I've always used git to determine which files should go into the gem package:
gem.files = `git ls-files`.split "\n"
Unfortunately, this approach has recently proved to be inappropriate. I need a self-contained, pure-Ruby solution.
My first idea was to simply glob the entire directory, but that alone is likely to include unwanted files. So, after researching the problem, I came up with this:
# example.gemspec
directory = File.dirname File.expand_path __FILE__
dotfiles = %w(.gitignore .rvmrc)
ignore_file = '.gitignore'
file_list = []
Dir.chdir directory do
ignored = File.readlines(ignore_file).map(&:chomp).reject { |glob| glob =~ /\A(#|\s*\z)/ }
file_list.replace Dir['**/**'] + dotfiles
file_list.delete_if do |file|
File.directory?(file) or ignored.any? { |glob| File.fnmatch? glob, file }
end
end
# Later...
gem.files = file_list
That seems a bit complex for a gemspec. It also does not fully support gitignore's pattern format. It currently seems to work but I'd rather not run into problems later.
Is there a simpler but robust way to compute the gem's list of files? Most gems apparently use git ls-files, and the ones that don't either use a solution similar to mine or specify the files manually.
The easiest solution depending on rake to list all files from a directory, but exclude everything in the .gitignore file:
require 'rake/file_list'
Rake::FileList['**/*'].exclude(*File.read('.gitignore').split)
Official rubygems solution, list and exclude manually:
require 'rake'
spec.files = FileList['lib/*.rb',
'bin/*',
'[A-Z]*',
'test/*'].to_a
# or without Rake...
spec.files = Dir['lib/*.rb'] + Dir['bin/*']
spec.files += Dir['[A-Z]*'] + Dir['test/**/*']
spec.files.reject! { |fn| fn.include? "CVS" }
Bundler solution, list manually:
s.files = Dir.glob("{lib,exe}/**/*", File::FNM_DOTMATCH).reject {|f| File.directory?(f) }
Note: rejecting directories is useless as gem will ignore them by default.
Vagrant solution to mimic git ls-files and taking care of .gitignore in pure ruby:
# The following block of code determines the files that should be included
# in the gem. It does this by reading all the files in the directory where
# this gemspec is, and parsing out the ignored files from the gitignore.
# Note that the entire gitignore(5) syntax is not supported, specifically
# the "!" syntax, but it should mostly work correctly.
root_path = File.dirname(__FILE__)
all_files = Dir.chdir(root_path) { Dir.glob("**/{*,.*}") }
all_files.reject! { |file| [".", ".."].include?(File.basename(file)) }
all_files.reject! { |file| file.start_with?("website/") }
all_files.reject! { |file| file.start_with?("test/") }
gitignore_path = File.join(root_path, ".gitignore")
gitignore = File.readlines(gitignore_path)
gitignore.map! { |line| line.chomp.strip }
gitignore.reject! { |line| line.empty? || line =~ /^(#|!)/ }
unignored_files = all_files.reject do |file|
# Ignore any directories, the gemspec only cares about files
next true if File.directory?(file)
# Ignore any paths that match anything in the gitignore. We do
# two tests here:
#
# - First, test to see if the entire path matches the gitignore.
# - Second, match if the basename does, this makes it so that things
# like '.DS_Store' will match sub-directories too (same behavior
# as git).
#
gitignore.any? do |ignore|
File.fnmatch(ignore, file, File::FNM_PATHNAME) ||
File.fnmatch(ignore, File.basename(file), File::FNM_PATHNAME)
end
end
Using pathspec gem Match Path Specifications, such as .gitignore, in Ruby!
See https://github.com/highb/pathspec-ruby