]>
Commit | Line | Data |
---|---|---|
1 | #!/usr/bin/env perl | |
2 | ############################################################################# | |
3 | # Name: regex.pl | |
4 | # Purpose: Generate test code for wxRegEx from 'reg.test' | |
5 | # Author: Mike Wetherell | |
6 | # RCS-ID: $Id$ | |
7 | # Copyright: (c) Mike Wetherell | |
8 | # Licence: wxWidgets licence | |
9 | ############################################################################# | |
10 | ||
11 | # | |
12 | # Notes: | |
13 | # See './regex.pl -h' for usage | |
14 | # | |
15 | # Output at the moment is C++ using the cppunit testing framework. The | |
16 | # language/framework specifics are separated, with the following 5 | |
17 | # subs as an interface: 'begin_output', 'begin_section', 'write_test', | |
18 | # 'end_section' and 'end_output'. So for a different language/framework, | |
19 | # implement 5 new similar subs. | |
20 | # | |
21 | # I've avoided using 'use encoding "UTF-8"', since this wasn't available | |
22 | # in perl 5.6.x. Instead I've used some hacks like 'pack "U0C*"'. Versions | |
23 | # earler than perl 5.6.0 aren't going to work. | |
24 | # | |
25 | ||
26 | use strict; | |
27 | use warnings; | |
28 | use File::Basename; | |
29 | #use encoding "UTF-8"; # enable in the future when perl 5.6.x is just a memory | |
30 | ||
31 | # if 0 output is wide characters, if 1 output is utf8 encoded | |
32 | my $utf = 1; | |
33 | ||
34 | # quote a parameter (C++ helper) | |
35 | # | |
36 | sub quotecxx { | |
37 | my %esc = ( "\a" => "a", "\b" => "b", "\f" => "f", | |
38 | "\n" => "n", "\r" => "r", "\t" => "t", | |
39 | "\013" => "v", '"' => '"', "\\" => "\\" ); | |
40 | ||
41 | # working around lack of 'use encoding' | |
42 | if (!$utf) { | |
43 | $_ = pack "U0C*", unpack "C*", $_; | |
44 | use utf8; | |
45 | } | |
46 | ||
47 | s/[\000-\037"\\\177-\x{ffff}]/ | |
48 | if ($esc{$&}) { | |
49 | "\\$esc{$&}"; | |
50 | } elsif (ord($&) > 0x9f && !$utf) { | |
51 | sprintf "\\u%04x", ord($&); | |
52 | } else { | |
53 | sprintf "\\%03o", ord($&); | |
54 | } | |
55 | /ge; | |
56 | ||
57 | # working around lack of 'use encoding' | |
58 | if (!$utf) { | |
59 | no utf8; | |
60 | $_ = pack "C*", unpack "C*", $_; | |
61 | } | |
62 | ||
63 | return ($utf ? '"' : 'L"') . $_ . '"' | |
64 | } | |
65 | ||
66 | # start writing the output code (C++ interface) | |
67 | # | |
68 | sub begin_output { | |
69 | my ($from, $instructions) = @_; | |
70 | ||
71 | # embed it in the comment | |
72 | $from = "\n$from"; | |
73 | $from =~ s/^(?: )?/ * /mg; | |
74 | ||
75 | # $instructions contains information about the flags etc. | |
76 | if ($instructions) { | |
77 | $instructions = "\n$instructions"; | |
78 | $instructions =~ s/^(?: )?/ * /mg; | |
79 | } | |
80 | ||
81 | my $u = $utf ? " (UTF-8 encoded)" : ""; | |
82 | ||
83 | print <<EOT; | |
84 | /* | |
85 | * Test data for wxRegEx$u | |
86 | $from$instructions */ | |
87 | ||
88 | EOT | |
89 | } | |
90 | ||
91 | my @classes; | |
92 | ||
93 | # start a new section (C++ interface) | |
94 | # | |
95 | sub begin_section { | |
96 | my ($id, $title) = @_; | |
97 | my $class = "regextest_$id"; | |
98 | $class =~ s/\W/_/g; | |
99 | push @classes, [$id, $class]; | |
100 | ||
101 | print <<EOT; | |
102 | ||
103 | /* | |
104 | * $id $title | |
105 | */ | |
106 | ||
107 | class $class : public RegExTestSuite | |
