opt
/
alt
/
ruby26
/
lib64
/
ruby
/
2.6.0
/
Go to Home Directory
+
Upload
Create File
root@0UT1S:~$
Execute
By Order of Mr.0UT1S
[DIR] ..
N/A
[DIR] bundler
N/A
[DIR] cgi
N/A
[DIR] csv
N/A
[DIR] digest
N/A
[DIR] drb
N/A
[DIR] e2mmap
N/A
[DIR] fiddle
N/A
[DIR] fileutils
N/A
[DIR] forwardable
N/A
[DIR] io
N/A
[DIR] irb
N/A
[DIR] json
N/A
[DIR] matrix
N/A
[DIR] net
N/A
[DIR] openssl
N/A
[DIR] optparse
N/A
[DIR] psych
N/A
[DIR] racc
N/A
[DIR] rdoc
N/A
[DIR] rexml
N/A
[DIR] rinda
N/A
[DIR] ripper
N/A
[DIR] rss
N/A
[DIR] rubygems
N/A
[DIR] shell
N/A
[DIR] syslog
N/A
[DIR] thwait
N/A
[DIR] tracer
N/A
[DIR] unicode_normalize
N/A
[DIR] uri
N/A
[DIR] webrick
N/A
[DIR] x86_64-linux
N/A
[DIR] yaml
N/A
English.rb
6.31 KB
Rename
Delete
abbrev.rb
3.49 KB
Rename
Delete
base64.rb
3.30 KB
Rename
Delete
benchmark.rb
18.07 KB
Rename
Delete
bigdecimal.rb
179 bytes
Rename
Delete
bundler.rb
17.82 KB
Rename
Delete
cgi.rb
9.80 KB
Rename
Delete
cmath.rb
9.49 KB
Rename
Delete
coverage.rb
368 bytes
Rename
Delete
csv.rb
52.00 KB
Rename
Delete
date.rb
1.01 KB
Rename
Delete
debug.rb
29.97 KB
Rename
Delete
delegate.rb
10.44 KB
Rename
Delete
digest.rb
2.83 KB
Rename
Delete
drb.rb
50 bytes
Rename
Delete
e2mmap.rb
3.94 KB
Rename
Delete
erb.rb
28.73 KB
Rename
Delete
expect.rb
2.17 KB
Rename
Delete
fiddle.rb
1.68 KB
Rename
Delete
fileutils.rb
47.16 KB
Rename
Delete
find.rb
2.47 KB
Rename
Delete
forwardable.rb
8.49 KB
Rename
Delete
getoptlong.rb
15.42 KB
Rename
Delete
ipaddr.rb
19.56 KB
Rename
Delete
irb.rb
22.53 KB
Rename
Delete
json.rb
1.77 KB
Rename
Delete
kconv.rb
5.77 KB
Rename
Delete
logger.rb
23.47 KB
Rename
Delete
matrix.rb
60.17 KB
Rename
Delete
mkmf.rb
84.87 KB
Rename
Delete
monitor.rb
7.87 KB
Rename
Delete
mutex_m.rb
2.16 KB
Rename
Delete
observer.rb
5.83 KB
Rename
Delete
open-uri.rb
25.13 KB
Rename
Delete
open3.rb
21.81 KB
Rename
Delete
openssl.rb
469 bytes
Rename
Delete
optionparser.rb
59 bytes
Rename
Delete
optparse.rb
57.74 KB
Rename
Delete
ostruct.rb
10.65 KB
Rename
Delete
pathname.rb
16.17 KB
Rename
Delete
pp.rb
15.08 KB
Rename
Delete
prettyprint.rb
15.89 KB
Rename
Delete
prime.rb
12.37 KB
Rename
Delete
profile.rb
235 bytes
Rename
Delete
profiler.rb
4.54 KB
Rename
Delete
pstore.rb
14.70 KB
Rename
Delete
psych.rb
21.11 KB
Rename
Delete
rdoc.rb
4.88 KB
Rename
Delete
resolv-replace.rb
1.76 KB
Rename
Delete
resolv.rb
73.44 KB
Rename
Delete
ripper.rb
2.44 KB
Rename
Delete
rss.rb
2.87 KB
Rename
Delete
rubygems.rb
36.11 KB
Rename
Delete
scanf.rb
23.55 KB
Rename
Delete
securerandom.rb
8.98 KB
Rename
Delete
set.rb
24.07 KB
Rename
Delete
shell.rb
11.39 KB
Rename
Delete
shellwords.rb
6.66 KB
Rename
Delete
singleton.rb
4.04 KB
Rename
Delete
socket.rb
43.51 KB
Rename
Delete
sync.rb
7.23 KB
Rename
Delete
tempfile.rb
11.21 KB
Rename
Delete
thwait.rb
3.33 KB
Rename
Delete
time.rb
23.56 KB
Rename
Delete
timeout.rb
3.81 KB
Rename
Delete
tmpdir.rb
4.25 KB
Rename
Delete
tracer.rb
6.42 KB
Rename
Delete
tsort.rb
14.30 KB
Rename
Delete
un.rb
9.95 KB
Rename
Delete
uri.rb
3.12 KB
Rename
Delete
weakref.rb
1.44 KB
Rename
Delete
webrick.rb
6.72 KB
Rename
Delete
yaml.rb
1.81 KB
Rename
Delete
# frozen_string_literal: true # # This class implements a pretty printing algorithm. It finds line breaks and # nice indentations for grouped structure. # # By default, the class assumes that primitive elements are strings and each # byte in the strings have single column in width. But it can be used for # other situations by giving suitable arguments for some methods: # * newline object and space generation block for PrettyPrint.new # * optional width argument for PrettyPrint#text # * PrettyPrint#breakable # # There are several candidate uses: # * text formatting using proportional fonts # * multibyte characters which has columns different to number of bytes # * non-string formatting # # == Bugs # * Box based formatting? # * Other (better) model/algorithm? # # Report any bugs at http://bugs.ruby-lang.org # # == References # Christian Lindig, Strictly Pretty, March 2000, # http://www.st.cs.uni-sb.de/~lindig/papers/#pretty # # Philip Wadler, A prettier printer, March 1998, # http://homepages.inf.ed.ac.uk/wadler/topics/language-design.html#prettier # # == Author # Tanaka Akira <akr@fsij.org> # class PrettyPrint # This is a convenience method which is same as follows: # # begin # q = PrettyPrint.new(output, maxwidth, newline, &genspace) # ... # q.flush # output # end # def PrettyPrint.format(output=''.dup, maxwidth=79, newline="\n", genspace=lambda {|n| ' ' * n}) q = PrettyPrint.new(output, maxwidth, newline, &genspace) yield q q.flush output end # This is similar to PrettyPrint::format but the result has no breaks. # # +maxwidth+, +newline+ and +genspace+ are ignored. # # The invocation of +breakable+ in the block doesn't break a line and is # treated as just an invocation of +text+. # def PrettyPrint.singleline_format(output=''.dup, maxwidth=nil, newline=nil, genspace=nil) q = SingleLine.new(output) yield q output end # Creates a buffer for pretty printing. # # +output+ is an output target. If it is not specified, '' is assumed. It # should have a << method which accepts the first argument +obj+ of # PrettyPrint#text, the first argument +sep+ of PrettyPrint#breakable, the # first argument +newline+ of PrettyPrint.new, and the result of a given # block for PrettyPrint.new. # # +maxwidth+ specifies maximum line length. If it is not specified, 79 is # assumed. However actual outputs may overflow +maxwidth+ if long # non-breakable texts are provided. # # +newline+ is used for line breaks. "\n" is used if it is not specified. # # The block is used to generate spaces. {|width| ' ' * width} is used if it # is not given. # def initialize(output=''.dup, maxwidth=79, newline="\n", &genspace) @output = output @maxwidth = maxwidth @newline = newline @genspace = genspace || lambda {|n| ' ' * n} @output_width = 0 @buffer_width = 0 @buffer = [] root_group = Group.new(0) @group_stack = [root_group] @group_queue = GroupQueue.new(root_group) @indent = 0 end # The output object. # # This defaults to '', and should accept the << method attr_reader :output # The maximum width of a line, before it is separated in to a newline # # This defaults to 79, and should be a Fixnum attr_reader :maxwidth # The value that is appended to +output+ to add a new line. # # This defaults to "\n", and should be String attr_reader :newline # A lambda or Proc, that takes one argument, of a Fixnum, and returns # the corresponding number of spaces. # # By default this is: # lambda {|n| ' ' * n} attr_reader :genspace # The number of spaces to be indented attr_reader :indent # The PrettyPrint::GroupQueue of groups in stack to be pretty printed attr_reader :group_queue # Returns the group most recently added to the stack. # # Contrived example: # out = "" # => "" # q = PrettyPrint.new(out) # => #<PrettyPrint:0x82f85c0 @output="", @maxwidth=79, @newline="\n", @genspace=#<Proc:0x82f8368@/home/vbatts/.rvm/rubies/ruby-head/lib/ruby/2.0.0/prettyprint.rb:82 (lambda)>, @output_width=0, @buffer_width=0, @buffer=[], @group_stack=[#<PrettyPrint::Group:0x82f8138 @depth=0, @breakables=[], @break=false>], @group_queue=#<PrettyPrint::GroupQueue:0x82fb7c0 @queue=[[#<PrettyPrint::Group:0x82f8138 @depth=0, @breakables=[], @break=false>]]>, @indent=0> # q.group { # q.text q.current_group.inspect # q.text q.newline # q.group(q.current_group.depth + 1) { # q.text q.current_group.inspect # q.text q.newline # q.group(q.current_group.depth + 1) { # q.text q.current_group.inspect # q.text q.newline # q.group(q.current_group.depth + 1) { # q.text q.current_group.inspect # q.text q.newline # } # } # } # } # => 284 # puts out # #<PrettyPrint::Group:0x8354758 @depth=1, @breakables=[], @break=false> # #<PrettyPrint::Group:0x8354550 @depth=2, @breakables=[], @break=false> # #<PrettyPrint::Group:0x83541cc @depth=3, @breakables=[], @break=false> # #<PrettyPrint::Group:0x8347e54 @depth=4, @breakables=[], @break=false> def current_group @group_stack.last end # Breaks the buffer into lines that are shorter than #maxwidth def break_outmost_groups while @maxwidth < @output_width + @buffer_width return unless group = @group_queue.deq until group.breakables.empty? data = @buffer.shift @output_width = data.output(@output, @output_width) @buffer_width -= data.width end while !