Say, I have Gemfile like following.
source "GEM_REPOSITORY"
gem 'gem_A'
# gem_A has no additional dependency
gem 'gem_B'
# gem_B depends on gem_B_1 and gem_B_2
When I run bundle install, I want Bundler to do the following.
If a gem already exists in "local system-wide gems", it copies the gem from local.
If a gem doesn't exist in local, it looks for GEM_REPOSITORY.
I looked for some related articles, and found some of likely-answers like
Ruby Bundler multiple sources in Gemfile
SOURCE PRIORITY
But none of the above looks like the answer for me.
Using source repository priority does't work. Because in the example above, if dependent gem (say, gem_B_1) exits in local but the target gem (gem_B) doesn't exist in local, it'll download both of above from the remote repository.
Are there any work around for doing this?
If not, don't you guys think it's necessary considering the cost of the implementation and the effect?
This is the current behavior.
When running gem install, directly or via bundle install, gem will first build a dependency graph with all the needed gems. If the gem is found locally it will use it, otherwise it will try to download it from the specified source.
If you want, try it yourself.
bundle gem gem_a
bundle gem gem_b
cd gem_a
vim gem_a.gemspec
add
spec.add_dependency 'multi_json', '~> 1.10.1'
or any dependency you want to the gem and run bundle install.
cd ../gem_b
vim Gemfile
and add
gem 'gem_a', path: '../gem_a'
then run
bundle install --verbose
you will see that the multi_json or whatever dependency of gem_a uses the local version and does not download anything.
This is of course also true for gems from remote sources.
Related
How do I configure bundler so that when I run bundle install it looks for gems under /my/custom/path first and if it doesn't find there then try to fetch them from a ruby gem remote repository hosted under https://a.nice.host and downloads those into ./local/relative/path (relative to cwd for example). I would like to avoid the bundler looking at default gem installation system path or rubygems.org
The syntax for sourcing a gem from a local folder is:
gem 'some-gem-name', path: '/my/custom/path'
And the syntax for specifying a custom source is:
gem 'another-gem-name', source: 'https://a.nice.host'
And to install gems into a specific local folder, you can run:
bundle install --path ./local/relative/path
Now, that's probably all the tools you need, in truth... (And in fact, especially for that last requirement, you may instead wish to look into rvm gemsets, or using bundle install --deployment.)
But you did also ask about "looking in a local folder first, and only falling back to a remote source if it doesn't exist". That's quite an odd requirement (usually you'd only want to explicitly opt-in to fetching gems from a local path?!), but to answer this question as you've asked it...
A Gemfile is literally just ruby code! So you can define this logic using... You guessed it, ruby! For example:
if File.exists?('/my/custom/path')
gem 'some-gem-name', path: '/my/custom/path'
else
gem 'some-gem-name', source: 'https://a.nice.host'
end
If this (unusual) pattern needs to be repeated in multiple places, you could wrap it into some helper method.
For more information on the configuration options of bundler, please see the documentation.
I'm developing a couple of private gems and I think I don't understand correctly the PATH/GEM_PATH and/or Bundler/RVM installation flow, would love if someone could chip in.
I have a repository with two gems (A & B for simplicity sake). I've developed the gems using the scaffolding + following the guidelines provided by this bundler tutorial.
Thanks to the Bundler project I have a few Rakefile tasks like rake build, rake install, rake install:local and rake release. Because of the private nature of these gems I can't release them to RubyGems (and we haven't looked into hosting our rubygems).
My machines are using RVM to manage ruby versions and Bundler version 1.15.1
What I want to do: Assuming a new machine/developer trying out the project, ideally we would cd into each of the subfolders (currently 2, gem A and gem B), run rake install and after that we should have the gems available system wide for the current user.
What is happening: The gems are built and work properly, but they are only available inside the subfolder of each gem i.e. gem A is only available inside the subfolder A and gem B is only available inside subfolder B.
What I've tried: So, after rake build/install/install:local a new .gem file is generated under pkg. I've tried to manually install the "compiled" file using gem install pkg/A.gem, gem install --local pkg/A.gem and gem install --local --user-install pkg/A.gem without success. (there are plenty of SO questions/answers about this)
I believe this has something to do with the PATH variables, but like I said before I don't fully understand the way they are managed. I get the following results from these commands:
# Our gem
> gem which A
/home/ubuntu/.rvm/gems/ruby-2.4.0/gems/A-0.1.8/lib/A.rb
# Pry, available globally
> gem which pry
/home/ubuntu/.rvm/gems/ruby-2.4.0/gems/pry-0.11.1/lib/pry.rb
I've been lost and frustrated for far too long now, any help is appreciated. Also open to hear suggestions of better private gem installation flows :)
Yes, it has something to do with your PATH variables. Your installation seems to be good.
