mediawiki_dump_tools/wikiq

#!/usr/bin/env python3

# original wikiq headers are: title articleid revid date_time anon
# editor editor_id minor text_size text_entropy text_md5 reversion
# additions_size deletions_size

import argparse
import sys
import os, os.path
import re
from datetime import datetime

from subprocess import Popen, PIPE
from collections import deque
from hashlib import sha1

from mwxml import Dump

from deltas.tokenizers import wikitext_split
import mwpersistence
import mwreverts
from urllib.parse import quote
TO_ENCODE = ('title', 'editor')
PERSISTENCE_RADIUS=7
from deltas import SequenceMatcher
from deltas import SegmentMatcher

from dataclasses import dataclass
import pyarrow as pa
import pyarrow.parquet as pq

class PersistMethod:
    none = 0
    sequence = 1
    segment = 2
    legacy = 3

def calculate_persistence(tokens_added):
    return(sum([(len(x.revisions)-1) for x in tokens_added]),
           len(tokens_added))


class WikiqIterator():
    def __init__(self, fh, collapse_user=False):
        self.fh = fh
        self.collapse_user = collapse_user
        self.mwiterator = Dump.from_file(self.fh)
        self.namespace_map = { ns.id : ns.name for ns in
                               self.mwiterator.site_info.namespaces }
        self.__pages = self.load_pages()

    def load_pages(self):
        for page in self.mwiterator:
            yield WikiqPage(page,
                            namespace_map = self.namespace_map,
                            collapse_user=self.collapse_user)

    def __iter__(self):
        return self.__pages

    def __next__(self):
        return next(self._pages)

class WikiqPage():
    __slots__ = ('id', 'title', 'namespace', 'redirect',
                 'restrictions', 'mwpage', '__revisions',
                 'collapse_user')
    
    def __init__(self, page, namespace_map, collapse_user=False):
        self.id = page.id
        self.namespace = page.namespace
        # following mwxml, we assume namespace 0 in cases where
        # page.namespace is inconsistent with namespace_map
        if page.namespace not in namespace_map:
            self.title = page.title
            page.namespace = 0
        if page.namespace != 0:
            self.title = ':'.join([namespace_map[page.namespace], page.title])
        else:
            self.title = page.title
        self.restrictions = page.restrictions
        self.collapse_user = collapse_user
        self.mwpage = page
        self.__revisions = self.rev_list()

    def rev_list(self):
        # Outline for how we want to handle collapse_user=True
        # iteration   rev.user   prev_rev.user   add prev_rev?
        #         0          A            None           Never
        #         1          A               A           False
        #         2          B               A            True
        #         3          A               B            True
        #         4          A               A           False
        # Post-loop                          A          Always
        for i, rev in enumerate(self.mwpage):
            # never yield the first time
            if i == 0:
                if self.collapse_user: 
                    collapsed_revs = 1
                    rev.collapsed_revs = collapsed_revs

            else:
                if self.collapse_user:
                    # yield if this is the last edit in a seq by a user and reset
                    # also yield if we do know who the user is

                    if rev.deleted.user or prev_rev.deleted.user:
                        yield prev_rev
                        collapsed_revs = 1
                        rev.collapsed_revs = collapsed_revs

                    elif not rev.user.text == prev_rev.user.text:
                        yield prev_rev
                        collapsed_revs = 1
                        rev.collapsed_revs = collapsed_revs
                    # otherwise, add one to the counter
                    else:
                        collapsed_revs += 1
                        rev.collapsed_revs = collapsed_revs
                # if collapse_user is false, we always yield
                else:
                    yield prev_rev

            prev_rev = rev

        # also yield the final time
        yield prev_rev

    def __iter__(self):
        return self.__revisions

    def __next__(self):
        return next(self.__revisions)


class RegexPair(object):
    def __init__(self, pattern, label):
        self.pattern = re.compile(pattern)
        self.label = label
        self.has_groups = bool(self.pattern.groupindex)
        if self.has_groups:
            self.capture_groups = list(self.pattern.groupindex.keys())
            
    def _make_key(self, cap_group):
        return ("{}_{}".format(self.label, cap_group))

    def matchmake(self, content, rev_data):
        
        temp_dict = {}
        # if there are named capture groups in the regex
        if self.has_groups:

