Class
Constants
No documentation available
No documentation available
No documentation available
No documentation available
No documentation available
No documentation available
No documentation available
No documentation available
No documentation available
Class Methods
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 107
def self.calculate_width(str, allow_escape_code = false)
if allow_escape_code
width = 0
rest = str.encode(Encoding::UTF_8)
in_zero_width = false
rest.scan(WIDTH_SCANNER) do |non_printing_start, non_printing_end, csi, osc, gc|
case
when non_printing_start
in_zero_width = true
when non_printing_end
in_zero_width = false
when csi, osc
when gc
unless in_zero_width
width += get_mbchar_width(gc)
end
end
end
width
else
str.encode(Encoding::UTF_8).grapheme_clusters.inject(0) { |w, gc|
w + get_mbchar_width(gc)
}
end
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 361
def self.ed_transpose_words(line, byte_pointer)
right_word_start = nil
size = get_next_mbchar_size(line, byte_pointer)
mbchar = line.byteslice(byte_pointer, size)
if size.zero?
# ' aaa bbb [cursor]'
byte_size = 0
while 0 < (byte_pointer + byte_size)
size = get_prev_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size - size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\p{Word}/
byte_size -= size
end
while 0 < (byte_pointer + byte_size)
size = get_prev_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size - size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
byte_size -= size
end
right_word_start = byte_pointer + byte_size
byte_size = 0
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
byte_size += size
end
after_start = byte_pointer + byte_size
elsif mbchar.encode(Encoding::UTF_8) =~ /\p{Word}/
# ' aaa bb[cursor]b'
byte_size = 0
while 0 < (byte_pointer + byte_size)
size = get_prev_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size - size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
byte_size -= size
end
right_word_start = byte_pointer + byte_size
byte_size = 0
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
byte_size += size
end
after_start = byte_pointer + byte_size
else
byte_size = 0
while (line.bytesize - 1) > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\p{Word}/
byte_size += size
end
if (byte_pointer + byte_size) == (line.bytesize - 1)
# ' aaa bbb [cursor] '
after_start = line.bytesize
while 0 < (byte_pointer + byte_size)
size = get_prev_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size - size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\p{Word}/
byte_size -= size
end
while 0 < (byte_pointer + byte_size)
size = get_prev_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size - size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
byte_size -= size
end
right_word_start = byte_pointer + byte_size
else
# ' aaa [cursor] bbb '
right_word_start = byte_pointer + byte_size
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
byte_size += size
end
after_start = byte_pointer + byte_size
end
end
byte_size = right_word_start - byte_pointer
while 0 < (byte_pointer + byte_size)
size = get_prev_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size - size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\p{Word}/
byte_size -= size
end
middle_start = byte_pointer + byte_size
byte_size = middle_start - byte_pointer
while 0 < (byte_pointer + byte_size)
size = get_prev_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size - size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
byte_size -= size
end
left_word_start = byte_pointer + byte_size
[left_word_start, middle_start, right_word_start, after_start]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 321
def self.em_backward_word(line, byte_pointer)
width = 0
byte_size = 0
while 0 < (byte_pointer - byte_size)
size = get_prev_mbchar_size(line, byte_pointer - byte_size)
mbchar = line.byteslice(byte_pointer - byte_size - size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\p{Word}/
width += get_mbchar_width(mbchar)
byte_size += size
end
while 0 < (byte_pointer - byte_size)
size = get_prev_mbchar_size(line, byte_pointer - byte_size)
mbchar = line.byteslice(byte_pointer - byte_size - size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
width += get_mbchar_width(mbchar)
byte_size += size
end
[byte_size, width]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 341
def self.em_big_backward_word(line, byte_pointer)
width = 0
byte_size = 0
while 0 < (byte_pointer - byte_size)
size = get_prev_mbchar_size(line, byte_pointer - byte_size)
mbchar = line.byteslice(byte_pointer - byte_size - size, size)
