy and you've had a cup of coffee or two while the mingw32 tool-chain builds installs we can build our gem for Windows.

Building Ruby Using the MinGW Cross-Compiler

This piece was straight from Cross-compiling Ruby extensions for win32: rcovrt. I used Mauricio's cross-compile.sh as a base and altered it to work on OS X.

% cat cross-compile.sh
#!/bin/sh
env ac_cv_func_getpgrp_void=no \
    ac_cv_func_setpgrp_void=yes \
    rb_cv_negative_time_t=no \
    ac_cv_func_memcmp_working=yes \
    rb_cv_binary_elf=no \
    ./configure \
    --host=i386-mingw32 \
    --target=i386-mingw32 \
    --build=i686-darwin9.2.2 \
    --prefix=${HOME}/ruby-mingw32
make ruby
make rubyw.exe
make install

Stick this in a freshly extracted ruby source distribution and run it. This builds an i386-mingw32 version of ruby and installs it in ${HOME}/ruby-mingw32.

The whole purpose of this little exercise is a side effect. I don't really need to run this build of ruby, I only need it to exist so I may build against it.

Building a Windows Gem

Now that I have a mingw32 build of ruby I can build my extensions against it. Normally when building an extensions I do rake ext:build. Which under the covers does:

cd ext/
ruby extconf.rb
make

For a Windows build, we need to have the extconf.rb produced Makefile build against the i386-mingw32 ruby. This is done by taking the rbconfig.rb file from the ruby-mingw32 installation and putting it in the ext/ directory. I keep a copy of this as ext/rbconfig-mingw.rb. The build process then becomes:

cd ext
cp rbconfig-mingw.rb rbconfig.rb
ruby -I. extconf.rb
make
rm -f rbconfig.rb

Using -I. forces the current working directory to the front of the $LOAD_PATH and mkmf will therefore load the rbconfig.rb in the ext/ directory instead of the one from the global ruby installation. This forces all the Config::CONFIG accesses in mkmf to use the environment of the mingw build and make a i386-mingw Makefile.

The make command will then build against the i386-mingw32 ruby installation and the final loadable library can be shipped in a platform dependent gem.

Packaging the Gem

I have a top level gemspec.rb file in all of my projects which holds the global Gem Specification for the project. For the Windows binary gem I need an additional spec that is almost the same as the default spec, without the extensions. Here is the snippet I added for Amalgalite.

# create a new spec based upon the normal spec
Amalgalite::GEM_SPEC_WIN = Amalgalite::GEM_SPEC.clone

# set the platform to be compatible with the official 
# Windows release of Ruby
win_platform = ::Gem::Platform.new( "i386-mswin32_60" )
Amalgalite::GEM_SPEC_WIN.platform = win_platform

# turn off the extensions, since this is a binary release
Amalgalite::GEM_SPEC_WIN.extensions = []

# add the binary extension to the normal file list
Amalgalite::GEM_SPEC_WIN.files +=  ["lib/amalgalite3.so"]

The gem is then packaged using an additional rake gem packaging task.

desc "package the windows gem"
task :package_win => "ext:build_win" do
  cp "ext/amalgalite3.so", "lib", :verbose => true
  Gem::Builder.new( Amalgalite::GEM_SPEC_WIN ).build
  mv Dir["*.gem"].first, "pkg"
end 

This all culminates in the ability to publish a new version of a gem for my supported platforms to rubyforge with a simple rake dist:rubyforge

Clone the Amalgalite repo or Hitimes repo and look around. Let me know if you have any questions or comments.

Comments (View)

Managing a Gem Repository with Stickler (2008-10-09)

One of the items Fernand and I mentioned in our Lone Star talk was Stickler. This was a project I had originally started on back in June and it lay dormant for a little while.

The first public version (0.1.0) is now released.

Why should I use Stickler?

