VirtualBox

source: vbox/trunk/src/libs/libxml2-2.13.2/genChRanges.py@ 106165

最後變更 在這個檔案從106165是 105420,由 vboxsync 提交於 4 月 前

libxml2-2.12.6: Applied and adjusted our libxml2 changes to 2.12.6. bugref:10730

  • 屬性 svn:eol-style 設為 native
  • 屬性 svn:executable 設為 *
檔案大小: 17.7 KB
 
1#!/usr/bin/env python3
2#
3# Portions of this script have been (shamelessly) stolen from the
4# prior work of Daniel Veillard (genUnicode.py)
5#
6# I, however, take full credit for any bugs, errors or difficulties :-)
7#
8# William Brack
9# October 2003
10#
11# 18 October 2003
12# Modified to maintain binary compatibility with previous library versions
13# by adding a suffix 'Q' ('quick') to the macro generated for the original,
14# function, and adding generation of a function (with the original name) which
15# instantiates the macro.
16#
17
18import sys
19import time
20
21#
22# A routine to take a list of yes/no (1, 0) values and turn it
23# into a list of ranges. This will later be used to determine whether
24# to generate single-byte lookup tables, or inline comparisons
25#
26def makeRange(lst):
27 ret = []
28 pos = 0
29 while pos < len(lst):
30 try: # index generates exception if not present
31 s = lst[pos:].index(1) # look for start of next range
32 except:
33 break # if no more, finished
34 pos += s # pointer to start of possible range
35 try:
36 e = lst[pos:].index(0) # look for end of range
37 e += pos
38 except: # if no end, set to end of list
39 e = len(lst)
40 ret.append((pos, e-1)) # append range tuple to list
41 pos = e + 1 # ready to check for next range
42 return ret
43
44sources = "chvalid.def" # input filename
45
46# minTableSize gives the minimum number of ranges which must be present
47# before a 256-byte lookup table is produced. If there are less than this
48# number, a macro with inline comparisons is generated
49minTableSize = 6
50
51# dictionary of functions, key=name, element contains char-map and range-list
52Functs = {}
53
54state = 0
55
56try:
57 defines = open("chvalid.def", "r")
58except:
59 print("Missing chvalid.def, aborting ...")
60 sys.exit(1)
61
62#
63# The lines in the .def file have three types:-
64# name: Defines a new function block
65# ur: Defines individual or ranges of unicode values
66# end: Indicates the end of the function block
67#
68# These lines are processed below.
69#
70for line in defines.readlines():
71 # ignore blank lines, or lines beginning with '#'
72 if line[0] == '#':
73 continue
74 line = line.strip()
75 if line == '':
76 continue
77 # split line into space-separated fields, then split on type
78 try:
79 fields = line.split(' ')
80 #
81 # name line:
82 # validate any previous function block already ended
83 # validate this function not already defined
84 # initialize an entry in the function dicitonary
85 # including a mask table with no values yet defined
86 #
87 if fields[0] == 'name':
88 name = fields[1]
89 if state != 0:
90 print("'name' %s found before previous name" \
91 "completed" % (fields[1]))
92 continue
93 state = 1
94 if name in Functs:
95 print("name '%s' already present - may give" \
96 " wrong results" % (name))
97 else:
98 # dict entry with two list elements (chdata, rangedata)
99 Functs[name] = [ [], [] ]
100 for v in range(256):
101 Functs[name][0].append(0)
102 #
103 # end line:
104 # validate there was a preceding function name line
105 # set state to show no current function active
106 #
107 elif fields[0] == 'end':
108 if state == 0:
109 print("'end' found outside of function block")
110 continue
111 state = 0
112
113 #
114 # ur line:
115 # validate function has been defined
116 # process remaining fields on the line, which may be either
117 # individual unicode values or ranges of values
118 #
119 elif fields[0] == 'ur':
120 if state != 1:
121 raise Exception("'ur' found outside of 'name' block")
122 for el in fields[1:]:
123 pos = el.find('..')
124 # pos <=0 means not a range, so must be individual value
125 if pos <= 0:
126 # cheap handling of hex or decimal values
127 if el[0:2] == '0x':
128 value = int(el[2:],16)
129 elif el[0] == "'":
130 value = ord(el[1])
131 else:
132 value = int(el)
133 if ((value < 0) | (value > 0x1fffff)):
134 raise Exception('Illegal value (%s) in ch for'\
135 ' name %s' % (el,name))
136 # for ur we have only ranges (makes things simpler),
137 # so convert val to range
138 currange = (value, value)
139 # pos > 0 means this is a range, so isolate/validate
140 # the interval
141 else:
142 # split the range into it's first-val, last-val
143 (first, last) = el.split("..")