            # if there are matches of some sort in this revision content, fill the lists for each cap_group
            if self.pattern.search(content) is not None:
                m = self.pattern.finditer(content)
                matchobjects = list(m)

                for cap_group in self.capture_groups:
                    key = self._make_key(cap_group)
                    temp_list = []
                    for match in matchobjects:
                        # we only want to add the match for the capture group if the match is not None
                        if match.group(cap_group) != None:
                            temp_list.append(match.group(cap_group))

                    # if temp_list of matches is empty just make that column None
                    if len(temp_list)==0:
                        temp_dict[key] = None
                    # else we put in the list we made in the for-loop above
                    else:
                        temp_dict[key] = ', '.join(temp_list)

            # there are no matches at all in this revision content, we default values to None
            else:
                for cap_group in self.capture_groups:
                    key = self._make_key(cap_group)
                    temp_dict[key] = None

        # there are no capture groups, we just search for all the matches of the regex
        else:
            #given that there are matches to be made
            if type(content) in(str, bytes):
                if self.pattern.search(content) is not None:
                    m = self.pattern.findall(content)
                    temp_dict[self.label] = ', '.join(m)
                else:
                    temp_dict[self.label] = None

        # update rev_data with our new columns
        for k, v in temp_dict:
            rev_data.setattr(k,v)

        return rev_data

@dataclass()
class RevDataBase():
    revid: int 
    date_time: datetime
    articleid: int
    editorid: int
    title: str
    namespace: int
    deleted: bool
    text_chars: int = None
    revert: bool = None
    reverteds: list[int] = None
    sha1: str = None
    minor: bool = None
    editor: str = None
    anon: bool = None
    collapsed_revs:int = None

    pa_schema_fields = [
        pa.field("revid", pa.int64),
        pa.field("date_time",pa.timestamp('ms')),
        pa.field("articleid",pa.int64()),
        pa.field("editorid",pa.int64()),
        pa.field("title",pa.string()),
        pa.field("namespace",pa.int32()),
        pa.field("deleted",pa.binary()),
        pa.field("test_chars",pa.int32()),
        pa.field("revert",pa.binary()),
        pa.field("reverteds",pa.list_(pa.int64())),
        pa.field("sha1",pa.string()),
        pa.field("minor",pa.binary()),
        pa.field("editor",pa.string()),
        pa.field("anon",pa.binary())
    ]

    def to_pyarrow(self):
        return pa.array(self.astuple(), map(self.pa_schema_fields, pa.field.type))
        

    def to_tsv_row(self):
        
        row = []
        for f in self.fields():
            val = getattr(self, f.name)
            if getattr(self, f.name) is None:
                row.append("")
            elif f.type == bool:
                row.append("TRUE" if val else "FALSE")

            elif f.type == datetime:
                row.append(val.strftime('%Y-%m-%d %H:%M:%S'))

            elif f.name in {'editor','title'}:
                s = '"' + val + '"'
                if f.name in TO_ENCODE:
                    row.append(quote(str(val)))

            elif f.type == list[int]:
                row.append('"' + ",".join([str(x) for x in val]) + '"')

            elif f.type == str:
                if f.name in TO_ENCODE:
                    row.append(quote(str(val)))
            else:
                row.append(val)

        return '\t'.join(row)
    
    # def __init__(revid: int,
    #              date_time: datetime,
    #              articleid: int,
    #              editorid: int,
    #              title: str,
    #              namespace: int,
    #              deleted: bool,
    #              test_chars: int,
    #              revert: bool,
    #              reverteds: list[bool],
    #              sha1: str,
    #              minor: bool,
    #              editor: str,
    #              anon: bool):

        

@dataclass()
class RevDataCollapse(RevDataBase):
    collapsed_revs:int = None
    pa_collapsed_revs_schema = pa.field('collapsed_revs',pa.int64())
    pa_schema_fields = RevDataBase.pa_schema_fields + pa_collapsed_revs_schema
    pa_schema = pa.schema(pa_schema_fields)    