One of the problems that we have had at work is inadvertently upgrading a gem on a production system. For example, you are adding a gem to a system that depends on activerecord, >= 2.0.2. This is great, until active_record 2.1.1 comes out and on your production system you type gem install somegem and it updates your active record installation, as a side effect. Yes, you can just quickly uninstall it, but it is annoying.

Another issue that some installations have, is production boxes are unable to talk directly to the Internet. This is a problem when doing a gem installation using rubygems. You either have to run your own gem server, or install the gems locally.

Installation and Setup

Stickler currently has a basic workflow. First, install and setup your Stickler repository.

jeremy@aramis:~ % sudo gem install stickler
  ============================================================

  Thank you for installing Stickler!

  * Create a new stickler repository:
      stickler setup /path/to/repo

  * Look at the help:
      stickler help

  ============================================================
Successfully installed stickler-0.1.0
1 gem installed
Installing ri documentation for stickler-0.1.0...
Installing RDoc documentation for stickler-0.1.0...

jeremy@aramis:~ % stickler setup /tmp/stickler
created repository root /tmp/stickler
created directory /tmp/stickler/gems
created directory /tmp/stickler/log
created directory /tmp/stickler/specifications
created directory /tmp/stickler/dist
created directory /tmp/stickler/cache
copied in default configuration to /tmp/stickler/stickler.yml
Setting up sources
 * loading http://gems.rubyforge.org/

A Stickler repository is a specific directory that holds all the gem files that are managed by Stickler for your personal distribution. In this case, Stickler is now installed as a gem and we have setup a repository in /tmp/stickler.

Populating the Repository

A repository is great, and completely useless unless it is populated with gems to distribute. So that is what we will do. The best way to operate on a Stickler repository is to be in it. You can also use the --directory commandline option on every command, but that can get old. Therefore, we'll operate from the root of the repository.

cd /tmp/stickler

First lets look at the state of the repository using the info command.

jeremy@aramis:/tmp/stickler % stickler info
Setting up sources
 * loading http://gems.rubyforge.org/

Upstream Sources
----------------

  http://gems.rubyforge.org/ : 16651 gems available
                             : 0 gems existing

Configured gems (in stickler.yml)
---------------------------------


Existing gems
-------------

Yes, that definitely is empty. We do a fair number of merb apps, so lets add merb to our repository. It is a 2 step process to add a gem to the repository via the commandline. We first pick the requirement operator, and then the version. In our case, we want to stick with a specific version of merb so we'll add merb, = 0.9.8 to our repository

jeremy@aramis:/tmp/stickler % stickler add gem merb
Setting up sources
 * loading http://gems.rubyforge.org/

You need to pick the merb Requirement to configure Stickler.
This involves picking one of the following Requirement operators
See http://docs.rubygems.org/read/chapter/16#page74 for operator info.

You need to (1) pick an operator and (2) pick a requirement.
The most common operators are >=, > and ~>
1. =  Equals version
2. != Not equal to version
3. >  Greater than version
4. <  Less than version
5. >= Greater than or equal to
6. <= Less than or equal to
7. ~> Approximately greater than
(1) Pick an operator ? 1

Now to pick a requirement.  Based upon your chosen operator '=',
These are the available version of the merb gem.
1. = 0.9.8   7. = 0.9.2   13. = 0.4.1  19. = 0.3.0  25. = 0.0.6
2. = 0.9.7   8. = 0.5.3   14. = 0.4.0  20. = 0.2.0  26. = 0.0.5
3. = 0.9.6   9. = 0.5.2   15. = 0.3.7  21. = 0.1.0  27. = 0.0.4
4. = 0.9.5   10. = 0.5.1  16. = 0.3.4  22. = 0.0.9  28. = 0.0.3
5. = 0.9.4   11. = 0.5.0  17. = 0.3.3  23. = 0.0.8  29. = 0.0.2
6. = 0.9.3   12. = 0.4.2  18. = 0.3.1  24. = 0.0.7  30. = 0.0.1
(2) Pick a requirement ? 1