break if mbchar =~ /\S/
width += get_mbchar_width(mbchar)
byte_size += size
end
while 0 < (byte_pointer - byte_size)
size = get_prev_mbchar_size(line, byte_pointer - byte_size)
mbchar = line.byteslice(byte_pointer - byte_size - size, size)
break if mbchar =~ /\s/
width += get_mbchar_width(mbchar)
byte_size += size
end
[byte_size, width]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 272
def self.em_forward_word(line, byte_pointer)
width = 0
byte_size = 0
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\p{Word}/
width += get_mbchar_width(mbchar)
byte_size += size
end
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
width += get_mbchar_width(mbchar)
byte_size += size
end
[byte_size, width]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 292
def self.em_forward_word_with_capitalization(line, byte_pointer)
width = 0
byte_size = 0
new_str = String.new
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\p{Word}/
new_str += mbchar
width += get_mbchar_width(mbchar)
byte_size += size
end
first = true
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar.encode(Encoding::UTF_8) =~ /\P{Word}/
if first
new_str += mbchar.upcase
first = false
else
new_str += mbchar.downcase
end
width += get_mbchar_width(mbchar)
byte_size += size
end
[byte_size, width, new_str]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 44
def self.escape_for_print(str)
str.chars.map! { |gr|
case gr
when -"\n"
gr
when -"\t"
-' '
else
EscapedPairs[gr.ord] || gr
end
}.join
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 87
def self.get_mbchar_width(mbchar)
ord = mbchar.ord
if (0x00 <= ord and ord <= 0x1F) # in EscapedPairs
return 2
elsif (0x20 <= ord and ord <= 0x7E) # printable ASCII chars
return 1
end
m = mbchar.encode(Encoding::UTF_8).match(MBCharWidthRE)
case
when m.nil? then 1 # TODO should be U+FFFD � REPLACEMENT CHARACTER
when m[:width_2_1], m[:width_2_2], m[:width_2_3] then 2
when m[:width_3] then 3
when m[:width_0] then 0
when m[:width_1] then 1
when m[:ambiguous_width] then Reline.ambiguous_width
else
nil
end
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 258
def self.get_next_mbchar_size(line, byte_pointer)
grapheme = line.byteslice(byte_pointer..-1).grapheme_clusters.first
grapheme ? grapheme.bytesize : 0
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 263
def self.get_prev_mbchar_size(line, byte_pointer)
if byte_pointer.zero?
0
else
grapheme = line.byteslice(0..(byte_pointer - 1)).grapheme_clusters.last
grapheme ? grapheme.bytesize : 0
end
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 133
def self.split_by_width(str, max_width, encoding = str.encoding, offset: 0)
lines = [String.new(encoding: encoding)]
height = 1
width = offset
rest = str.encode(Encoding::UTF_8)
in_zero_width = false
seq = String.new(encoding: encoding)
rest.scan(WIDTH_SCANNER) do |non_printing_start, non_printing_end, csi, osc, gc|
case
when non_printing_start
in_zero_width = true
lines.last << NON_PRINTING_START
when non_printing_end
in_zero_width = false
lines.last << NON_PRINTING_END
when csi
lines.last << csi
unless in_zero_width
if csi == -"\e[m" || csi == -"\e[0m"
seq.clear
else
seq << csi
end
end
when osc
lines.last << osc
seq << osc
when gc
unless in_zero_width
mbchar_width = get_mbchar_width(gc)
if (width += mbchar_width) > max_width
width = mbchar_width
lines << nil
lines << seq.dup
height += 1
end
end
lines.last << gc
end
end
# The cursor moves to next line in first
if width == max_width
lines << nil
lines << String.new(encoding: encoding)
height += 1
end
[lines, height]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 187
def self.take_mbchar_range(str, start_col, width, cover_begin: false, cover_end: false, padding: false)
chunk = String.new(encoding: str.encoding)
end_col = start_col + width
total_width = 0
rest = str.encode(Encoding::UTF_8)
in_zero_width = false
chunk_start_col = nil
chunk_end_col = nil
has_csi = false
rest.scan(WIDTH_SCANNER) do |non_printing_start, non_printing_end, csi, osc, gc|
case
when non_printing_start
in_zero_width = true
chunk << NON_PRINTING_START
when non_printing_end
in_zero_width = false
chunk << NON_PRINTING_END
when csi
has_csi = true
chunk << csi
when osc
chunk << osc
when gc
if in_zero_width
chunk << gc
next
end
mbchar_width = get_mbchar_width(gc)
prev_width = total_width
total_width += mbchar_width
if (cover_begin || padding ? total_width <= start_col : prev_width < start_col)
# Current character haven't reached start_col yet
next
elsif padding && !cover_begin && prev_width < start_col && start_col < total_width
# Add preceding padding. This padding might have background color.
chunk << ' '
chunk_start_col ||= start_col
chunk_end_col = total_width
next
elsif (cover_end ? prev_width < end_col : total_width <= end_col)
# Current character is in the range
chunk << gc
chunk_start_col ||= prev_width
chunk_end_col = total_width
break if total_width >= end_col
else
# Current character exceeds end_col
if padding && end_col < total_width
# Add succeeding padding. This padding might have background color.
chunk << ' '
chunk_start_col ||= prev_width
chunk_end_col = end_col
end
break
end
end
end
chunk_start_col ||= start_col
chunk_end_col ||= start_col
if padding && chunk_end_col < end_col
# Append padding. This padding should not include background color.
chunk << "\e[0m" if has_csi
chunk << ' ' * (end_col - chunk_end_col)
chunk_end_col = end_col
end
[chunk, chunk_start_col, chunk_end_col - chunk_start_col]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 183
def self.take_range(str, start_col, max_width)
take_mbchar_range(str, start_col, max_width).first
end
Take a chunk of a String
cut by width with escape sequences.
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 650
def self.vi_backward_word(line, byte_pointer)
width = 0
byte_size = 0
while 0 < (byte_pointer - byte_size)
size = get_prev_mbchar_size(line, byte_pointer - byte_size)
mbchar = line.byteslice(byte_pointer - byte_size - size, size)
if mbchar =~ /\S/
if mbchar =~ /\w/
started_by = :word
else
started_by = :non_word_printable
end
break
end
width += get_mbchar_width(mbchar)
byte_size += size
end
while 0 < (byte_pointer - byte_size)
size = get_prev_mbchar_size(line, byte_pointer - byte_size)
mbchar = line.byteslice(byte_pointer - byte_size - size, size)
case started_by
when :word
break if mbchar =~ /\W/
when :non_word_printable
break if mbchar =~ /[\w\s]/
end
width += get_mbchar_width(mbchar)
byte_size += size
end
[byte_size, width]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 512
def self.vi_big_backward_word(line, byte_pointer)
width = 0
byte_size = 0
while 0 < (byte_pointer - byte_size)
size = get_prev_mbchar_size(line, byte_pointer - byte_size)
mbchar = line.byteslice(byte_pointer - byte_size - size, size)
break if mbchar =~ /\S/
width += get_mbchar_width(mbchar)
byte_size += size
end
while 0 < (byte_pointer - byte_size)
size = get_prev_mbchar_size(line, byte_pointer - byte_size)
mbchar = line.byteslice(byte_pointer - byte_size - size, size)
break if mbchar =~ /\s/
width += get_mbchar_width(mbchar)
byte_size += size
end
[byte_size, width]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 482
def self.vi_big_forward_end_word(line, byte_pointer)
if (line.bytesize - 1) > byte_pointer
size = get_next_mbchar_size(line, byte_pointer)
mbchar = line.byteslice(byte_pointer, size)
width = get_mbchar_width(mbchar)
byte_size = size
else
return [0, 0]
end
while (line.bytesize - 1) > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar =~ /\S/
width += get_mbchar_width(mbchar)
byte_size += size
end
prev_width = width
prev_byte_size = byte_size
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar =~ /\s/
prev_width = width
prev_byte_size = byte_size
width += get_mbchar_width(mbchar)
byte_size += size
end
[prev_byte_size, prev_width]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 462
def self.vi_big_forward_word(line, byte_pointer)
width = 0
byte_size = 0
while (line.bytesize - 1) > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar =~ /\s/
width += get_mbchar_width(mbchar)
byte_size += size
end
while (line.bytesize - 1) > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar =~ /\S/
width += get_mbchar_width(mbchar)
byte_size += size
end
[byte_size, width]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 682
def self.vi_first_print(line)
width = 0
byte_size = 0
while (line.bytesize - 1) > byte_size
size = get_next_mbchar_size(line, byte_size)
mbchar = line.byteslice(byte_size, size)
if mbchar =~ /\S/
break
end
width += get_mbchar_width(mbchar)
byte_size += size
end
[byte_size, width]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 573
def self.vi_forward_end_word(line, byte_pointer)
if (line.bytesize - 1) > byte_pointer
size = get_next_mbchar_size(line, byte_pointer)
mbchar = line.byteslice(byte_pointer, size)
if mbchar =~ /\w/
started_by = :word
elsif mbchar =~ /\s/
started_by = :space
else
started_by = :non_word_printable
end
width = get_mbchar_width(mbchar)
byte_size = size
else
return [0, 0]
end
if (line.bytesize - 1) > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
if mbchar =~ /\w/
second = :word
elsif mbchar =~ /\s/
second = :space
else
second = :non_word_printable
end
second_width = get_mbchar_width(mbchar)
second_byte_size = size
else
return [byte_size, width]
end
if second == :space
width += second_width
byte_size += second_byte_size
while (line.bytesize - 1) > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
if mbchar =~ /\S/
if mbchar =~ /\w/
started_by = :word
else
started_by = :non_word_printable
end
break
end
width += get_mbchar_width(mbchar)
byte_size += size
end
else
case [started_by, second]
when [:word, :non_word_printable], [:non_word_printable, :word]
started_by = second
else
width += second_width
byte_size += second_byte_size
started_by = second
end
end
prev_width = width
prev_byte_size = byte_size
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
case started_by
when :word
break if mbchar =~ /\W/
when :non_word_printable
break if mbchar =~ /[\w\s]/
end
prev_width = width
prev_byte_size = byte_size
width += get_mbchar_width(mbchar)
byte_size += size
end
[prev_byte_size, prev_width]
end
No documentation available
lib/reline/unicode.rb
View on GitHub
# File tmp/rubies/ruby-master/lib/reline/unicode.rb, line 532
def self.vi_forward_word(line, byte_pointer, drop_terminate_spaces = false)
if line.bytesize > byte_pointer
size = get_next_mbchar_size(line, byte_pointer)
mbchar = line.byteslice(byte_pointer, size)
if mbchar =~ /\w/
started_by = :word
elsif mbchar =~ /\s/
started_by = :space
else
started_by = :non_word_printable
end
width = get_mbchar_width(mbchar)
byte_size = size
else
return [0, 0]
end
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
case started_by
when :word
break if mbchar =~ /\W/
when :space
break if mbchar =~ /\S/
when :non_word_printable
break if mbchar =~ /\w|\s/
end
width += get_mbchar_width(mbchar)
byte_size += size
end
return [byte_size, width] if drop_terminate_spaces
while line.bytesize > (byte_pointer + byte_size)
size = get_next_mbchar_size(line, byte_pointer + byte_size)
mbchar = line.byteslice(byte_pointer + byte_size, size)
break if mbchar =~ /\S/
width += get_mbchar_width(mbchar)
byte_size += size
end
[byte_size, width]
end
No documentation available