144 # convert values from text into binary
145 if first[0:2] == '0x':
146 start = int(first[2:],16)
147 elif first[0] == "'":
148 start = ord(first[1])
149 else:
150 start = int(first)
151 if last[0:2] == '0x':
152 end = int(last[2:],16)
153 elif last[0] == "'":
154 end = ord(last[1])
155 else:
156 end = int(last)
157 if (start < 0) | (end > 0x1fffff) | (start > end):
158 raise Exception("Invalid range '%s'" % el)
159 currange = (start, end)
160 # common path - 'currange' has the range, now take care of it
161 # We split on single-byte values vs. multibyte
162 if currange[1] < 0x100: # single-byte
163 for ch in range(currange[0],currange[1]+1):
164 # validate that value not previously defined
165 if Functs[name][0][ch]:
166 msg = "Duplicate ch value '%s' for name '%s'" % (el, name)
167 raise Exception(msg)
168 Functs[name][0][ch] = 1
169 else: # multi-byte
170 if currange in Functs[name][1]:
171 raise Exception("range already defined in" \
172 " function")
173 else:
174 Functs[name][1].append(currange)
175
176 except:
177 print("Failed to process line: %s" % (line))
178 raise
179#
180# At this point, the entire definition file has been processed. Now we
181# enter the output phase, where we generate the two files chvalid.c and'
182# chvalid.h
183#
184# To do this, we first output the 'static' data (heading, fixed
185# definitions, etc.), then output the 'dynamic' data (the results
186# of the above processing), and finally output closing 'static' data
187# (e.g. the subroutine to process the ranges)
188#
189
190#
191# Generate the headings:
192#
193try:
194 header = open("include/libxml/chvalid.h", "w")
195except:
196 print("Failed to open include/libxml/chvalid.h")
197 sys.exit(1)
198
199try:
200 output = open("chvalid.c", "w")
201except:
202 print("Failed to open chvalid.c")
203 sys.exit(1)
204
205date = time.asctime(time.localtime(time.time()))
206
207header.write(
208"""/*
209 * Summary: Unicode character range checking
210 * Description: this module exports interfaces for the character
211 * range validation APIs
212 *
213 * This file is automatically generated from the cvs source
214 * definition files using the genChRanges.py Python script
215 *
216 * Generation date: %s
217 * Sources: %s
218 * Author: William Brack <[email protected]>
219 */
220
221#ifndef __XML_CHVALID_H__
222#define __XML_CHVALID_H__
223
224#include <libxml/xmlversion.h>
225#include <libxml/xmlstring.h>
226
227#ifdef __cplusplus
228extern "C" {
229#endif
230
231/*
232 * Define our typedefs and structures
233 *
234 */
235typedef struct _xmlChSRange xmlChSRange;
236typedef xmlChSRange *xmlChSRangePtr;
237struct _xmlChSRange {
238 unsigned short\tlow;
239 unsigned short\thigh;
240};
241
242typedef struct _xmlChLRange xmlChLRange;
243typedef xmlChLRange *xmlChLRangePtr;
244struct _xmlChLRange {
245 unsigned int\tlow;
246 unsigned int\thigh;
247};
248
249typedef struct _xmlChRangeGroup xmlChRangeGroup;
250typedef xmlChRangeGroup *xmlChRangeGroupPtr;
251struct _xmlChRangeGroup {
252 int\t\t\tnbShortRange;
253 int\t\t\tnbLongRange;
254 const xmlChSRange\t*shortRange;\t/* points to an array of ranges */
255 const xmlChLRange\t*longRange;
256};
257
258/**
259 * Range checking routine
260 */
261XMLPUBFUN int
262\t\txmlCharInRange(unsigned int val, const xmlChRangeGroup *group);
263
264""" % (date, sources));
265output.write(
266"""/*
267 * chvalid.c:\tthis module implements the character range
268 *\t\tvalidation APIs
269 *
270 * This file is automatically generated from the cvs source
271 * definition files using the genChRanges.py Python script
272 *
273 * Generation date: %s
274 * Sources: %s
275 * William Brack <[email protected]>
276 */
277
278#define IN_LIBXML
279#include "libxml.h"
280#include <libxml/chvalid.h>
281
282#include <stddef.h>
283
284/*
285 * The initial tables ({func_name}_tab) are used to validate whether a
286 * single-byte character is within the specified group. Each table
287 * contains 256 bytes, with each byte representing one of the 256
288 * possible characters. If the table byte is set, the character is
289 * allowed.
290 *
291 */
292""" % (date, sources));
293
294#
295# Now output the generated data.