Resolving gem dependencies for merb (= 0.9.8, runtime) ...
Adding ParseTree-2.1.1-x86-mswin32
Adding ruby2ruby-1.1.9
Adding RubyInline-3.7.0
Adding ParseTree-2.2.0
Adding merb-action-args-0.9.8
Adding merb-assets-0.9.8
Adding highline-1.4.0
Adding diff-lcs-1.1.2
Adding templater-0.3.2
Adding merb-gen-0.9.8
Adding haml-2.0.3
Adding merb-haml-0.9.8
Adding builder-2.1.2
Adding merb-builder-0.9.8
Adding mailfactory-1.4.0
Adding merb-mailer-0.9.8
Adding merb-parts-0.9.8
Adding merb-cache-0.9.8
Adding merb-slices-0.9.8
Adding merb-jquery-0.9.8
Adding extlib-0.9.7
Adding abstract-1.0.0
Adding erubis-2.6.2
Adding json_pure-1.1.3
Adding rubyforge-1.0.0
Adding rake-0.8.3
Adding hoe-1.8.0
Adding rspec-1.1.8
Adding rack-0.4.0
Adding mime-types-1.15
Adding hpricot-0.6.161-java
Adding hpricot-0.6.161
Adding hpricot-0.6-x86-mswin32
Adding thor-0.9.6
Adding merb-core-0.9.8
Adding merb-helpers-0.9.8
Adding merb-more-0.9.8
Adding mongrel-1.1.5-java
Adding mongrel-1.1.1-java
Adding daemons-1.0.10
Adding fastthread-1.0.1
Adding fastthread-1.0.1-x86-mswin32
Adding mongrel-1.1.5
Adding mongrel-1.1.3-x86-mswin32
Adding mongrel-1.1.5-x86-mswin32-60
Adding mongrel-1.1.2-x86-mswin32
Adding gem_plugin-0.2.3
Adding cgi_multipart_eof_fix-2.5.0
Adding mongrel-1.1.5-x86-mingw32
Adding merb-0.9.8

Wow, that's a lot of gems added with for merb. And this is more than would be normally installed if you did a gem install merb. Stickler assumes you are redistributing gems, and as such gets every available platform for a gem when you add it to the repository. And then it recurses through all runtime and development dependencies. And now when we look at the repository state we see:

Setting up sources
 * loading http://gems.rubyforge.org/

Upstream Sources
----------------

  http://gems.rubyforge.org/ : 16652 gems available
                             : 50 gems existing

Configured gems (in stickler.yml)
---------------------------------

merb : = 0.9.8

Existing gems
-------------

ParseTree-2.1.1-x86-mswin32
ParseTree-2.2.0
RubyInline-3.7.0
abstract-1.0.0
builder-2.1.2
cgi_multipart_eof_fix-2.5.0
daemons-1.0.10
diff-lcs-1.1.2
erubis-2.6.2
extlib-0.9.7
fastthread-1.0.1
fastthread-1.0.1-x86-mswin32
gem_plugin-0.2.3
haml-2.0.3
highline-1.4.0
hoe-1.8.0
hpricot-0.6-x86-mswin32
hpricot-0.6.161
hpricot-0.6.161-java
json_pure-1.1.3
mailfactory-1.4.0
merb-0.9.8
merb-action-args-0.9.8
merb-assets-0.9.8
merb-builder-0.9.8
merb-cache-0.9.8
merb-core-0.9.8
merb-gen-0.9.8
merb-haml-0.9.8
merb-helpers-0.9.8
merb-jquery-0.9.8
merb-mailer-0.9.8
merb-more-0.9.8
merb-parts-0.9.8
merb-slices-0.9.8
mime-types-1.15
mongrel-1.1.1-java
mongrel-1.1.2-x86-mswin32
mongrel-1.1.3-x86-mswin32
mongrel-1.1.5
mongrel-1.1.5-java
mongrel-1.1.5-x86-mingw32
mongrel-1.1.5-x86-mswin32-60
rack-0.4.0
rake-0.8.3
rspec-1.1.8
ruby2ruby-1.1.9
rubyforge-1.0.0
templater-0.3.2
thor-0.9.6

There is a difference between configured gems and existing gems. Configured gems are those you have specifically requested to be in the repository. Existing gems are those that are added to support the configured gems and the configured gems themselves.