@buffer.empty? && Text === @buffer.first text = @buffer.shift @output_width = text.output(@output, @output_width) @buffer_width -= text.width end end end # This adds +obj+ as a text of +width+ columns in width. # # If +width+ is not specified, obj.length is used. # def text(obj, width=obj.length) if @buffer.empty? @output << obj @output_width += width else text = @buffer.last unless Text === text text = Text.new @buffer << text end text.add(obj, width) @buffer_width += width break_outmost_groups end end # This is similar to #breakable except # the decision to break or not is determined individually. # # Two #fill_breakable under a group may cause 4 results: # (break,break), (break,non-break), (non-break,break), (non-break,non-break). # This is different to #breakable because two #breakable under a group # may cause 2 results: # (break,break), (non-break,non-break). # # The text +sep+ is inserted if a line is not broken at this point. # # If +sep+ is not specified, " " is used. # # If +width+ is not specified, +sep.length+ is used. You will have to # specify this when +sep+ is a multibyte character, for example. # def fill_breakable(sep=' ', width=sep.length) group { breakable sep, width } end # This says "you can break a line here if necessary", and a +width+\-column # text +sep+ is inserted if a line is not broken at the point. # # If +sep+ is not specified, " " is used. # # If +width+ is not specified, +sep.length+ is used. You will have to # specify this when +sep+ is a multibyte character, for example. # def breakable(sep=' ', width=sep.length) group = @group_stack.last if group.break? flush @output << @newline @output << @genspace.call(@indent) @output_width = @indent @buffer_width = 0 else @buffer << Breakable.new(sep, width, self) @buffer_width += width break_outmost_groups end end # Groups line break hints added in the block. The line break hints are all # to be used or not. # # If +indent+ is specified, the method call is regarded as nested by # nest(indent) { ... }. # # If +open_obj+ is specified, <tt>text open_obj, open_width</tt> is called # before grouping. If +close_obj+ is specified, <tt>text close_obj, # close_width</tt> is called after grouping. # def group(indent=0, open_obj='', close_obj='', open_width=open_obj.length, close_width=close_obj.length) text open_obj, open_width group_sub { nest(indent) { yield } } text close_obj, close_width end # Takes a block and queues a new group that is indented 1 level further. def group_sub group = Group.new(@group_stack.last.depth + 1) @group_stack.push group @group_queue.enq group begin yield ensure @group_stack.pop if group.breakables.empty? @group_queue.delete group end end end # Increases left margin after newline with +indent+ for line breaks added in # the block. # def nest(indent) @indent += indent begin yield ensure @indent -= indent end end # outputs buffered data. # def flush @buffer.each {|data| @output_width = data.output(@output, @output_width) } @buffer.clear @buffer_width = 0 end # The Text class is the means by which to collect strings from objects. # # This class is intended for internal use of the PrettyPrint buffers. class Text # :nodoc: # Creates a new text object. # # This constructor takes no arguments. # # The workflow is to append a PrettyPrint::Text object to the buffer, and # being able to call the buffer.last() to reference it. # # As there are objects, use PrettyPrint::Text#add to include the objects # and the width to utilized by the String version of this object. def initialize @objs = [] @width = 0 end # The total width of the objects included in this Text object. attr_reader :width # Render the String text of the objects that have been added to this Text object. # # Output the text to +out+, and increment the width to +output_width+ def output(out, output_width) @objs.each {|obj| out << obj} output_width + @width end # Include +obj+ in the objects to be pretty printed, and increment # this Text object's total width by +width+ def add(obj, width) @objs << obj @width += width end end # The Breakable class is used for breaking up object information # # This class is intended for internal use of the PrettyPrint buffers. class Breakable # :nodoc: # Create a new Breakable object. # # Arguments: # * +sep+ String of the separator # * +width+ Fixnum width of the +sep+ # * +q+ parent PrettyPrint object, to base from def initialize(sep, width, q) @obj = sep @width = width @pp = q @indent = q.indent @group = q.current_group @group.breakables.push self end # Holds the separator String # # The +sep+ argument from ::new attr_reader :obj # The width of +obj+ / +sep+ attr_reader :width # The number of spaces to indent. # # This is inferred from +q+ within PrettyPrint, passed in ::new attr_reader :indent # Render the String text of the objects that have been added to this # Breakable object. # # Output the text to +out+, and increment the width to +output_width+ def output(out, output_width) @group.breakables.shift if @group.break? out << @pp.newline out << @pp.genspace.call(@indent) @indent else @pp.group_queue.delete @group if @group.breakables.empty? out << @obj output_width + @width end end end # The Group class is used for making indentation easier. # # While this class does neither the breaking into newlines nor indentation, # it is used in a stack (as well as a queue) within PrettyPrint, to group # objects. # # For information on using groups, see PrettyPrint#group # # This class is intended for internal use of the PrettyPrint buffers. class Group # :nodoc: # Create a Group object # # Arguments: # * +depth+ - this group's relation to previous groups def initialize(depth) @depth = depth @breakables = [] @break = false end # This group's relation to previous groups attr_reader :depth # Array to hold the Breakable objects for this Group attr_reader :breakables # Makes a break for this Group, and returns true def break @break = true end # Boolean of whether this Group has made a break def break? @break end # Boolean of whether this Group has been queried for being first # # This is used as a predicate, and ought to be called first. def first? if defined? @first false else @first = false true end end end # The GroupQueue class is used for managing the queue of Group to be pretty # printed. # # This queue groups the Group objects, based on their depth. # # This class is intended for internal use of the PrettyPrint buffers. class GroupQueue # :nodoc: # Create a GroupQueue object # # Arguments: # * +groups+ - one or more PrettyPrint::Group objects def initialize(*groups) @queue = [] groups.each {|g| enq g} end # Enqueue +group+ # # This does not strictly append the group to the end of the queue, # but instead adds it in line, base on the +group.depth+ def enq(group) depth = group.depth @queue << [] until depth < @queue.length @queue[depth] << group end # Returns the outer group of the queue def deq @queue.each {|gs| (gs.length-1).downto(0) {|i| unless gs[i].breakables.empty? group = gs.slice!(i, 1).first group.break return group end } gs.each {|group| group.break} gs.clear } return nil end # Remote +group+ from this queue def delete(group) @queue[group.depth].delete(group) end end # PrettyPrint::SingleLine is used by PrettyPrint.singleline_format # # It is passed to be similar to a PrettyPrint object itself, by responding to: # * #text # * #breakable # * #nest # * #group # * #flush # * #first? # # but instead, the output has no line breaks # class SingleLine # Create a PrettyPrint::SingleLine object # # Arguments: # * +output+ - String (or similar) to store rendered text. Needs to respond to '<<' # * +maxwidth+ - Argument position expected to be here for compatibility. # This argument is a noop. # * +newline+ - Argument position expected to be here for compatibility. # This argument is a noop. def initialize(output, maxwidth=nil, newline=nil) @output = output @first = [true] end # Add +obj+ to the text to be output. # # +width+ argument is here for compatibility. It is a noop argument. def text(obj, width=nil) @output << obj end # Appends +sep+ to the text to be output. By default +sep+ is ' ' # # +width+ argument is here for compatibility. It is a noop argument. def breakable(sep=' ', width=nil) @output << sep end # Takes +indent+ arg, but does nothing with it. # # Yields to a block. def nest(indent) # :nodoc: yield end # Opens a block for grouping objects to be pretty printed. # # Arguments: # * +indent+ - noop argument. Present for compatibility. # * +open_obj+ - text appended before the &blok. Default is '' # * +close_obj+ - text appended after the &blok. Default is '' # * +open_width+ - noop argument. Present for compatibility. # * +close_width+ - noop argument. Present for compatibility. def group(indent=nil, open_obj='', close_obj='', open_width=nil, close_width=nil) @first.push true @output << open_obj yield @output << close_obj @first.pop end # Method present for compatibility, but is a noop def flush # :nodoc: end # This is used as a predicate, and ought to be called first. def first? result = @first[-1] @first[-1] = false result end end end
Save