108 | { | |
109 | public: | |
110 | $class() : RegExTestSuite("regex.$id") { } | |
111 | static Test *suite(); | |
112 | }; | |
113 | ||
114 | Test *$class\::suite() | |
115 | { | |
116 | RegExTestSuite *suite = new $class; | |
117 | ||
118 | EOT | |
119 | } | |
120 | ||
121 | # output a test line (C++ interface) | |
122 | # | |
123 | sub write_test { | |
124 | my @args = @_; | |
125 | $_ = quotecxx for @args; | |
126 | print " suite->add(" . (join ', ', @args) . ", NULL);\n"; | |
127 | } | |
128 | ||
129 | # end a section (C++ interface) | |
130 | # | |
131 | sub end_section { | |
132 | my ($id, $class) = @{$classes[$#classes]}; | |
133 | ||
134 | print <<EOT; | |
135 | ||
136 | return suite; | |
137 | } | |
138 | ||
139 | CPPUNIT_TEST_SUITE_NAMED_REGISTRATION($class, "regex.$id"); | |
140 | ||
141 | EOT | |
142 | } | |
143 | ||
144 | # finish off the output (C++ interface) | |
145 | # | |
146 | sub end_output { | |
147 | print <<EOT; | |
148 | ||
149 | /* | |
150 | * A suite containing all the above suites | |
151 | */ | |
152 | ||
153 | class regextest : public TestSuite | |
154 | { | |
155 | public: | |
156 | regextest() : TestSuite("regex") { } | |
157 | static Test *suite(); | |
158 | }; | |
159 | ||
160 | Test *regextest::suite() | |
161 | { | |
162 | TestSuite *suite = new regextest; | |
163 | ||
164 | EOT | |
165 | print " suite->addTest(".$_->[1]."::suite());\n" for @classes; | |
166 | ||
167 | print <<EOT; | |
168 | ||
169 | return suite; | |
170 | } | |
171 | ||
172 | CPPUNIT_TEST_SUITE_NAMED_REGISTRATION(regextest, "regex"); | |
173 | CPPUNIT_TEST_SUITE_REGISTRATION(regextest); | |
174 | EOT | |
175 | } | |
176 | ||
177 | # Parse a tcl string. Handles curly quoting and double quoting. | |
178 | # | |
179 | sub parsetcl { | |
180 | my ($curly, $quote); | |
181 | # recursively defined expression that can parse balanced braces | |
182 | # warning: uses experimental features of perl, see perlop(1) | |
183 | $curly = qr/\{(?:(?>(?:\\[{}]|[^{}])+)|(??{$curly}))*\}/; | |
184 | $quote = qr/"(?:\\"|[^"])*"/; | |
185 | my @tokens = shift =~ /($curly|$quote|\S+)/g; | |
186 | ||
187 | # now remove braces/quotes and unescape any escapes | |
188 | for (@tokens) { | |
189 | if (s/^{(.*)}$/$1/) { | |
190 | # for curly quoting, only unescape \{ and \} | |
191 | s/\\([{}])/$1/g; | |
192 | } else { | |
193 | s/^"(.*)"$/$1/; | |
194 | ||
195 | # unescape any escapes | |
196 | my %esc = ( "a" => "\a", "b" => "\b", "f" => "\f", | |
197 | "n" => "\n", "r" => "\r", "t" => "\t", | |
198 | "v" => "\013" ); | |
199 | my $x = qr/[[:xdigit:]]/; | |
200 | ||
201 | s/\\([0-7]{1,3}|x$x+|u$x{1,4}|.)/ | |
202 | if ($1 =~ m{^([0-7]+)}) { | |
203 | chr(oct($1)); | |
204 | } elsif ($1 =~ m{^x($x+)}) { | |
205 | pack("C0U", hex($1) & 0xff); | |
206 | } elsif ($1 =~ m{^u($x+)}) { | |
207 | pack("C0U", hex($1)); | |
208 | } elsif ($esc{$1}) { | |
209 | $esc{$1}; | |
210 | } else { | |
211 | $1; | |
212 | } | |
213 | /ge; | |
214 | } | |
215 | } | |
216 | ||
217 | return @tokens; | |
218 | } | |
219 | ||
220 | # helpers which keep track of whether begin_section has been called, so that | |
221 | # end_section can be called when appropriate | |
222 | # | |
223 | my @doing = ("0", ""); | |
224 | my $in_section = 0; | |
225 | ||
226 | sub handle_doing { | |
227 | end_section if $in_section; | |
228 | $in_section = 0; | |