Adding Additional Sources

Part of the reason for Stickler is to merge upstream gem repositories into your own internal repository. Say you use the github repository and want to add a gem or two from it to your internal system.

jeremy@aramis:/tmp/stickler % stickler add source http://gems.github.com
Setting up sources
 * loading http://gems.rubyforge.org/
 * loading http://gems.github.com/
http://gems.github.com added to sources

That was easy enough. Now lets add a gem from github and watch the dependencies roll in.

jeremy@aramis:/tmp/stickler % stickler add gem aniero-tire_swing
Setting up sources
 * loading http://gems.rubyforge.org/
 * loading http://gems.github.com/

You need to pick the aniero-tire_swing Requirement to configure Stickler.
This involves picking one of the following Requirement operators
See http://docs.rubygems.org/read/chapter/16#page74 for operator info.

You need to (1) pick an operator and (2) pick a requirement.
The most common operators are >=, > and ~>
1. =  Equals version
2. != Not equal to version
3. >  Greater than version
4. <  Less than version
5. >= Greater than or equal to
6. <= Less than or equal to
7. ~> Approximately greater than
(1) Pick an operator ? =

Now to pick a requirement.  Based upon your chosen operator '=',
These are the available version of the aniero-tire_swing gem.
1. = 0.0.3  2. = 0.0.2
(2) Pick a requirement ? 1

Resolving gem dependencies for aniero-tire_swing (= 0.0.3, runtime) ...
Adding polyglot-0.2.3
Adding treetop-1.2.4
Adding attributes-5.0.1
Adding activesupport-2.1.1
Adding aniero-tire_swing-0.0.3

You can see that aneiro-tire_swing was added from github, and it also resolved the dependencies to gems that reside on rubyforge. Looking at info shows us:

jeremy@aramis:/tmp/stickler % stickler info
Setting up sources
 * loading http://gems.rubyforge.org/
 * loading http://gems.github.com/

Upstream Sources
----------------

  http://gems.rubyforge.org/ : 16652 gems available
                             : 54 gems existing
     http://gems.github.com/ : 2463 gems available
                             : 1 gems existing

Configured gems (in stickler.yml)
---------------------------------

aniero-tire_swing : = 0.0.3
merb : = 0.9.8
...

Batch Addition

It can get repetitive to manually add gem after gem from the commandline. When you know exactly what you need, crack open the stickler.yml file and add a few gems. For instance, lets say that we are in the process of updating all our rails apps to 2.1. In the meantime we want to make sure we do not inadvertently update our production systems to 2.1. They are still at 2.0.2. So we add all the rails gems at the 2.0.2 level to the stickler.yml file in the repository.

---
downstream_source: http://gems.example.com/
sources:
- http://gems.rubyforge.org/
- http://gems.github.com/
gems:
  aniero-tire_swing:
  - = 0.0.3
  merb:
  - = 0.9.8
  activerecord: = 2.0.2
  activesupport: = 2.0.2
  actionmailer: = 2.0.2
  actionpack: = 2.0.2
  activeresource: = 2.0.2
  rails: = 2.0.2

We can now use the sync command to add all of them in at once.

jeremy@aramis:/tmp/stickler % stickler sync
Setting up sources
 * loading http://gems.rubyforge.org/
 * loading http://gems.github.com/

