2 * An XML parser for Wikipedia Data dumps.
3 * Converts XML files to tab-separated values files readable by spreadsheets
4 * and statistical packages.
16 #include "dtl/dtl.hpp"
24 // timestamp of the form 2003-11-07T00:43:23Z
25 #define DATE_LENGTH 10
27 #define TIMESTAMP_LENGTH 20
29 #define MEGABYTE 1048576
30 #define FIELD_BUFFER_SIZE 1024
32 // this can be changed at runtime if we encounter an article larger than 10mb
33 size_t text_buffer_size = 10 * MEGABYTE;
36 TITLE, ARTICLEID, REVISION, REVID, TIMESTAMP, CONTRIBUTOR,
37 EDITOR, EDITORID, MINOR, COMMENT, UNUSED, TEXT
40 enum block { TITLE_BLOCK, REVISION_BLOCK, CONTRIBUTOR_BLOCK, SKIP };
42 enum outtype { FULL, SIMPLE };
46 // pointers to once-allocated buffers
58 vector<string> last_text_tokens;
59 vector<pcrecpp::RE> title_regexes;
60 vector<string> diff_regex_names;
61 vector<pcrecpp::RE> diff_regexes;
62 map<string, string> revision_md5; // used for detecting reversions
64 // track string size of the elements, to prevent O(N^2) processing in charhndl
65 // when we have to take strlen for every character which we append to the buffer
67 size_t articleid_size;
71 size_t timestamp_size;
80 enum elements element;
82 enum outtype output_type;
87 /* free_data and clean_data
88 * Takes a pointer to the data struct and an integer {0,1} indicating if the
89 * title data needs to be cleared as well.
90 * Also, frees memory dynamically allocated to store data.
93 clean_data(revisionData *data, int title)
95 // reset title (if we are switching articles)
97 data->title[0] = '\0';
98 data->articleid[0] = '\0';
100 data->articleid_size = 0;
104 data->revid[0] = '\0';
105 data->date[0] = '\0';
106 data->time[0] = '\0';
107 data->timestamp[0] = '\0';
108 data->anon[0] = '\0';
109 data->editor[0] = '\0';
110 data->editorid[0] = '\0';
111 data->comment[0] = '\0';
112 data->text[0] = '\0';
114 // reset length tracking
115 data->revid_size = 0;
118 data->timestamp_size = 0;
120 data->editor_size = 0;
121 data->editorid_size = 0;
122 data->comment_size = 0;
125 // reset flags and element type info
127 data->element = UNUSED;
133 free_data(revisionData *data, int title)
136 //printf("freeing article\n");
138 free(data->articleid);
143 free(data->timestamp);
146 free(data->editorid);
149 data->last_text_tokens.clear();
152 void cleanup_revision(revisionData *data) {
156 void cleanup_article(revisionData *data) {
158 data->last_text_tokens.clear();
159 data->revision_md5.clear();
164 init_data(revisionData *data, outtype output_type)
166 data->text = (char*) malloc(text_buffer_size);
167 data->comment = (char*) malloc(FIELD_BUFFER_SIZE);
168 data->title = (char*) malloc(FIELD_BUFFER_SIZE);
169 data->articleid = (char*) malloc(FIELD_BUFFER_SIZE);
170 data->revid = (char*) malloc(FIELD_BUFFER_SIZE);
171 data->date = (char*) malloc(FIELD_BUFFER_SIZE);
172 data->time = (char*) malloc(FIELD_BUFFER_SIZE);
173 data->timestamp = (char*) malloc(FIELD_BUFFER_SIZE);
174 data->anon = (char*) malloc(FIELD_BUFFER_SIZE);
175 data->editor = (char*) malloc(FIELD_BUFFER_SIZE);
176 data->editorid = (char*) malloc(FIELD_BUFFER_SIZE);
179 // resets the data fields, null terminates strings, sets lengths
182 data->output_type = output_type;
185 /* for debugging only, prints out the state of the data struct
188 print_state(revisionData *data)
190 printf("element = %i\n", data->element);
191 printf("output_type = %i\n", data->output_type);
192 printf("title = %s\n", data->title);
193 printf("articleid = %s\n", data->articleid);
194 printf("revid = %s\n", data->revid);
195 printf("date = %s\n", data->date);
196 printf("time = %s\n", data->time);
197 printf("anon = %s\n", data->anon);
198 printf("editor = %s\n", data->editor);
199 printf("editorid = %s\n", data->editorid);
200 printf("minor = %s\n", (data->minor ? "1" : "0"));
201 printf("comment = %s\n", data->comment);
