1
0
Fork 0
mirror of https://github.com/ruby/ruby.git synced 2022-11-09 12:17:21 -05:00
ruby--ruby/spec/ruby/language/string_spec.rb
2022-04-25 14:53:54 +02:00

299 lines
8.7 KiB
Ruby

# -*- encoding: binary -*-
require_relative '../spec_helper'
# TODO: rewrite these horrid specs. it "are..." seriously?!
describe "Ruby character strings" do
before :each do
@ip = 'xxx' # used for interpolation
$ip = 'xxx'
end
it "don't get interpolated when put in single quotes" do
'#{@ip}'.should == '#{@ip}'
end
it 'get interpolated with #{} when put in double quotes' do
"#{@ip}".should == 'xxx'
end
it "interpolate instance variables just with the # character" do
"#@ip".should == 'xxx'
end
it "interpolate global variables just with the # character" do
"#$ip".should == 'xxx'
end
it "allows underscore as part of a variable name in a simple interpolation" do
@my_ip = 'xxx'
"#@my_ip".should == 'xxx'
end
it "does not interpolate invalid variable names" do
"#@".should == '#@'
"#$%".should == '#$%'
end
it "has characters [.(=?!# end simple # interpolation" do
"#@ip[".should == 'xxx['
"#@ip.".should == 'xxx.'
"#@ip(".should == 'xxx('
"#@ip=".should == 'xxx='
"#@ip?".should == 'xxx?'
"#@ip!".should == 'xxx!'
"#@ip#@ip".should == 'xxxxxx'
end
it "don't get confused by partial interpolation character sequences" do
"#@".should == '#@'
"#@ ".should == '#@ '
"#@@".should == '#@@'
"#@@ ".should == '#@@ '
"#$ ".should == '#$ '
"#\$".should == '#$'
end
it "allows using non-alnum characters as string delimiters" do
%(hey #{@ip}).should == "hey xxx"
%[hey #{@ip}].should == "hey xxx"
%{hey #{@ip}}.should == "hey xxx"
%<hey #{@ip}>.should == "hey xxx"
%!hey #{@ip}!.should == "hey xxx"
%@hey #{@ip}@.should == "hey xxx"
%#hey hey#.should == "hey hey"
%%hey #{@ip}%.should == "hey xxx"
%^hey #{@ip}^.should == "hey xxx"
%&hey #{@ip}&.should == "hey xxx"
%*hey #{@ip}*.should == "hey xxx"
%-hey #{@ip}-.should == "hey xxx"
%_hey #{@ip}_.should == "hey xxx"
%=hey #{@ip}=.should == "hey xxx"
%+hey #{@ip}+.should == "hey xxx"
%~hey #{@ip}~.should == "hey xxx"
%:hey #{@ip}:.should == "hey xxx"
%;hey #{@ip};.should == "hey xxx"
%"hey #{@ip}".should == "hey xxx"
%|hey #{@ip}|.should == "hey xxx"
%?hey #{@ip}?.should == "hey xxx"
%/hey #{@ip}/.should == "hey xxx"
%,hey #{@ip},.should == "hey xxx"
%.hey #{@ip}..should == "hey xxx"
# surprised? huh
%'hey #{@ip}'.should == "hey xxx"
%\hey #{@ip}\.should == "hey xxx"
%`hey #{@ip}`.should == "hey xxx"
%$hey #{@ip}$.should == "hey xxx"
end
it "using percent with 'q', stopping interpolation" do
%q(#{@ip}).should == '#{@ip}'
end
it "using percent with 'Q' to interpolate" do
%Q(#{@ip}).should == 'xxx'
end
# The backslashes :
#
# \t (tab), \n (newline), \r (carriage return), \f (form feed), \b
# (backspace), \a (bell), \e (escape), \s (whitespace), \nnn (octal),
# \xnn (hexadecimal), \cx (control x), \C-x (control x), \M-x (meta x),
# \M-\C-x (meta control x)
it "backslashes follow the same rules as interpolation" do
"\t\n\r\f\b\a\e\s\075\x62\cx".should == "\t\n\r\f\b\a\e =b\030"
'\t\n\r\f\b\a\e =b\030'.should == "\\t\\n\\r\\f\\b\\a\\e =b\\030"
end
it "calls #to_s when the object is not a String" do
obj = mock('to_s')
obj.stub!(:to_s).and_return('42')
"#{obj}".should == '42'
end
it "calls #to_s as a private method" do
obj = mock('to_s')
obj.stub!(:to_s).and_return('42')
class << obj
private :to_s
end
"#{obj}".should == '42'
end
it "uses an internal representation when #to_s doesn't return a String" do
obj = mock('to_s')
obj.stub!(:to_s).and_return(42)
