Skip to content
Merged
Show file tree
Hide file tree
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Next Next commit
Add RBS signature and testing
  • Loading branch information
ksss committed Mar 30, 2025
commit d193a0cdec630f71543712e7720634c736efeec3
1 change: 1 addition & 0 deletions Gemfile
Original file line number Diff line number Diff line change
Expand Up @@ -2,3 +2,4 @@ source "https://rubygems.org"

gem "rake"
gem "test-unit"
gem "rbs"
355 changes: 355 additions & 0 deletions sig/base64.rbs
Original file line number Diff line number Diff line change
@@ -0,0 +1,355 @@
# <!-- rdoc-file=lib/base64.rb -->
# Module Base64 provides methods for:
#
# * Encoding a binary string (containing non-ASCII characters) as a string of
# printable ASCII characters.
# * Decoding such an encoded string.
#
# Base64 is commonly used in contexts where binary data is not allowed or
# supported:
#
# * Images in HTML or CSS files, or in URLs.
# * Email attachments.
#
# A Base64-encoded string is about one-third larger that its source. See the
# [Wikipedia article](https://en.wikipedia.org/wiki/Base64) for more
# information.
#
# This module provides three pairs of encode/decode methods. Your choices among
# these methods should depend on:
#
# * Which character set is to be used for encoding and decoding.
# * Whether "padding" is to be used.
# * Whether encoded strings are to contain newlines.
#
# Note: Examples on this page assume that the including program has executed:
#
# require 'base64'
#
# ## Encoding Character Sets
#
# A Base64-encoded string consists only of characters from a 64-character set:
#
# * `('A'..'Z')`.
# * `('a'..'z')`.
# * `('0'..'9')`.
# * `=`, the 'padding' character.
# * Either:
# * `%w[+ /]`:
# [RFC-2045-compliant](https://datatracker.ietf.org/doc/html/rfc2045);
# *not* safe for URLs.
# * `%w[- _]`:
# [RFC-4648-compliant](https://datatracker.ietf.org/doc/html/rfc4648);
# safe for URLs.
#
# If you are working with Base64-encoded strings that will come from or be put
# into URLs, you should choose this encoder-decoder pair of RFC-4648-compliant
# methods:
#
# * Base64.urlsafe_encode64 and Base64.urlsafe_decode64.
#
# Otherwise, you may choose any of the pairs in this module, including the pair
# above, or the RFC-2045-compliant pairs:
#
# * Base64.encode64 and Base64.decode64.
# * Base64.strict_encode64 and Base64.strict_decode64.
#
# ## Padding
#
# Base64-encoding changes a triplet of input bytes into a quartet of output
# characters.
#
# **Padding in Encode Methods**
#
# Padding -- extending an encoded string with zero, one, or two trailing `=`
# characters -- is performed by methods Base64.encode64, Base64.strict_encode64,
# and, by default, Base64.urlsafe_encode64:
#
# Base64.encode64('s') # => "cw==\n"
# Base64.strict_encode64('s') # => "cw=="
# Base64.urlsafe_encode64('s') # => "cw=="
# Base64.urlsafe_encode64('s', padding: false) # => "cw"
#
# When padding is performed, the encoded string is always of length *4n*, where
# `n` is a non-negative integer:
#
# * Input bytes of length *3n* generate unpadded output characters of length
# *4n*:
#
# # n = 1: 3 bytes => 4 characters.
# Base64.strict_encode64('123') # => "MDEy"
# # n = 2: 6 bytes => 8 characters.