@dataclass()
class RevDataPersistence(RevDataBase):
    token_revs:int = None
    tokens_added:int = None
    tokens_removed:int = None
    tokens_window:int = None
    pa_persistence_schema_fields = [
        pa.field(token_revs, pa.int64()),
        pa.field(tokens_added, pa.int64()),
        pa.field(tokens_removed, pa.int64()),
        pa.tokens_window, pa.int64()]
        
    pa_schema_fields = RevDataBase.pa_schema_fields  + pa_persistence_schema_fields

@dataclass()
class RevDataCollapsePersistence(RevDataCollapse, RevDataPersistence):
    pa_scehma_fields = RevDataCollapse.pa_schema_fields + RevDataPersistence.pa_persistence_schema_fields

class WikiqParser():
    def __init__(self, input_file, output_file, regex_match_revision, regex_match_comment, regex_revision_label, regex_comment_label, collapse_user=False, persist=None, urlencode=False, namespaces = None, revert_radius=15, output_parquet=True, parquet_buffer_size=2000):
        """ 
        Parameters:
           persist : what persistence method to use. Takes a PersistMethod value
        """
        self.input_file = input_file

        self.collapse_user = collapse_user
        self.persist = persist
        self.namespaces = []
        self.urlencode = urlencode
        self.revert_radius = revert_radius
        
        self.output_buffer = []
        self.output_buffer_size = output_buffer_size

        if namespaces is not None:
            self.namespace_filter = set(namespaces)
        else:
            self.namespace_filter = None

        self.regex_schemas = []
        self.regex_revision_pairs = self.make_matchmake_pairs(regex_match_revision, regex_revision_label)
        self.regex_comment_pairs = self.make_matchmake_pairs(regex_match_comment, regex_comment_label)

        if self.collapse_user is True:
            if self.persist == PersistMethod.none:
                revdata_type = RevDataCollapse
            else:
                revdata_type = RevDataCollapsePersistence
        elif self.persist != PersistMethod.none:
            revdata_type = RevDataPersistence
        else:
            revdata_type = RevDataBase

        regex_fields = [(field.name, list[str]), for field in self.regex_schemas]
        self.revdata_type = dataclasses.make_dataclass('RevData_Parser',
                                                       fields=map(regex_fields,
                                                                  lambda pa_field: (pa_field.name,
                                                                                    list[string],
                                                                                    field(default=None))),
                                                       bases=(revdata_type))
        
        self.revdata_type.pa_schema_fields = revdata_type.pa_schema_fields + regex_fields
                        
        if output_parquet is True:
            self.output_parquet = True
            self.pq_writer = None
            self.output_file = output_file
        else:
            self.output_file = open(output_file,'w')


    def make_matchmake_pairs(self, patterns, labels):
        if (patterns is not None and labels is not None) and \
           (len(patterns) == len(labels)):
            result = []
            for pattern, label in zip(patterns, labels):
                result.append(RegexPair(pattern, label))
                self.regex_schemas.append(pa.field(label, pa.list_(pa.string())))

            return result
        elif (patterns is None and labels is None):
            return []
        else:
            sys.exit('Each regular expression *must* come with a corresponding label and vice versa.')

    def matchmake(self, rev, rev_data):
        rev_data = self.matchmake_revision(rev.text, rev_data)
        rev_data = self.matchmake_comment(rev.comment, rev_data)
        return rev_data

    def matchmake_revision(self, text, rev_data):
         return self.matchmake_pairs(text, rev_data, self.regex_revision_pairs)

    def matchmake_comment(self, comment, rev_data):
        return self.matchmake_pairs(comment, rev_data, self.regex_comment_pairs)

    def matchmake_pairs(self, text, rev_data, pairs):
        for pair in pairs:
            rev_data = pair.matchmake(text, rev_data)
        return rev_data

    def __get_namespace_from_title(self, title):
        default_ns = None

        for ns in self.namespaces:
            # skip if the namespace is not defined
            if ns == None:
                default_ns = self.namespaces[ns]
                continue

            if title.startswith(ns + ":"):
                return self.namespaces[ns]