229 | @doing = @_; | |
230 | } | |
231 | ||
232 | sub handle_test { | |
233 | begin_section(@doing) if !$in_section; | |
234 | $in_section = 1; | |
235 | write_test @_; | |
236 | } | |
237 | ||
238 | sub handle_end { | |
239 | end_section if $in_section; | |
240 | $in_section = 0; | |
241 | end_output; | |
242 | } | |
243 | ||
244 | # 'main' - start by parsing the command lines options. | |
245 | # | |
246 | my $badoption = !@ARGV; | |
247 | my $utfdefault = $utf; | |
248 | my $outputname; | |
249 | ||
250 | for (my $i = 0; $i < @ARGV; ) { | |
251 | if ($ARGV[$i] !~ m{^-.}) { | |
252 | $i++; | |
253 | next; | |
254 | } | |
255 | ||
256 | if ($ARGV[$i] eq '--') { | |
257 | splice @ARGV, $i, 1; | |
258 | last; | |
259 | } | |
260 | ||
261 | if ($ARGV[$i] =~ s{^-(.*)o(.*)$}{-$1}i) { # -o : output file | |
262 | $outputname = $2 || splice @ARGV, $i + 1, 1; | |
263 | } | |
264 | ||
265 | for (split //, substr($ARGV[$i], 1)) { | |
266 | if (/u/i) { # -u : utf-8 output | |
267 | $utf = 1; | |
268 | } elsif (/w/i) { # -w : wide char output | |
269 | $utf = 0; | |
270 | } else { | |
271 | $badoption = 1; | |
272 | } | |
273 | } | |
274 | ||
275 | splice @ARGV, $i, 1; | |
276 | } | |
277 | ||
278 | # Display help | |
279 | # | |
280 | if ($badoption) { | |
281 | my $prog = basename $0; | |
282 | my ($w, $u) = (" (default)", " "); | |
283 | ($w, $u) = ($u, $w) if $utfdefault; | |
284 | ||
285 | print <<EOT; | |
286 | Usage: $prog [-u|-w] [-o OUTPUT] [FILE...] | |
287 | Generate test code for wxRegEx from 'reg.test' | |
288 | Example: $prog -o regex.inc reg.test wxreg.test | |
289 | ||
290 | -w$w Output will be wide characters. | |
291 | -u$u Output will be UTF-8 encoded. | |
292 | ||
293 | Input files should be in UTF-8. If no input files are specified input is | |
294 | read from stdin. If no output file is specified output is written to stdout. | |
295 | See the comments in reg.test for details of the input file format. | |
296 | EOT | |
297 | exit 0; | |
298 | } | |
299 | ||
300 | # Open the output file | |
301 | # | |
302 | open STDOUT, ">$outputname" if $outputname; | |
303 | ||
304 | # Read in the files and initially parse just the comments for copyright | |
305 | # information and instructions on the tests | |
306 | # | |
307 | my @input; # slurped input files stripped of comments | |
308 | my $files = ""; # copyright info from the input comments | |
309 | my $instructions = ""; # test instructions from the input comments | |
310 | ||
311 | do { | |
312 | my $inputname = basename $ARGV[0] if @ARGV; | |
313 | ||
314 | # slurp input | |
315 | undef $/; | |
316 | my $in = <>; | |
317 | ||
318 | # remove escaped newlines | |
319 | $in =~ s/(?<!\\)\\\n//g; | |
320 | ||
321 | # record the copyrights of the input files | |
322 | for ($in =~ /^#[\t ]*(.*copyright.*)$/mig) { | |
323 | s/[\s:]+/ /g; | |
324 | $files .= " "; | |
325 | $files .= $inputname . ": " if $inputname && $inputname ne '-'; | |
326 | $files .= "$_\n"; | |
327 | } | |
328 | ||
329 | # Parse the comments for instructions on the tests, which look like this: | |
330 | # i successful match with -indices (used in checking things like | |
331 | # nonparticipating subexpressions) | |
332 | if (!$instructions) { | |
333 | my $sp = qr{\t| +}; # tab or three or more spaces | |
334 | my @instructions = $in =~ | |
335 | /\n( | |
336 | (?: | |