296# We try to produce the best execution times. Tests have shown that validation
297# with direct table lookup is, when there are a "small" number of valid items,
298# still not as fast as a sequence of inline compares. So, if the single-byte
299# portion of a range has a "small" number of ranges, we output a macro for inline
300# compares, otherwise we output a 256-byte table and a macro to use it.
301#
302
303fkeys = sorted(Functs.keys())
304
305for f in fkeys:
306
307# First we convert the specified single-byte values into a group of ranges.
308# If the total number of such ranges is less than minTableSize, we generate
309# an inline macro for direct comparisons; if greater, we generate a lookup
310# table.
311 if max(Functs[f][0]) > 0: # only check if at least one entry
312 rangeTable = makeRange(Functs[f][0])
313 numRanges = len(rangeTable)
314 if numRanges >= minTableSize: # table is worthwhile
315 header.write("XMLPUBVAR const unsigned char %s_tab[256];\n" % f)
316 header.write("""
317/**
318 * %s_ch:
319 * @c: char to validate
320 *
321 * Automatically generated by genChRanges.py
322 */
323""" % f)
324 header.write("#define %s_ch(c)\t(%s_tab[(c)])\n" % (f, f))
325
326 # write the constant data to the code file
327 output.write("const unsigned char %s_tab[256] = {\n" % f)
328 pline = " "
329 for n in range(255):
330 pline += " 0x%02x," % Functs[f][0][n]
331 if len(pline) > 72:
332 output.write(pline + "\n")
333 pline = " "
334 output.write(pline + " 0x%02x };\n\n" % Functs[f][0][255])
335
336 else: # inline check is used
337 # first another little optimisation - if space is present,
338 # put it at the front of the list so it is checked first
339 try:
340 ix = rangeTable.remove((0x20, 0x20))
341 rangeTable.insert(0, (0x20, 0x20))
342 except:
343 pass
344 firstFlag = 1
345
346 header.write("""
347/**
348 * %s_ch:
349 * @c: char to validate
350 *
351 * Automatically generated by genChRanges.py
352 */
353""" % f)
354 # okay, I'm tired of the messy lineup - let's automate it!
355 pline = "#define %s_ch(c)" % f
356 # 'ntab' is number of tabs needed to position to col. 33 from name end
357 ntab = 4 - (len(pline)) // 8
358 if ntab < 0:
359 ntab = 0
360 just = ""
361 for i in range(ntab):
362 just += "\t"
363 pline = pline + just + "("
364 for rg in rangeTable:
365 if not firstFlag:
366 pline += " || \\\n\t\t\t\t "
367 else:
368 firstFlag = 0
369 if rg[0] == rg[1]: # single value - check equal
370 pline += "((c) == 0x%x)" % rg[0]
371 else: # value range
372 # since we are doing char, also change range ending in 0xff
373 if rg[1] != 0xff:
374 pline += "((0x%x <= (c)) &&" % rg[0]
375 pline += " ((c) <= 0x%x))" % rg[1]
376 else:
377 pline += " (0x%x <= (c))" % rg[0]
378 pline += ")\n"
379 header.write(pline)
380
381 header.write("""
382/**
383 * %sQ:
384 * @c: char to validate
385 *
386 * Automatically generated by genChRanges.py
387 */
388""" % f)
389 pline = "#define %sQ(c)" % f
390 ntab = 4 - (len(pline)) // 8
391 if ntab < 0:
392 ntab = 0
393 just = ""
394 for i in range(ntab):
395 just += "\t"
396 header.write(pline + just + "(((c) < 0x100) ? \\\n\t\t\t\t ")
397 if max(Functs[f][0]) > 0:
398 header.write("%s_ch((c)) :" % f)
399 else:
400 header.write("0 :")
401
402 # if no ranges defined, value invalid if >= 0x100
403 numRanges = len(Functs[f][1])
404 if numRanges == 0:
405 header.write(" 0)\n\n")
406 else:
407 if numRanges >= minTableSize:
408 header.write(" \\\n\t\t\t\t xmlCharInRange((c), &%sGroup))\n\n" % f)
409 else: # if < minTableSize, generate inline code
410 firstFlag = 1
411 for rg in Functs[f][1]:
412 if not firstFlag:
413 pline += " || \\\n\t\t\t\t "
414 else:
415 firstFlag = 0
416 pline = "\\\n\t\t\t\t("
417 if rg[0] == rg[1]: # single value - check equal
418 pline += "((c) == 0x%x)" % rg[0]
419 else: # value range
420 pline += "((0x%x <= (c)) &&" % rg[0]