Making sure that all gems listed in configuration are available

Resolving gem dependencies for rails (= 2.0.2, runtime) ...
Adding activerecord-2.0.2
Adding actionpack-2.0.2
Adding actionmailer-2.0.2
Adding activesupport-2.0.2
Adding activeresource-2.0.2
Adding rails-2.0.2
Resolving gem dependencies for activerecord (= 2.0.2, runtime) ...
Resolving gem dependencies for activeresource (= 2.0.2, runtime) ...
Resolving gem dependencies for aniero-tire_swing (= 0.0.3, runtime) ...
Resolving gem dependencies for actionpack (= 2.0.2, runtime) ...
Resolving gem dependencies for actionmailer (= 2.0.2, runtime) ...
Resolving gem dependencies for merb (= 0.9.8, runtime) ...
Resolving gem dependencies for activesupport (= 2.0.2, runtime) ...

You can see that it added in all the gems in and then it also made sure that aneiro-tire_swing and merb were also synced up. If you need to rebuild the entire repository from scratch, use stickler sync --rebuild and it will wipe out the current gems and specifications in the repo and download them again.

Distribution of Gems

Populating the repository is only good if you can get your internal systems to utilize the repo. This is where the generate commands come into play. The generate index command does exactly what a gem generate_index does, but it uses your repository as the base.

Setting up sources
 * loading http://gems.rubyforge.org/
 * loading http://gems.github.com/
Generating rubygems index in /private/tmp/stickler/dist
Loading 61 gems from /private/tmp/stickler/dist
...............WARNING:  Skipping misnamed gem: /private/tmp/stickler/dist/gems/fastthread-1.0.1-x86-mswin32.gem => fastthread-1.0.1-x86-mswin32 (fastthread-1.0.1-mswin32)
.....WARNING:  Skipping misnamed gem: /private/tmp/stickler/dist/gems/hpricot-0.6-x86-mswin32.gem => hpricot-0.6-x86-mswin32 (hpricot-0.6-mswin32)
WARNING:  Skipping misnamed gem: /private/tmp/stickler/dist/gems/hpricot-0.6.161-java.gem => hpricot-0.6.161-java (hpricot-0.6.161-jruby)
..................WARNING:  Skipping misnamed gem: /private/tmp/stickler/dist/gems/mongrel-1.1.1-java.gem => mongrel-1.1.1-java (mongrel-1.1.1-jruby)
WARNING:  Skipping misnamed gem: /private/tmp/stickler/dist/gems/mongrel-1.1.2-x86-mswin32.gem => mongrel-1.1.2-x86-mswin32 (mongrel-1.1.2-mswin32)
WARNING:  Skipping misnamed gem: /private/tmp/stickler/dist/gems/mongrel-1.1.3-x86-mswin32.gem => mongrel-1.1.3-x86-mswin32 (mongrel-1.1.3-i386-mswin32)
....WARNING:  Skipping misnamed gem: /private/tmp/stickler/dist/gems/ParseTree-2.1.1-x86-mswin32.gem => ParseTree-2.1.1-x86-mswin32 (ParseTree-2.1.1-i386-mswin32)
............
Loaded all gems
Generating quick index gemspecs for 54 gems
......................................................
Complete
Generating specs index
Generating latest specs index
Generating quick index
Generating latest index
Generating Marshal master index
Generating YAML master index for 54 gems (this may take a while)
......................................................
Complete
Compressing indicies

The generated, distributable gem repository based upon your stickler repository is generated in the dist sub directory in your repository. This directory can be rsynced to a web server in your infrastructure, or run a web server on this machine and point its document root, or a directory alias to the dist directory.

An extra item you may do is globally configure your rubygems installations to automatically use your internal gem server as the canonical source instead of http://gems.rubyforge.org/. Stickler provides a quick command to generate this top level configuration, and the file contains the information on where to put it.

jeremy@aramis:/tmp/stickler % stickler generate sysconfig > gemrc
Setting up sources
 * loading http://gems.rubyforge.org/
 * loading http://gems.github.com/
Generating configuration to stdout

jeremy@aramis:/tmp/stickler % cat gemrc
#
# This is the system wide configuration to be used by
# rubygem clients that install gems from the repository
# located at :
#
#   http://gems.example.com/
#
# On Unix like machines install in
#
#   /etc/gemrc
#
# On Windows machines install in
#
#   C:\Documents and Settings\All Users\Application Data\gemrc
#
---
:sources:
- http://gems.example.com/

Conclusions

Hopefully you can find Stickler useful in your infrastructure. Please let me know about features and bugs.

Comments (View)

Hitimes are upon us (2008-09-14)

Saturday afternoon at the Lone Star Ruby Conf I was discussing something, I forget exactly what, with Bruce and he mentioned that he was looking for a really fast way to measure elapsed time of a piece of code.

The naive approach would be call Time.now before and after the call, or maybe use the Time.elapse method from Facets. Both of those approaches are perfectly acceptable, and return results down to the µsecond. Can we do better? Yes.

We have a goal of being faster than 2 successive Time.now calls. Each operating system has a different way accessing a high-resolution timer.

• Unixes - the POSIX calls to clock_gettime()
• OS X - UpTime()
• Windows - QueryPerformanceCounter()

Unify the C level interface to all of those, wrap it up in a Ruby extension and the result is the just-released Hitimes, a high-resolution timer library in Ruby for when you want to do measuring at the nanosecond level.

gem install hitimes

The simplest and lightest weight way to use the library is via the Interval class. This does one really simple thing. It measures an interval of time. You can measure code in a block:

require 'hitimes'
require 'open-uri'
duration = Hitimes::Interval.measure do
            bruce = open("http://www.codefluency.com/").read
           end
puts duration # => 0.261841618

Or measure a particular piece of code for performance:

interval = Hitimes::Interval.new
# do something
interval.start
# some code to measure
duration = interval.stop

There are other approaches using Interval, and a Timer class for measuring series of intervals and reporting statistics.

Comments (View)

Post Lone Star Talk Thoughts (2008-09-08)

Fernand and I presented a talk at Lone Star Ruby Conf last Friday entitled Building and Managing a Ruby Infrastructure. It appears we hit upon a subject that people are starting encounter. Over the next couple of weeks I'll post a few more articles explaining in more detail a few of the key concepts. For now, a few links to things that we mentioned in the talk.

• Bones -- A great project templating tool, extendible to use your own proect templates.
• Rational Versioning Policy -- in which ~> a.k.a "Twiddle Wakka" is explained.
• Stickler -- organize and maintain an internal gem distribution server. Still in planning stages.
• take a look at the gem index and gem server commands

Comments (View)

Amalgalite Pthread Error (2008-07-05)

If you happen to get an error like the following when running amalgalite, it was a bug in how amalgalite was built. I wasn't matching the thread support in amalgalite with the threadsupport of the installed ruby.

amalgalite-0.2.0/ext/amalgalite3.so: undefined symbol: pthread_mutexattr_init

For future reference this is how you can see how ruby was compiled on my iMac

>> puts "#{RUBY_VERSION} #{RUBY_PATCHLEVEL} (#{RUBY_RELEASE_DATE}) #{RUBY_PLATFORM}"
1.8.6 114 (2008-03-03) i686-darwin9.2.2
>> require 'rbconfig'
>> puts Config::CONFIG['configure_args']
 '--prefix=/opt/local' '--enable-shared' '--mandir=/opt/local/share/man'
 '--enable-pthread' '--without-tk' 'CC=/usr/bin/gcc-4.0' 'CFLAGS=-O2'
 'LDFLAGS=-L/opt/local/lib' 'CPPFLAGS=-I/opt/local/include'
 'CPP=/usr/bin/cpp-4.0'

versus my OpenBSD server:

>> puts "#{RUBY_VERSION} #{RUBY_PATCHLEVEL} (#{RUBY_RELEASE_DATE}) #{RUBY_PLATFORM}"
1.8.6 110 (2007-09-23) x86_64-openbsd4.2
>> require 'rbconfig'
>>puts Config::CONFIG['configure_args']
 '--enable-shared' '--enable-ipv6' '--with-dbm-type=bogus'
 '--with-opt-dir=/usr/local' '--with-tcl-include=/usr/local/include/tcl8.4'
 '--with-tk-include=/usr/local/include/tk8.4' '--with-X11-dir=/usr/X11R6'
 '--prefix=/usr/local' '--sysconfdir=/etc' '--mandir=/usr/local/man'
 '--infodir=/usr/local/info' 'CC=cc' 'CFLAGS=-O2 -pipe'
 'CPPFLAGS=-DOPENSSL_NO_STATIC_ENGINE' 'YACC=yacc'

You can see that on my iMac ruby was compiled with @--enable-pthread@ but on my OpenBSD machine that configuration option is missing. I didn't take this into account when building Amalgalite.

Version 0.2.1 was just released which fixes this bug.

gem install amalgalite

Comments (View)

Amalgalite 0.2.0 Released (2008-07-04)

Now with more examples!

SQLite has the ability to do incremental IO on items that are BLOB's. I just finished adding support for this in Amalgalite. This means you can SELECT a blob, and from the result set, transfer the data directly to an IO stream.

A quick example. Assume we store files in filestore.db that has the following schema:

CREATE TABLE files(
  id   INTEGER PRIMARY KEY AUTOINCREMENT,
  path VARCHAR UNIQUE
  data BLOB
)

You can then write all the files in this table to a different tree with this piece of code. In this case, the files are not loaded into memory, the are copied directly from the pages in the database for the BLOBs to the output files.

new_root_dir = "/tmp"
db = Amalgalite::Database.new( "filestore.db" )
db.execute( "SELECT * FROM files" ) do |row|
  # unsafe path expansion
  dest_path = File.expand_path( File.join( new_root_dir, row['path']) )
  row['data'].write_to_file( dest_path )
end
db.close

To get this behavior you need to make sure that column that is SQLite's rowid column is part of the result set. If one of your columns is INTEGER PRIMARY KEY it is that one. Otherwise you must specifically select ROWID, OID. The code above will still work if the rowid column is not specified, but Amalgalite will fully read it into memory before being written to the file.

Take a look at the Amalgalite::Blob api and the example.

Comments (View)

Amalgalite 0.1.0 Released (2008-06-21)

I've been a fan of SQLite for a long time. Many years ago I considered writing an SQLite extension for Ruby. Jamis beat me to it, and I've been happily using ruby-sqlite since it arrived on the scene.

Since then, SQLite itself has greatly improved. There are new features, an evolution in the API, and most of all, The SQLite Amalgamation. This is the entire SQLite database engine as a single C source code file.

Perfect! Why not make a ruby extension that embeds SQLite directly in the extension?

Amalgalite is born.

Install as a gem

gem install amalgalite

Or grab from github

git clone git://github.com/copiousfreetime/amalgalite.git

Give it a whirl and let me know what you think.

require 'rubygems'
require 'amalgalite'
puts "using SQLite v#{Amalgalite::SQLite3::Version}" # => "using SQLite v3.5.9"

Current Functionality

I would consider Amalgalite beta right now. It is fully usable; but it does not contain all the features I want yet.

The normal stuff:

• open or create databases
• run SQL in single statements or in batches.
• use prepared statements with parameters
• access resultsets by index or by column name
• transactions
• type conversion between database types and Ruby types

The fun/happy/cool stuff, and not available in sqlite3-ruby:

• trace information → Amalgalite::TraceTap
• profile information → Amalgalite::ProfileTap
• read only databases
• a type conversion protocol → Amalgalite::TypeMaps

Future plans

• ActiveRecord driver
• DataMapper driver
• Sequel driver
• progress and authorizer callbacks
• amalgalite command line tool similar to the sqlite3 commandline tool
• utf-16 support
• blob support
• and more

Comments (View)