337 | \#$sp\S?$sp\S[^\n]+\n # instruction line | |
338 | (?:\#$sp$sp\S[^\n]+\n)* # continuation lines (if any) | |
339 | )+ | |
340 | )/gx; | |
341 | ||
342 | if (@instructions) { | |
343 | $instructions[0] = "Test types:\n$instructions[0]"; | |
344 | if (@instructions > 1) { | |
345 | $instructions[1] = "Flag characters:\n$instructions[1]"; | |
346 | } | |
347 | $instructions = join "\n", @instructions; | |
348 | $instructions =~ s/^#([^\t]?)/ $1/mg; | |
349 | } | |
350 | } | |
351 | ||
352 | # @input is the input of all files (stipped of comments) | |
353 | $in =~ s/^#.*$//mg; | |
354 | push @input, $in; | |
355 | ||
356 | } while $ARGV[0]; | |
357 | ||
358 | # Make a string naming the generator, the input files and copyright info | |
359 | # | |
360 | my $from = "Generated " . localtime() . " by " . basename $0; | |
361 | $from =~ s/[\s]+/ /g; | |
362 | if ($files) { | |
363 | if ($files =~ /:/) { | |
364 | $from .= " from the following files:"; | |
365 | } else { | |
366 | $from .= " from work with the following copyright:"; | |
367 | } | |
368 | } | |
369 | $from = join("\n", $from =~ /(.{0,76}(?:\s|$))/g); # word-wrap | |
370 | $from .= "\n$files" if $files; | |
371 | ||
372 | # Now start to print the code | |
373 | # | |
374 | begin_output $from, $instructions; | |
375 | ||
376 | # numbers for 'extra' sections | |
377 | my $extra = 1; | |
378 | ||
379 | for (@input) | |
380 | { | |
381 | # Print the main tests | |
382 | # | |
383 | # Test lines look like this: | |
384 | # m 3 b {\(a\)b} ab ab a | |
385 | # | |
386 | # Also looks for heading lines, e.g.: | |
387 | # doing 4 "parentheses" | |
388 | # | |
389 | for (split "\n") { | |
390 | if (/^doing\s+(\S+)\s+(\S.*)/) { | |
391 | handle_doing parsetcl "$1 $2"; | |
392 | } elsif (/^[efimp]\s/) { | |
393 | handle_test parsetcl $_; | |
394 | } | |
395 | } | |
396 | ||
397 | # Extra tests | |
398 | # | |
399 | # The expression below matches something like this: | |
400 | # test reg-33.8 {Bug 505048} { | |
401 | # regexp -inline {\A\s*[^b]*b} ab | |
402 | # } ab | |
403 | # | |
404 | # The three subexpressions then return these parts: | |
405 | # $extras[$i] = '{Bug 505048}', | |
406 | # $extras[$i + 1] = '-inline {\A\s*[^b]*b} ab' | |
407 | # $extras[$i + 2] = 'ab' | |
408 | # | |
409 | my @extras = /\ntest\s+\S+\s*(\{.*?\})\s*\{\n # line 1 | |
410 | \s*regexp\s+([^\n]+)\n # line 2 | |
411 | \}\s*(\S[^\n]*)/gx; # line 3 | |
412 | ||
413 | handle_doing "extra_" . $extra++, "checks for bug fixes" if @extras; | |
414 | ||
415 | for (my $i = 0; $i < @extras; $i += 3) { | |
416 | my $id = $extras[$i]; | |
417 | ||
418 | # further parse the middle line into options and the rest (i.e. $args) | |
419 | my ($opts, $args) = $extras[$i + 1] =~ /^\s*((?:-\S+\s+)*)([^\s-].*)/; | |
420 | ||
421 | my @args = parsetcl $args; | |
422 | $#args = 1; # only want the first two | |
423 | ||
424 | # now handle the options | |
425 | my $test = $opts =~ /-indices/ ? 'i' : $extras[$i + 2] ? 'm' : 'f'; | |
426 | my $results = $opts =~ /-inline/ && $test ne 'f' ? $extras[$i+2] : ''; | |
427 | ||
428 | # get them all in the right order and print | |
429 | unshift @args, $test, parsetcl($id), $results ? '-' : 'o'; | |
430 | push @args, parsetcl(parsetcl($results)) if $results; | |
431 | handle_test @args; | |
432 | } | |
433 | } | |
434 | ||
435 | # finish | |
436 | # | |
437 | handle_end; |