# Base64.strict_encode64('123456') # => "MDEyMzQ1"
#
# * Input bytes of length *3n+1* generate padded output characters of length
# *4(n+1)*, with two padding characters at the end:
#
# # n = 1: 4 bytes => 8 characters.
# Base64.strict_encode64('1234') # => "MDEyMw=="
# # n = 2: 7 bytes => 12 characters.
# Base64.strict_encode64('1234567') # => "MDEyMzQ1Ng=="
#
# * Input bytes of length *3n+2* generate padded output characters of length
# *4(n+1)*, with one padding character at the end:
#
# # n = 1: 5 bytes => 8 characters.
# Base64.strict_encode64('12345') # => "MDEyMzQ="
# # n = 2: 8 bytes => 12 characters.
# Base64.strict_encode64('12345678') # => "MDEyMzQ1Njc="
#
# When padding is suppressed, for a positive integer *n*:
#
# * Input bytes of length *3n* generate unpadded output characters of length
# *4n*:
#
# # n = 1: 3 bytes => 4 characters.
# Base64.urlsafe_encode64('123', padding: false) # => "MDEy"
# # n = 2: 6 bytes => 8 characters.
# Base64.urlsafe_encode64('123456', padding: false) # => "MDEyMzQ1"
#
# * Input bytes of length *3n+1* generate unpadded output characters of length
# *4n+2*, with two padding characters at the end:
#
# # n = 1: 4 bytes => 6 characters.
# Base64.urlsafe_encode64('1234', padding: false) # => "MDEyMw"
# # n = 2: 7 bytes => 10 characters.
# Base64.urlsafe_encode64('1234567', padding: false) # => "MDEyMzQ1Ng"
#
# * Input bytes of length *3n+2* generate unpadded output characters of length
# *4n+3*, with one padding character at the end:
#
# # n = 1: 5 bytes => 7 characters.
# Base64.urlsafe_encode64('12345', padding: false) # => "MDEyMzQ"
# # m = 2: 8 bytes => 11 characters.
# Base64.urlsafe_encode64('12345678', padding: false) # => "MDEyMzQ1Njc"
#
# **Padding in Decode Methods**
#
# All of the Base64 decode methods support (but do not require) padding.
#
# Method Base64.decode64 does not check the size of the padding:
#
# Base64.decode64("MDEyMzQ1Njc") # => "01234567"
# Base64.decode64("MDEyMzQ1Njc=") # => "01234567"
# Base64.decode64("MDEyMzQ1Njc==") # => "01234567"
#
# Method Base64.strict_decode64 strictly enforces padding size:
#
# Base64.strict_decode64("MDEyMzQ1Njc") # Raises ArgumentError
# Base64.strict_decode64("MDEyMzQ1Njc=") # => "01234567"
# Base64.strict_decode64("MDEyMzQ1Njc==") # Raises ArgumentError
#
# Method Base64.urlsafe_decode64 allows padding in `str`, which if present, must
# be correct: see [Padding](Base64.html#module-Base64-label-Padding), above:
#
# Base64.urlsafe_decode64("MDEyMzQ1Njc") # => "01234567"
# Base64.urlsafe_decode64("MDEyMzQ1Njc=") # => "01234567"
# Base64.urlsafe_decode64("MDEyMzQ1Njc==") # Raises ArgumentError.
#
# ## Newlines
#
# An encoded string returned by Base64.encode64 or Base64.urlsafe_encode64 has
# an embedded newline character after each 60-character sequence, and, if
# non-empty, at the end:
#
# # No newline if empty.
# encoded = Base64.encode64("\x00" * 0)
# encoded.index("\n") # => nil
#
# # Newline at end of short output.
# encoded = Base64.encode64("\x00" * 1)
# encoded.size # => 4
# encoded.index("\n") # => 4
#
# # Newline at end of longer output.
# encoded = Base64.encode64("\x00" * 45)
# encoded.size # => 60
# encoded.index("\n") # => 60
#
# # Newlines embedded and at end of still longer output.
# encoded = Base64.encode64("\x00" * 46)
# encoded.size # => 65
# encoded.rindex("\n") # => 65
# encoded.split("\n").map {|s| s.size } # => [60, 4]
#
# The string to be encoded may itself contain newlines, which are encoded as
# Base64:
#
# # Base64.encode64("\n\n\n") # => "CgoK\n"
# s = "This is line 1\nThis is line 2\n"
# Base64.encode64(s) # => "VGhpcyBpcyBsaW5lIDEKVGhpcyBpcyBsaW5lIDIK\n"
#
module Base64
# <!--
# rdoc-file=lib/base64.rb
# - decode64(str)
# -->
# Returns a string containing the decoding of an RFC-2045-compliant
# Base64-encoded string `str`:
#
# s = "VGhpcyBpcyBsaW5lIDEKVGhpcyBpcyBsaW5lIDIK\n"
# Base64.decode64(s) # => "This is line 1\nThis is line 2\n"
#
# Non-Base64 characters in `str` are ignored; see [Encoding Character
# Set](Base64.html#module-Base64-label-Encoding+Character+Sets) above: these
# include newline characters and characters `-` and `/`:
#
# Base64.decode64("\x00\n-_") # => ""
#
# Padding in `str` (even if incorrect) is ignored:
#
# Base64.decode64("MDEyMzQ1Njc") # => "01234567"
# Base64.decode64("MDEyMzQ1Njc=") # => "01234567"
# Base64.decode64("MDEyMzQ1Njc==") # => "01234567"
#
def self?.decode64: (String str) -> String

# <!--
# rdoc-file=lib/base64.rb
# - encode64(bin)
# -->
# Returns a string containing the RFC-2045-compliant Base64-encoding of `bin`.
#
# Per RFC 2045, the returned string may contain the URL-unsafe characters `+` or
# `/`; see [Encoding Character
# Set](Base64.html#module-Base64-label-Encoding+Character+Sets) above:
#
# Base64.encode64("\xFB\xEF\xBE") # => "++++\n"
# Base64.encode64("\xFF\xFF\xFF") # => "////\n"
#
# The returned string may include padding; see
# [Padding](Base64.html#module-Base64-label-Padding) above.
#
# Base64.encode64('*') # => "Kg==\n"
#
# The returned string ends with a newline character, and if sufficiently long
# will have one or more embedded newline characters; see
# [Newlines](Base64.html#module-Base64-label-Newlines) above:
#
# Base64.encode64('*') # => "Kg==\n"
# Base64.encode64('*' * 46)
# # => "KioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioq\nKg==\n"
#
# The string to be encoded may itself contain newlines, which will be encoded as
# ordinary Base64:
#
# Base64.encode64("\n\n\n") # => "CgoK\n"
# s = "This is line 1\nThis is line 2\n"
# Base64.encode64(s) # => "VGhpcyBpcyBsaW5lIDEKVGhpcyBpcyBsaW5lIDIK\n"
#
def self?.encode64: (String bin) -> String

# <!--
# rdoc-file=lib/base64.rb
# - strict_decode64(str)
# -->
# Returns a string containing the decoding of an RFC-2045-compliant
# Base64-encoded string `str`:
#
# s = "VGhpcyBpcyBsaW5lIDEKVGhpcyBpcyBsaW5lIDIK"
# Base64.strict_decode64(s) # => "This is line 1\nThis is line 2\n"
#
# Non-Base64 characters in `str` not allowed; see [Encoding Character
# Set](Base64.html#module-Base64-label-Encoding+Character+Sets) above: these
# include newline characters and characters `-` and `/`:
#
# Base64.strict_decode64("\n") # Raises ArgumentError
# Base64.strict_decode64('-') # Raises ArgumentError
# Base64.strict_decode64('_') # Raises ArgumentError
#
# Padding in `str`, if present, must be correct:
#
# Base64.strict_decode64("MDEyMzQ1Njc") # Raises ArgumentError
# Base64.strict_decode64("MDEyMzQ1Njc=") # => "01234567"
# Base64.strict_decode64("MDEyMzQ1Njc==") # Raises ArgumentError
#
def self?.strict_decode64: (String str) -> String