202 printf("text = %s\n", data->text);
209 * write a line of comma-separated value formatted data to standard out
211 * title,articleid,revid,date,time,anon,editor,editorid,minor,comment
212 * (str) (int) (int) (str)(str)(bin)(str) (int) (bin) (str)
214 * it is called right before cleanup_revision() and cleanup_article()
217 write_row(revisionData *data)
222 md5_byte_t digest[16];
223 char md5_hex_output[2 * 16 + 1];
225 md5_append(&state, (const md5_byte_t *)data->text, data->text_size);
226 md5_finish(&state, digest);
228 for (di = 0; di < 16; ++di) {
229 sprintf(md5_hex_output + di * 2, "%02x", digest[di]);
233 map<string, string>::iterator prev_revision = data->revision_md5.find(md5_hex_output);
234 if (prev_revision != data->revision_md5.end()) {
235 reverted_to = prev_revision->second; // id of previous revision
237 data->revision_md5[md5_hex_output] = data->revid;
239 string text = string(data->text, data->text_size);
240 vector<string> text_tokens;
243 while ((pos = text.find_first_of(" \n\t\r", pos)) != string::npos) {
244 //cout << "\"\"\"" << text.substr(start, pos - start) << "\"\"\"" << endl;
245 text_tokens.push_back(text.substr(start, pos - start));
250 // look to see if (a) we've passed in a list of /any/ title_regexes
251 // and (b) if all of the title_regex_matches match
252 // if (a) is true and (b) is not, we return
253 bool any_title_regex_match = false;
254 if (!data->title_regexes.empty()) {
255 for (vector<pcrecpp::RE>::iterator r = data->title_regexes.begin(); r != data->title_regexes.end(); ++r) {
256 pcrecpp::RE& title_regex = *r;
257 if (title_regex.PartialMatch(data->title)) {
258 any_title_regex_match = true;
262 if (!any_title_regex_match) {
267 //vector<string> additions;
268 //vector<string> deletions;
272 vector<bool> diff_regex_matches_adds;
273 vector<bool> diff_regex_matches_dels;
275 if (data->last_text_tokens.empty()) {
276 additions = data->text;
280 dtl::Diff< string, vector<string> > d(data->last_text_tokens, text_tokens);
281 //d.onOnlyEditDistance();
284 vector<pair<string, dtl::elemInfo> > ses_v = d.getSes().getSequence();
285 for (vector<pair<string, dtl::elemInfo> >::iterator sit=ses_v.begin(); sit!=ses_v.end(); ++sit) {
286 switch (sit->second.type) {
288 //cout << "ADD: \"" << sit->first << "\"" << endl;
289 additions += sit->first;
291 case dtl::SES_DELETE:
292 //cout << "DEL: \"" << sit->first << "\"" << endl;
293 deletions += sit->first;
299 if (!additions.empty()) {
300 //cout << "ADD: " << additions << endl;
301 for (vector<pcrecpp::RE>::iterator r = data->diff_regexes.begin(); r != data->diff_regexes.end(); ++r) {
302 pcrecpp::RE& diff_regex = *r;
303 diff_regex_matches_adds.push_back(diff_regex.PartialMatch(additions));
307 if (!deletions.empty()) {
308 //cout << "DEL: " << deletions << endl;
309 for (vector<pcrecpp::RE>::iterator r = data->diff_regexes.begin(); r != data->diff_regexes.end(); ++r) {
310 pcrecpp::RE& diff_regex = *r;
311 diff_regex_matches_dels.push_back(diff_regex.PartialMatch(deletions));
315 data->last_text_tokens = text_tokens;
318 // print line of tsv output
320 << data->title << "\t"
321 << data->articleid << "\t"
322 << data->revid << "\t"
324 << data->time << "\t"
325 << ((data->editor[0] != '\0') ? "FALSE" : "TRUE") << "\t"
326 << data->editor << "\t"
327 << data->editorid << "\t"
328 << ((data->minor) ? "TRUE" : "FALSE") << "\t"
329 << (unsigned int) data->text_size << "\t"
330 << shannon_H(data->text, data->text_size) << "\t"
331 << md5_hex_output << "\t"
332 << reverted_to << "\t"
333 << (int) additions.size() << "\t"
334 << (int) deletions.size();
336 for (int n = 0; n < data->diff_regex_names.size(); ++n) {
337 cout << "\t" << ((!diff_regex_matches_adds.empty() && diff_regex_matches_adds.at(n)) ? "TRUE" : "FALSE")
338 << "\t" << ((!diff_regex_matches_dels.empty() && diff_regex_matches_dels.at(n)) ? "TRUE" : "FALSE");
