mirror of
https://github.com/ruby/ruby.git
synced 2022-11-09 12:17:21 -05:00

https://github.com/ruby/bigdecimal/commit/c2b22cc8b3 https://github.com/ruby/bigdecimal/commit/8cbca8481d https://github.com/ruby/bigdecimal/commit/f05aecf673
44 lines
1.3 KiB
Ruby
44 lines
1.3 KiB
Ruby
# coding: utf-8
|
|
|
|
bigdecimal_version = '3.0.0'
|
|
|
|
Gem::Specification.new do |s|
|
|
s.name = "bigdecimal"
|
|
s.version = bigdecimal_version
|
|
s.authors = ["Kenta Murata", "Zachary Scott", "Shigeo Kobayashi"]
|
|
s.email = ["mrkn@mrkn.jp"]
|
|
|
|
s.summary = "Arbitrary-precision decimal floating-point number library."
|
|
s.description = "This library provides arbitrary-precision decimal floating-point number class."
|
|
s.homepage = "https://github.com/ruby/bigdecimal"
|
|
s.license = "Ruby"
|
|
|
|
s.require_paths = %w[lib]
|
|
s.extensions = %w[ext/bigdecimal/extconf.rb]
|
|
s.files = %w[
|
|
bigdecimal.gemspec
|
|
ext/bigdecimal/bigdecimal.c
|
|
ext/bigdecimal/bigdecimal.h
|
|
ext/bigdecimal/bits.h
|
|
ext/bigdecimal/feature.h
|
|
ext/bigdecimal/missing.h
|
|
ext/bigdecimal/static_assert.h
|
|
lib/bigdecimal.rb
|
|
lib/bigdecimal/jacobian.rb
|
|
lib/bigdecimal/ludcmp.rb
|
|
lib/bigdecimal/math.rb
|
|
lib/bigdecimal/newton.rb
|
|
lib/bigdecimal/util.rb
|
|
sample/linear.rb
|
|
sample/nlsolve.rb
|
|
sample/pi.rb
|
|
]
|
|
|
|
s.required_ruby_version = Gem::Requirement.new(">= 2.4.0")
|
|
|
|
s.add_development_dependency "fiddle"
|
|
s.add_development_dependency "rake", ">= 12.3.3"
|
|
s.add_development_dependency "rake-compiler", ">= 0.9"
|
|
s.add_development_dependency "minitest", "< 5.0.0"
|
|
s.add_development_dependency "pry"
|
|
end
|