# See rubyspec commit 787c132d by yugui. There is value in
# ensuring that this behavior works. So rather than removing
# this spec completely, the only thing that can be asserted
# is that if you interpolate an object that fails to return
# a String, you will still get a String and not raise an
# exception.
"#{obj}".should be_an_instance_of(String)
end
it "allows a dynamic string to parse a nested do...end block as an argument to a call without parens, interpolated" do
s = eval 'eval "#{proc do; 1; end.call}"'
s.should == 1
end
it "are produced from character shortcuts" do
?z.should == 'z'
end
it "are produced from control character shortcuts" do
# Control-Z
?\C-z.should == "\x1A"
# Meta-Z
?\M-z.should == "\xFA"
# Meta-Control-Z
?\M-\C-z.should == "\x9A"
end
describe "Unicode escaping" do
it "can be done with \\u and four hex digits" do
[ ["\u0000", 0x0000],
["\u2020", 0x2020]
].should be_computed_by(:ord)
end
it "can be done with \\u{} and one to six hex digits" do
[ ["\u{a}", 0xa],
["\u{ab}", 0xab],
["\u{abc}", 0xabc],
["\u{1abc}", 0x1abc],
["\u{12abc}", 0x12abc],
["\u{100000}", 0x100000]
].should be_computed_by(:ord)
end
# TODO: spec other source encodings
describe "with ASCII_8BIT source encoding" do
it "produces an ASCII string when escaping ASCII characters via \\u" do
"\u0000".encoding.should == Encoding::BINARY
end
it "produces an ASCII string when escaping ASCII characters via \\u{}" do
"\u{0000}".encoding.should == Encoding::BINARY
end
it "produces a UTF-8-encoded string when escaping non-ASCII characters via \\u" do
"\u1234".encoding.should == Encoding::UTF_8
end
it "produces a UTF-8-encoded string when escaping non-ASCII characters via \\u{}" do
"\u{1234}".encoding.should == Encoding::UTF_8
end
end
end
end
# TODO: rewrite all specs above this
describe "Ruby String literals" do
def str_concat
"foo" "bar" "baz"
end
def long_string_literals
"Beautiful is better than ugly." \
"Explicit is better than implicit."
end
it "on a single line with spaces in between are concatenated together" do
str_concat.should == "foobarbaz"
end
it "on multiple lines with newlines and backslash in between are concatenated together" do
long_string_literals.should == "Beautiful is better than ugly.Explicit is better than implicit."
end
describe "with a magic frozen comment" do
it "produce the same object each time" do
ruby_exe(fixture(__FILE__, "freeze_magic_comment_one_literal.rb")).chomp.should == "true"
end
it "produce the same object for literals with the same content" do
ruby_exe(fixture(__FILE__, "freeze_magic_comment_two_literals.rb")).chomp.should == "true"
end
it "produce the same object for literals with the same content in different files" do
ruby_exe(fixture(__FILE__, "freeze_magic_comment_across_files.rb")).chomp.should == "true"
end
it "produce different objects for literals with the same content in different files if the other file doesn't have the comment" do
ruby_exe(fixture(__FILE__, "freeze_magic_comment_across_files_no_comment.rb")).chomp.should == "true"
end
it "produce different objects for literals with the same content in different files if they have different encodings" do
ruby_exe(fixture(__FILE__, "freeze_magic_comment_across_files_diff_enc.rb")).chomp.should == "true"
end
end
end
describe "Ruby String interpolation" do
it "permits an empty expression" do
s = "#{}" # rubocop:disable Lint/EmptyInterpolation
s.should.empty?
s.should_not.frozen?
end
it "returns a string with the source encoding by default" do
"a#{"b"}c".encoding.should == Encoding::BINARY
eval('"a#{"b"}c"'.force_encoding("us-ascii")).encoding.should == Encoding::US_ASCII
eval("# coding: US-ASCII \n 'a#{"b"}c'").encoding.should == Encoding::US_ASCII
end
it "returns a string with the source encoding, even if the components have another encoding" do
a = "abc".force_encoding("euc-jp")
"#{a}".encoding.should == Encoding::BINARY
b = "abc".encode("utf-8")
"#{b}".encoding.should == Encoding::BINARY
end
it "raises an Encoding::CompatibilityError if the Encodings are not compatible" do
a = "\u3042"
b = "\xff".force_encoding "binary"
-> { "#{a} #{b}" }.should raise_error(Encoding::CompatibilityError)
end
it "creates a non-frozen String" do
code = <<~'RUBY'
"a#{6*7}c"
RUBY
eval(code).should_not.frozen?
end
ruby_version_is "3.0" do
it "creates a non-frozen String when # frozen-string-literal: true is used" do
code = <<~'RUBY'
# frozen-string-literal: true
"a#{6*7}c"
RUBY
eval(code).should_not.frozen?
end
end
ruby_version_is ""..."3.0" do
it "creates a frozen String when # frozen-string-literal: true is used" do
code = <<~'RUBY'
# frozen-string-literal: true
"a#{6*7}c"
RUBY
eval(code).should.frozen?
end
end
end