343 if (data->output_type == FULL) {
344 cout << "comment:" << data->comment << endl
345 << "text:" << endl << data->text << endl;
351 split_timestamp(revisionData *data)
353 char *t = data->timestamp;
354 strncpy(data->date, data->timestamp, DATE_LENGTH);
355 char *timeinstamp = &data->timestamp[DATE_LENGTH+1];
356 strncpy(data->time, timeinstamp, TIME_LENGTH);
359 // like strncat but with previously known length
361 strlcatn(char *dest, const char *src, size_t dest_len, size_t n)
365 for (i = 0 ; i < n && src[i] != '\0' ; i++)
366 dest[dest_len + i] = src[i];
367 dest[dest_len + i] = '\0';
373 charhndl(void* vdata, const XML_Char* s, int len)
375 revisionData* data = (revisionData*) vdata;
377 if (data->element != UNUSED && data->position != SKIP) {
378 switch (data->element) {
380 // check if we'd overflow our buffer
381 bufsz = data->text_size + len;
382 if (bufsz + 1 > text_buffer_size) {
383 data->text = (char*) realloc(data->text, bufsz + 1);
384 text_buffer_size = bufsz + 1;
386 strlcatn(data->text, s, data->text_size, len);
387 data->text_size = bufsz;
390 strlcatn(data->comment, s, data->comment_size, len);
391 data->comment_size += len;
394 strlcatn(data->title, s, data->title_size, len);
395 data->title_size += len;
398 // printf("articleid = %s\n", t);
399 strlcatn(data->articleid, s, data->articleid_size, len);
400 data->articleid_size += len;
403 // printf("revid = %s\n", t);
404 strlcatn(data->revid, s, data->revid_size, len);
405 data->revid_size += len;
408 strlcatn(data->timestamp, s, data->timestamp_size, len);
409 data->timestamp_size += len;
410 if (strlen(data->timestamp) == TIMESTAMP_LENGTH)
411 split_timestamp(data);
414 strlcatn(data->editor, s, data->editor_size, len);
415 data->editor_size += len;
418 //printf("editorid = %s\n", t);
419 strlcatn(data->editorid, s, data->editorid_size, len);
420 data->editorid_size += len;
422 /* the following are implied or skipped:
424 printf("found minor element\n"); doesn't work
425 break; minor tag is just a tag
434 start(void* vdata, const XML_Char* name, const XML_Char** attr)
436 revisionData* data = (revisionData*) vdata;
438 if (strcmp(name,"title") == 0) {
439 cleanup_article(data); // cleans up data from last article
440 data->element = TITLE;
441 data->position = TITLE_BLOCK;
442 } else if (data->position != SKIP) {
443 if (strcmp(name,"revision") == 0) {
444 data->element = REVISION;
445 data->position = REVISION_BLOCK;
446 } else if (strcmp(name, "contributor") == 0) {
447 data->element = CONTRIBUTOR;
448 data->position = CONTRIBUTOR_BLOCK;
449 } else if (strcmp(name,"id") == 0)
450 switch (data->position) {
452 data->element = ARTICLEID;
455 data->element = REVID;
457 case CONTRIBUTOR_BLOCK:
458 data->element = EDITORID;
462 // minor tag has no character data, so we parse here
463 else if (strcmp(name,"minor") == 0) {
464 data->element = MINOR;
467 else if (strcmp(name,"timestamp") == 0)
468 data->element = TIMESTAMP;
470 else if (strcmp(name, "username") == 0)
471 data->element = EDITOR;
473 else if (strcmp(name,"ip") == 0)
474 data->element = EDITORID;
476 else if (strcmp(name,"comment") == 0)
477 data->element = COMMENT;
479 else if (strcmp(name,"text") == 0)
480 data->element = TEXT;
482 else if (strcmp(name,"page") == 0
483 || strcmp(name,"mediawiki") == 0
484 || strcmp(name,"restrictions") == 0
485 || strcmp(name,"siteinfo") == 0)
486 data->element = UNUSED;
493 end(void* vdata, const XML_Char* name)
495 revisionData* data = (revisionData*) vdata;
496 if (strcmp(name, "revision") == 0 && data->position != SKIP) {
497 write_row(data); // crucial... :)
498 cleanup_revision(data); // also crucial
500 data->element = UNUSED; // sets our state to "not-in-useful"
501 } // thus avoiding unpleasant character data
502 // b/w tags (newlines etc.)