# <!--
# rdoc-file=lib/base64.rb
# - strict_encode64(bin)
# -->
# Returns a string containing the RFC-2045-compliant Base64-encoding of `bin`.
#
# Per RFC 2045, the returned string may contain the URL-unsafe characters `+` or
# `/`; see [Encoding Character
# Set](Base64.html#module-Base64-label-Encoding+Character+Sets) above:
#
# Base64.strict_encode64("\xFB\xEF\xBE") # => "++++\n"
# Base64.strict_encode64("\xFF\xFF\xFF") # => "////\n"
#
# The returned string may include padding; see
# [Padding](Base64.html#module-Base64-label-Padding) above.
#
# Base64.strict_encode64('*') # => "Kg==\n"
#
# The returned string will have no newline characters, regardless of its length;
# see [Newlines](Base64.html#module-Base64-label-Newlines) above:
#
# Base64.strict_encode64('*') # => "Kg=="
# Base64.strict_encode64('*' * 46)
# # => "KioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKg=="
#
# The string to be encoded may itself contain newlines, which will be encoded as
# ordinary Base64:
#
# Base64.strict_encode64("\n\n\n") # => "CgoK"
# s = "This is line 1\nThis is line 2\n"
# Base64.strict_encode64(s) # => "VGhpcyBpcyBsaW5lIDEKVGhpcyBpcyBsaW5lIDIK"
#
def self?.strict_encode64: (String bin) -> String

# <!--
# rdoc-file=lib/base64.rb
# - urlsafe_decode64(str)
# -->
# Returns the decoding of an RFC-4648-compliant Base64-encoded string `str`:
#
# `str` may not contain non-Base64 characters; see [Encoding Character
# Set](Base64.html#module-Base64-label-Encoding+Character+Sets) above:
#
# Base64.urlsafe_decode64('+') # Raises ArgumentError.
# Base64.urlsafe_decode64('/') # Raises ArgumentError.
# Base64.urlsafe_decode64("\n") # Raises ArgumentError.
#
# Padding in `str`, if present, must be correct: see
# [Padding](Base64.html#module-Base64-label-Padding), above:
#
# Base64.urlsafe_decode64("MDEyMzQ1Njc") # => "01234567"
# Base64.urlsafe_decode64("MDEyMzQ1Njc=") # => "01234567"
# Base64.urlsafe_decode64("MDEyMzQ1Njc==") # Raises ArgumentError.
#
def self?.urlsafe_decode64: (String str) -> String

# <!--
# rdoc-file=lib/base64.rb
# - urlsafe_encode64(bin, padding: true)
# -->
# Returns the RFC-4648-compliant Base64-encoding of `bin`.
#
# Per RFC 4648, the returned string will not contain the URL-unsafe characters
# `+` or `/`, but instead may contain the URL-safe characters `-` and `_`; see
# [Encoding Character
# Set](Base64.html#module-Base64-label-Encoding+Character+Sets) above:
#
# Base64.urlsafe_encode64("\xFB\xEF\xBE") # => "----"
# Base64.urlsafe_encode64("\xFF\xFF\xFF") # => "____"
#
# By default, the returned string may have padding; see
# [Padding](Base64.html#module-Base64-label-Padding), above:
#
# Base64.urlsafe_encode64('*') # => "Kg=="
#
# Optionally, you can suppress padding:
#
# Base64.urlsafe_encode64('*', padding: false) # => "Kg"
#
# The returned string will have no newline characters, regardless of its length;
# see [Newlines](Base64.html#module-Base64-label-Newlines) above:
#
# Base64.urlsafe_encode64('*') # => "Kg=="
# Base64.urlsafe_encode64('*' * 46)
# # => "KioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKg=="
#
def self?.urlsafe_encode64: (String bin, ?padding: boolish) -> String
end
Loading