I advise you to first affirm your gems installation path with:
echo $GEM_HOME
The double check your PATH to ensure its present and also confirm that the GEM home is also where the gem got installed into from the rake install
echo $PATH
If not, put it in your path and you should be fine with something like this:
echo PATH=$PATH:$GEM_HOME >> ~/.bashrc
source ~/.bashrc
Build your gem as per that guide you linked. You should end up with a gem file. Distribute this as you see fit (I use rsync/crontab to download newer gem versions but anything goes). User can install the gem as follows:
gem install --user-install /path/to/your/file.gem
This will install the gem in the user's ~/.gem/ruby/<version>/gems/<your-gem-name> directory.
Tried it with an empty gem (foodie, as in that example guide) and it works fine. But if you don't specify the --user-install parameter it will try to install in the system ruby dir (/usr/lib/ruby/gems...)
I want to put a debugger in a file for testing, but can't find where Bundler installs gems pulled from Github on my local machine.
I've looked at this thread http://bundler.io/v1.5/git.html which shows how to setup a local file repo to pull from, but I would rather avoid this as my situation is a one off debugging scenario.
I use rbenv for my ruby and gem management. When I pull in a gem from a git repo, it places the files for the gem here:
/usr/local/var/rbenv/versions/2.3.1/lib/ruby/gems/2.3.0/bundler/gems/gem-name-SHA/
On my machine bundler installed the gem in:
~/.rvm/gems/ruby-2.3.3/bundler/gems/gem-from-github
gem which [GEM] doesn't work on its own, but bundle exec gem which [GEM] does.
I am running my own gem server with geminabox. This gem server hosts private forks of several public gems, as well as my own gems. I have a Gemfile with the sources listed like so:
source https://rubygems.org
source http://my/gem/server
When Bundler installs a bundle, I would like it to always use the version available on my gem server if it exists and satisfies the version requirements. I can't get this to work. Instead, bundler seems to pull from rubygems unless it can't find a gem with the same name there. This may be because the versions on rubygems are higher. However, bundler's documentation makes no mention of version numbers. This is how it describes source priority:
SOURCE PRIORITY
When attempting to locate a gem to satisfy a gem requirement, bundler uses the following priority order:
The source explicitly attached to the gem (using :path or :git)
For implicit gems (dependencies of explicit gems), any git or path repository otherwise declared. This results in bundler prioritizing the ActiveSupport gem from the Rails git repository over ones from rubygems.org
The sources specified via source, searching each source in your Gemfile from last added to first added.
Also, I know it is possible to explicitly list a source for each gem in the Gemfile. I would prefer to avoid this, since I have many of my own gems, and it is annoying to make explicit entries in the Gemfile for dependencies of dependencies (which you have to do, because bundler will not read the Gemfile of dependencies).
How can I achieve my goal?
I'm using bundler and have a Gemfile that looks like this:
source 'http://rubygems.org'
gem 'sinatra', '1.3.1'
gem 'httparty'
# ...etc...
gem 'my_custom_gem', :git => 'git#github.com:me/my_custom_gem.git'
When I run bundle install it fetches the necessary gems, including my custom gem, and installs them in the system gem directory. So far, so good. However, an issue arises when I try to vendor them into a project-local directory. When I run
bundle install --path vendor
It creates the "vendor" directory in my project root and installs all the regular gems there. So I see directories like
vendor/ruby/1.8/gems/sinatra-1.3.1
vendor/ruby/1.8/gems/httparty-0.8.1
...etc...
But it does not vendor the gem specified with a 'git' parameter. I expect to see but do not see anything like
vendor/ruby/1.8/gems/my_custom_gem-1.0.0
It continues to use the system-installed version of this gem. Any explanation for this? Any clean way to get this custom gem vendored as well?
Not supported right now, hopefully coming in Bundler 1.1:
https://github.com/carlhuda/bundler/issues/67
For now you'll have to do:
cd vendor/ruby/1.8/gems/
git clone git://github.com/foo/foo.git
or similar