505 void print_usage(char* argv[]) {
506 cerr << "usage: <wikimedia dump xml> | " << argv[0] << "[options]" << endl
508 << "options:" << endl
509 << " -v verbose mode prints text and comments after each line of tab separated data" << endl
510 << " -n name of the following regex (e.g. -n name -r \"...\")" << endl
511 << " -r regex to check against additions and deletions" << endl
512 << " -t parse revisions only from pages whose titles match regex(es)" << endl
514 << "Takes a wikimedia data dump XML stream on standard in, and produces" << endl
515 << "a tab-separated stream of revisions on standard out:" << endl
517 << "title, articleid, revid, timestamp, anon, editor, editorid, minor," << endl
518 << "text_length, text_entropy, text_md5, reversion, additions_size, deletions_size" << endl
519 << ".... and additional fields for each regex executed against add/delete diffs" << endl
521 << "Boolean fields are TRUE/FALSE except in the case of reversion, which is blank" << endl
522 << "unless the article is a revert to a previous revision, in which case, it" << endl
523 << "contains the revision ID of the revision which was reverted to." << endl
525 << "author: Erik Garrison <erik@hypervolu.me>" << endl;
530 main(int argc, char *argv[])
533 enum outtype output_type;
535 // in "simple" output, we don't print text and comments
536 output_type = SIMPLE;
538 string diff_regex_name;
540 // the user data struct which is passed to callback functions
543 while ((c = getopt(argc, argv, "hvn:r:t:")) != -1)
553 diff_regex_name = optarg;
556 data.diff_regexes.push_back(pcrecpp::RE(optarg, pcrecpp::UTF8()));
557 data.diff_regex_names.push_back(diff_regex_name);
558 if (!diff_regex_name.empty()) {
559 diff_regex_name.clear();
567 data.title_regexes.push_back(pcrecpp::RE(optarg, pcrecpp::UTF8()));
571 if (dry_run) { // lets us print initialization options
572 printf("simple_output = %i\n", output_type);
576 // create a new instance of the expat parser
577 XML_Parser parser = XML_ParserCreate("UTF-8");
579 // initialize the elements of the struct to default values
580 init_data(&data, output_type);
583 // makes the parser pass "data" as the first argument to every callback
584 XML_SetUserData(parser, &data);
585 void (*startFnPtr)(void*, const XML_Char*, const XML_Char**) = start;
586 void (*endFnPtr)(void*, const XML_Char*) = end;
587 void (*charHandlerFnPtr)(void*, const XML_Char*, int) = charhndl;
589 // sets start and end to be the element start and end handlers
590 XML_SetElementHandler(parser, startFnPtr, endFnPtr);
591 // sets charhndl to be the callback for character data
592 XML_SetCharacterDataHandler(parser, charHandlerFnPtr);
599 cout << "title" << "\t"
600 << "articleid" << "\t"
606 << "editor_id" << "\t"
608 << "text_size" << "\t"
609 << "text_entropy" << "\t"
610 << "text_md5" << "\t"
611 << "reversion" << "\t"
612 << "additions_size" << "\t"
616 if (!data.diff_regexes.empty()) {
617 for (vector<pcrecpp::RE>::iterator r = data.diff_regexes.begin(); r != data.diff_regexes.end(); ++r, ++n) {
618 if (data.diff_regex_names.at(n).empty()) {
619 cout << "\t" << "regex_" << n << "_add"
620 << "\t" << "regex_" << n << "_del";
622 cout << "\t" << data.diff_regex_names.at(n) << "_add"
623 << "\t" << data.diff_regex_names.at(n) << "_del";
629 // shovel data into the parser
632 // read into buf a bufferfull of data from standard input
633 size_t len = fread(buf, 1, BUFSIZ, stdin);
634 done = len < BUFSIZ; // checks if we've got the last bufferfull
636 // passes the buffer of data to the parser and checks for error
637 // (this is where the callbacks are invoked)
638 if (XML_Parse(parser, buf, len, done) == XML_STATUS_ERROR) {
639 cerr << "XML ERROR: " << XML_ErrorString(XML_GetErrorCode(parser)) << " at line "
640 << (int) XML_GetCurrentLineNumber(parser) << endl;
646 XML_ParserFree(parser);