421 pline += " ((c) <= 0x%x))" % rg[1]
422 pline += "))\n\n"
423 header.write(pline)
424
425
426 if len(Functs[f][1]) > 0:
427 header.write("XMLPUBVAR const xmlChRangeGroup %sGroup;\n" % f)
428
429
430#
431# Next we do the unicode ranges
432#
433
434for f in fkeys:
435 if len(Functs[f][1]) > 0: # only generate if unicode ranges present
436 rangeTable = Functs[f][1]
437 rangeTable.sort() # ascending tuple sequence
438 numShort = 0
439 numLong = 0
440 for rg in rangeTable:
441 if rg[1] < 0x10000: # if short value
442 if numShort == 0: # first occurrence
443 pline = "static const xmlChSRange %s_srng[] = {" % f
444 else:
445 pline += ","
446 numShort += 1
447 if len(pline) > 60:
448 output.write(pline + "\n")
449 pline = " "
450 else:
451 pline += " "
452 pline += "{0x%x, 0x%x}" % (rg[0], rg[1])
453 else: # if long value
454 if numLong == 0: # first occurrence
455 if numShort > 0: # if there were shorts, finish them off
456 output.write(pline + "};\n")
457 pline = "static const xmlChLRange %s_lrng[] = { " % f
458 else:
459 pline += ", "
460 numLong += 1
461 if len(pline) > 60:
462 output.write(pline + "\n")
463 pline = " "
464 pline += "{0x%x, 0x%x}" % (rg[0], rg[1])
465 output.write(pline + "};\n") # finish off last group
466
467 pline = "const xmlChRangeGroup %sGroup =\n\t{%d, %d, " % (f, numShort, numLong)
468 if numShort > 0:
469 pline += "%s_srng" % f
470 else:
471 pline += "(xmlChSRangePtr)0"
472 if numLong > 0:
473 pline += ", %s_lrng" % f
474 else:
475 pline += ", (xmlChLRangePtr)0"
476
477 output.write(pline + "};\n\n")
478
479output.write(
480"""
481/**
482 * xmlCharInRange:
483 * @val: character to be validated
484 * @rptr: pointer to range to be used to validate
485 *
486 * Does a binary search of the range table to determine if char
487 * is valid
488 *
489 * Returns: true if character valid, false otherwise
490 */
491int
492xmlCharInRange (unsigned int val, const xmlChRangeGroup *rptr) {
493 int low, high, mid;
494 const xmlChSRange *sptr;
495 const xmlChLRange *lptr;
496
497 if (rptr == NULL) return(0);
498 if (val < 0x10000) {\t/* is val in 'short' or 'long' array? */
499\tif (rptr->nbShortRange == 0)
500\t return 0;
501\tlow = 0;
502\thigh = rptr->nbShortRange - 1;
503\tsptr = rptr->shortRange;
504\twhile (low <= high) {
505\t mid = (low + high) / 2;
506\t if ((unsigned short) val < sptr[mid].low) {
507\t\thigh = mid - 1;
508\t } else {
509\t\tif ((unsigned short) val > sptr[mid].high) {
510\t\t low = mid + 1;
511\t\t} else {
512\t\t return 1;
513\t\t}
514\t }
515\t}
516 } else {
517\tif (rptr->nbLongRange == 0) {
518\t return 0;
519\t}
520\tlow = 0;
521\thigh = rptr->nbLongRange - 1;
522\tlptr = rptr->longRange;
523\twhile (low <= high) {
524\t mid = (low + high) / 2;
525\t if (val < lptr[mid].low) {
526\t\thigh = mid - 1;
527\t } else {
528\t\tif (val > lptr[mid].high) {
529\t\t low = mid + 1;
530\t\t} else {
531\t\t return 1;
532\t\t}
533\t }
534\t}
535 }
536 return 0;
537}
538
539""");
540
541#
542# finally, generate the ABI compatibility functions
543#
544for f in fkeys:
545 output.write("""
546/**
547 * %s:
548 * @ch: character to validate
549 *
550 * This function is DEPRECATED.
551""" % f);
552 if max(Functs[f][0]) > 0:
553 output.write(" * Use %s_ch or %sQ instead" % (f, f))
554 else:
555 output.write(" * Use %sQ instead" % f)
556 output.write("""
557 *
558 * Returns true if argument valid, false otherwise
559 */
560""")
561 output.write("int\n%s(unsigned int ch) {\n return(%sQ(ch));\n}\n\n" % (f,f))
562 header.write("XMLPUBFUN int\n\t\t%s(unsigned int ch);\n" % f);
563#
564# Run complete - write trailers and close the output files
565#
566
567header.write("""
568#ifdef __cplusplus
569}
570#endif
571#endif /* __XML_CHVALID_H__ */
572""")
573
574header.close()
575
576output.close()
577
注意: 瀏覽 TracBrowser 來幫助您使用儲存庫瀏覽器

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette