3 # $Release Version: 1.1.2 $
8 # A product of the Austin Ruby Codefest (Austin, Texas, August 2002)
16 scanf for Ruby is an implementation of the C function scanf(3),
17 modified as necessary for Ruby compatibility.
19 The methods provided are String#scanf, IO#scanf, and
20 Kernel#scanf. Kernel#scanf is a wrapper around STDIN.scanf. IO#scanf
21 can be used on any IO stream, including file handles and sockets.
22 scanf can be called either with or without a block.
24 scanf for Ruby scans an input string or stream according to a
25 <b>format</b>, as described below ("Conversions"), and returns an
26 array of matches between the format and the input. The format is
27 defined in a string, and is similar (though not identical) to the
28 formats used in Kernel#printf and Kernel#sprintf.
30 The format may contain <b>conversion specifiers</b>, which tell scanf
31 what form (type) each particular matched substring should be converted
32 to (e.g., decimal integer, floating point number, literal string,
33 etc.) The matches and conversions take place from left to right, and
34 the conversions themselves are returned as an array.
36 The format string may also contain characters other than those in the
37 conversion specifiers. White space (blanks, tabs, or newlines) in the
38 format string matches any amount of white space, including none, in
39 the input. Everything else matches only itself.
41 Scanning stops, and scanf returns, when any input character fails to
42 match the specifications in the format string, or when input is
43 exhausted, or when everything in the format string has been
44 matched. All matches found up to the stopping point are returned in
45 the return array (or yielded to the block, if a block was given).
52 # String#scanf and IO#scanf take a single argument (a format string)
53 array = aString.scanf("%d%s")
54 array = anIO.scanf("%d%s")
56 # Kernel#scanf reads from STDIN
61 When called with a block, scanf keeps scanning the input, cycling back
62 to the beginning of the format string, and yields a new array of
63 conversions to the block every time the format string is matched
64 (including partial matches, but not including complete failures). The
65 actual return value of scanf when called with a block is an array
66 containing the results of all the executions of the block.
68 str = "123 abc 456 def 789 ghi"
69 str.scanf("%d%s") { |num,str| [ num * 2, str.upcase ] }
70 # => [[246, "ABC"], [912, "DEF"], [1578, "GHI"]]
74 The single argument to scanf is a format string, which generally
75 includes one or more conversion specifiers. Conversion specifiers
76 begin with the percent character ('%') and include information about
77 what scanf should next scan for (string, decimal number, single
80 There may be an optional maximum field width, expressed as a decimal
81 integer, between the % and the conversion. If no width is given, a
82 default of `infinity' is used (with the exception of the %c specifier;
83 see below). Otherwise, given a field width of <em>n</em> for a given
84 conversion, at most <em>n</em> characters are scanned in processing
85 that conversion. Before conversion begins, most conversions skip
86 white space in the input string; this white space is not counted
87 against the field width.
89 The following conversions are available. (See the files EXAMPLES
90 and <tt>tests/scanftests.rb</tt> for examples.)
93 Matches a literal `%'. That is, `%%' in the format string matches a
94 single input `%' character. No conversion is done, and the resulting
95 '%' is not included in the return array.
98 Matches an optionally signed decimal integer.
104 Matches an optionally signed integer. The integer is read in base
105 16 if it begins with `0x' or `0X', in base 8 if it begins with `0',
106 and in base 10 other- wise. Only characters that correspond to the
110 Matches an optionally signed octal integer.
113 Matches an optionally signed hexadecimal integer,
116 Matches an optionally signed floating-point number.
119 Matches a sequence of non-white-space character. The input string stops at
120 white space or at the maximum field width, whichever occurs first.
123 Matches a single character, or a sequence of <em>n</em> characters if a
124 field width of <em>n</em> is specified. The usual skip of leading white
125 space is suppressed. To skip white space first, use an explicit space in
129 Matches a nonempty sequence of characters from the specified set
130 of accepted characters. The usual skip of leading white space is
131 suppressed. This bracketed sub-expression is interpreted exactly like a
132 character class in a Ruby regular expression. (In fact, it is placed as-is
133 in a regular expression.) The matching against the input string ends with
134 the appearance of a character not in (or, with a circumflex, in) the set,
135 or when the field width runs out, whichever comes first.
137 ===Assignment suppression
139 To require that a particular match occur, but without including the result
140 in the return array, place the <b>assignment suppression flag</b>, which is
141 the star character ('*'), immediately after the leading '%' of a format
142 specifier (just before the field width, if any).
146 See the files <tt>EXAMPLES</tt> and <tt>tests/scanftests.rb</tt>.
148 ==scanf for Ruby compared with scanf in C
150 scanf for Ruby is based on the C function scanf(3), but with modifications,
151 dictated mainly by the underlying differences between the languages.
153 ===Unimplemented flags and specifiers
155 * The only flag implemented in scanf for Ruby is '<tt>*</tt>' (ignore
156 upcoming conversion). Many of the flags available in C versions of scanf(4)
157 have to do with the type of upcoming pointer arguments, and are literally
160 * The <tt>n</tt> specifier (store number of characters consumed so far in
161 next pointer) is not implemented.
163 * The <tt>p</tt> specifier (match a pointer value) is not implemented.
165 ===Altered specifiers
168 In scanf for Ruby, all of these specifiers scan for an optionally signed
169 integer, rather than for an unsigned integer like their C counterparts.
173 scanf for Ruby returns an array of successful conversions, whereas
174 scanf(3) returns the number of conversions successfully
175 completed. (See below for more details on scanf for Ruby's return
180 Without a block, scanf returns an array containing all the conversions
181 it has found. If none are found, scanf will return an empty array. An
182 unsuccesful match is never ignored, but rather always signals the end
183 of the scanning operation. If the first unsuccessful match takes place
184 after one or more successful matches have already taken place, the
185 returned array will contain the results of those successful matches.
187 With a block scanf returns a 'map'-like array of transformations from
188 the block -- that is, an array reflecting what the block did with each
189 yielded result from the iterative scanf operation. (See "Block
194 scanf for Ruby includes a suite of unit tests (requiring the
195 <tt>TestUnit</tt> package), which can be run with the command <tt>ruby
196 tests/scanftests.rb</tt> or the command <tt>make test</tt>.
198 ==Current limitations and bugs
200 When using IO#scanf under Windows, make sure you open your files in
203 File.open("filename", "rb")
205 so that scanf can keep track of characters correctly.
207 Support for character classes is reasonably complete (since it
208 essentially piggy-backs on Ruby's regular expression handling of
209 character classes), but users are advised that character class testing
210 has not been exhaustive, and that they should exercise some caution
211 in using any of the more complex and/or arcane character class
217 ===Rationale behind scanf for Ruby
219 The impetus for a scanf implementation in Ruby comes chiefly from the fact
220 that existing pattern matching operations, such as Regexp#match and
221 String#scan, return all results as strings, which have to be converted to
222 integers or floats explicitly in cases where what's ultimately wanted are
223 integer or float values.
225 ===Design of scanf for Ruby
227 scanf for Ruby is essentially a <format string>-to-<regular
228 expression> converter.
230 When scanf is called, a FormatString object is generated from the
231 format string ("%d%s...") argument. The FormatString object breaks the
232 format string down into atoms ("%d", "%5f", "blah", etc.), and from
233 each atom it creates a FormatSpecifier object, which it
236 Each FormatSpecifier has a regular expression fragment and a "handler"
237 associated with it. For example, the regular expression fragment
238 associated with the format "%d" is "([-+]?\d+)", and the handler
239 associated with it is a wrapper around String#to_i. scanf itself calls
240 FormatString#match, passing in the input string. FormatString#match
241 iterates through its FormatSpecifiers; for each one, it matches the
242 corresponding regular expression fragment against the string. If
243 there's a match, it sends the matched string to the handler associated
244 with the FormatSpecifier.
246 Thus, to follow up the "%d" example: if "123" occurs in the input
247 string when a FormatSpecifier consisting of "%d" is reached, the "123"
248 will be matched against "([-+]?\d+)", and the matched string will be
249 rendered into an integer by a call to to_i.
251 The rendered match is then saved to an accumulator array, and the
252 input string is reduced to the post-match substring. Thus the string
253 is "eaten" from the left as the FormatSpecifiers are applied in
254 sequence. (This is done to a duplicate string; the original string is
257 As soon as a regular expression fragment fails to match the string, or
258 when the FormatString object runs out of FormatSpecifiers, scanning
259 stops and results accumulated so far are returned in an array.
261 ==License and copyright
263 Copyright:: (c) 2002-2003 David Alan Black
264 License:: Distributed on the same licensing terms as Ruby itself
266 ==Warranty disclaimer
268 This software is provided "as is" and without any express or implied
269 warranties, including, without limitation, the implied warranties of
270 merchantibility and fitness for a particular purpose.
272 ==Credits and acknowledgements
274 scanf for Ruby was developed as the major activity of the Austin
275 Ruby Codefest (Austin, Texas, August 2002).
277 Principal author:: David Alan Black (mailto:dblack@superlink.net)
278 Co-author:: Hal Fulton (mailto:hal9000@hypermetrics.com)
279 Project contributors:: Nolan Darilek, Jason Johnston
281 Thanks to Hal Fulton for hosting the Codefest.
283 Thanks to Matz for suggestions about the class design.
285 Thanks to Gavin Sinclair for some feedback on the documentation.
287 The text for parts of this document, especially the Description and
288 Conversions sections, above, were adapted from the Linux Programmer's
289 Manual manpage for scanf(3), dated 1995-11-01.
291 ==Bugs and bug reports
293 scanf for Ruby is based on something of an amalgam of C scanf
294 implementations and documentation, rather than on a single canonical
295 description. Suggestions for features and behaviors which appear in
296 other scanfs, and would be meaningful in Ruby, are welcome, as are
297 reports of suspicious behaviors and/or bugs. (Please see "Credits and
298 acknowledgements", above, for email addresses.)
304 class FormatSpecifier
306 attr_reader :re_string, :matched_string, :conversion, :matched
310 def skip; /^\s*%\*/.match(@spec_string); end
312 def extract_float(s); s.to_f if s &&! skip; end
313 def extract_decimal(s); s.to_i if s &&! skip; end
314 def extract_hex(s); s.hex if s &&! skip; end
315 def extract_octal(s); s.oct if s &&! skip; end
316 def extract_integer(s); Integer(s) if s &&! skip; end
317 def extract_plain(s); s unless skip; end
319 def nil_proc(s); nil; end
328 /(?:\A|\S)%\*?\d*c|\[/.match(@spec_string)
335 @re_string, @handler =
339 when /%\*?(\[\[:[a-z]+:\]\])/
340 [ "(#{$1}+)", :extract_plain ]
343 when /%\*?(\d+)(\[\[:[a-z]+:\]\])/
344 [ "(#{$2}{1,#{$1}})", :extract_plain ]
347 when /%\*?\[([^\]]*)\]/
349 if /^\^/.match(yes) then no = yes[1..-1] else no = '^' + yes end
350 [ "([#{yes}]+)(?=[#{no}]|\\z)", :extract_plain ]
353 when /%\*?(\d+)\[([^\]]*)\]/
356 [ "([#{yes}]{1,#{w}})", :extract_plain ]
360 [ "([-+]?(?:(?:0[0-7]+)|(?:0[Xx]#{h}+)|(?:[1-9]\\d+)))", :extract_integer ]
366 if n > 1 then s += "[1-9]\\d{1,#{n-1}}|" end
367 if n > 1 then s += "0[0-7]{1,#{n-1}}|" end
368 if n > 2 then s += "[-+]0[0-7]{1,#{n-2}}|" end
369 if n > 2 then s += "[-+][1-9]\\d{1,#{n-2}}|" end
370 if n > 2 then s += "0[Xx]#{h}{1,#{n-2}}|" end
371 if n > 3 then s += "[-+]0[Xx]#{h}{1,#{n-3}}|" end
374 [ s, :extract_integer ]
378 [ '([-+]?\d+)', :extract_decimal ]
384 if n > 1 then s += "[-+]\\d{1,#{n-1}}|" end
386 [ s, :extract_decimal ]
390 [ "([-+]?(?:0[Xx])?#{h}+)", :extract_hex ]
396 if n > 3 then s += "[-+]0[Xx]#{h}{1,#{n-3}}|" end
397 if n > 2 then s += "0[Xx]#{h}{1,#{n-2}}|" end
398 if n > 1 then s += "[-+]#{h}{1,#{n-1}}|" end
405 [ '([-+]?[0-7]+)', :extract_octal ]
409 [ "([-+][0-7]{1,#{$1.to_i-1}}|[0-7]{1,#{$1}})", :extract_octal ]
413 [ '([-+]?((\d+(?>(?=[^\d.]|$)))|(\d*(\.(\d*([eE][-+]?\d+)?)))))', :extract_float ]
417 [ "(\\S{1,#{$1}})", :extract_float ]
421 [ "(\\S{1,#{$1}})", :extract_plain ]
425 [ '(\S+)', :extract_plain ]
429 [ "\\s*(.)", :extract_plain ]
433 [ "(.)", :extract_plain ]
435 # %5c (whitespace issues are handled by the count_*_space? methods)
437 [ "(.{1,#{$1}})", :extract_plain ]
441 [ '(\s*%)', :nil_proc ]
445 [ "(#{Regexp.escape(@spec_string)})", :nil_proc ]
448 @re_string = '\A' + @re_string
452 Regexp.new(@re_string,Regexp::MULTILINE)
458 s.sub!(/\A\s+/,'') unless count_space?
461 @conversion = send(@handler, res[1])
462 @matched_string = @conversion.to_s
469 @spec_string[/%\*?\d*([a-z\[])/, 1]
473 w = @spec_string[/%\*?(\d+)/, 1]
478 return false unless @matched
479 cc_no_width = letter == '[' &&! width
480 c_or_cc_width = (letter == 'c' || letter == '[') && width
481 width_left = c_or_cc_width && (matched_string.size < width)
483 return width_left || cc_no_width
490 attr_reader :string_left, :last_spec_tried,
491 :last_match_tried, :matched_count, :space
493 SPECIFIERS = 'diuXxofeEgsc'
495 # possible space, followed by...
497 # percent sign, followed by...
499 # another percent sign, or...
501 # optional assignment suppression flag
503 # optional maximum field width
505 # named character class, ...
507 # traditional character class, or...
511 # or miscellaneous characters
518 return unless /\S/.match(s)
519 @space = true if /\s\z/.match(s)
520 @specs.replace s.scan(REGEX).map {|spec| FormatSpecifier.new(spec) }
527 def prune(n=matched_count)
528 n.times { @specs.shift }
544 @specs.each_with_index do |spec,i|
546 @last_spec_tried = spec
547 @last_match_tried = spec.match(@string_left)
548 break unless @last_match_tried
551 accum << spec.conversion
553 @string_left = @last_match_tried.post_match
554 break if @string_left.empty?
563 # The trick here is doing a match where you grab one *line*
564 # of input at a time. The linebreak may or may not occur
565 # at the boundary where the string matches a format specifier.
566 # And if it does, some rule about whitespace may or may not
569 # That's why this is much more elaborate than the string
573 # Match succeeds (non-emptily)
574 # and the last attempted spec/string sub-match succeeded:
576 # could the last spec keep matching?
577 # yes: save interim results and continue (next line)
579 # The last attempted spec/string did not match:
581 # are we on the next-to-last spec in the string?
583 # is fmt_string.string_left all spaces?
584 # yes: does current spec care about input space?
586 # no: save interim results and continue
587 # no: continue [this state could be analyzed further]
592 return block_scanf(str,&b) if b
593 return [] unless str.size > 0
595 start_position = pos rescue 0
601 fstr = Scanf::FormatString.new(str)
604 if eof || (tty? &&! fstr.match(source_buffer))
605 final_result.concat(result_buffer)
609 source_buffer << gets
611 current_match = fstr.match(source_buffer)
613 spec = fstr.last_spec_tried
617 result_buffer.replace(current_match)
621 elsif (fstr.matched_count == fstr.spec_count - 1)
622 if /\A\s*\z/.match(fstr.string_left)
623 break if spec.count_space?
624 result_buffer.replace(current_match)
629 final_result.concat(current_match)
631 matched_so_far += source_buffer.size
632 source_buffer.replace(fstr.string_left)
633 matched_so_far -= source_buffer.size
634 break if fstr.last_spec
637 seek(start_position + matched_so_far, IO::SEEK_SET) rescue Errno::ESPIPE
638 soak_up_spaces if fstr.last_spec && fstr.space
648 until eof ||! c || /\S/.match(c.chr)
651 ungetc(c) if (c && /\S/.match(c.chr))
656 # Sub-ideal, since another FS gets created in scanf.
657 # But used here to determine the number of specifiers.
658 fstr = Scanf::FormatString.new(str)
659 last_spec = fstr.last_spec
662 break if current.empty?
663 final.push(yield(current))
664 end until eof || fstr.last_spec_tried == last_spec
676 if fstr.is_a? Scanf::FormatString
679 Scanf::FormatString.new(fstr)
685 def block_scanf(fstr,&b)
686 fs = Scanf::FormatString.new(fstr)
690 current = str.scanf(fs)
691 final.push(yield(current)) unless current.empty?
693 end until current.empty? || str.empty?