        # if we've made it this far with no matches, we return the default namespace
        return default_ns


    def process(self):

        # create a regex that creates the output filename
        # output_filename = re.sub(r'^.*/(enwiki\-\d+)\-.*p(\d+)p.*$',
        #                         r'output/wikiq-\1-\2.tsv',
        #                         input_filename)

        # Construct dump file iterator
        dump = WikiqIterator(self.input_file, collapse_user=self.collapse_user)

        # extract list of namspaces
        self.namespaces = {ns.name : ns.id for ns in dump.mwiterator.site_info.namespaces}

        page_count = 0
        rev_count = 0


        # Iterate through pages
        for page in dump:
            namespace = page.namespace if page.namespace is not None else self.__get_namespace_from_title(page.title)

            # skip namespaces not in the filter
            if self.namespace_filter is not None:
                if namespace not in self.namespace_filter:
                    continue

            rev_detector = mwreverts.Detector(radius = self.revert_radius)

            if self.persist != PersistMethod.none:
                window = deque(maxlen=PERSISTENCE_RADIUS)

                if self.persist == PersistMethod.sequence:
                    state = mwpersistence.DiffState(SequenceMatcher(tokenizer = wikitext_split),
                                                    revert_radius=PERSISTENCE_RADIUS)

                elif self.persist == PersistMethod.segment:
                    state = mwpersistence.DiffState(SegmentMatcher(tokenizer = wikitext_split),
                                                    revert_radius=PERSISTENCE_RADIUS)

                # self.persist == PersistMethod.legacy
                else:
                    from mw.lib import persistence
                    state = persistence.State()

            # Iterate through a page's revisions
            for rev in page:
                
                rev_data = self.revdata_type(revid = rev.id,
                                             date_time = rev.timestamp,
                                             articleid = page.id,
                                             editorid = "" if rev.deleted.user == True or rev.user.id is None else rev.user.id,
                                             title =  page.title,
                                             deleted = rev.deleted.text
                                             )

                rev_data = self.matchmake(rev, rev_data)

                if not rev.deleted.text:
                    # rev.text can be None if the page has no text
                    if not rev.text:
                        rev.text = ""
                    # if text exists, we'll check for a sha1 and generate one otherwise

                    if rev.sha1:
                        text_sha1 = rev.sha1
                    else:

                        text_sha1 = sha1(bytes(rev.text, "utf8")).hexdigest()
                    
                    rev_data.sha1 = text_sha1

                    # TODO rev.bytes doesn't work.. looks like a bug
                    rev_data.text_chars = len(rev.text)

                    # generate revert data
                    rev_data.revert = rev_detector.process(text_sha1, rev.id)
                    
                    if revert:
                        rev_data.reverteds = revert.reverteds

                # if the fact that the edit was minor can be hidden, this might be an issue
                rev_data.minor = rev.minor

                if not rev.deleted.user:
                    # wrap user-defined editors in quotes for fread
                    rev_data.editor = rev.user.text 
                    rev_data.anon = rev.user.id == None
                    
                #if re.match(r'^#redirect \[\[.*\]\]', rev.text, re.I):
                #    redirect = True
                #else:
                #    redirect = False
                
                #TODO missing: additions_size deletions_size
                
                # if collapse user was on, lets run that
                if self.collapse_user:
                    rev_data.collapsed_revs = rev.collapsed_revs

                if self.persist != PersistMethod.none:

                    if not rev.deleted.text:

                        if self.persist != PersistMethod.legacy:
                            _, tokens_added, tokens_removed = state.update(rev.text, rev.id)

                        else:
                            _, tokens_added, tokens_removed = state.process(rev.text, rev.id, text_sha1)
                            
                        window.append((rev.id, rev_data, tokens_added, tokens_removed))
                        
                        if len(window) == PERSISTENCE_RADIUS:
                            old_rev_id, old_rev_data, old_tokens_added, old_tokens_removed = window[0]
                            
                            num_token_revs, num_tokens = calculate_persistence(old_tokens_added)

                            rev_data.token_revs = num_token_revs
                            rev_data.tokens_added = num_tokens
                            rev_data.tokens_removed = len(old_tokens_removed)
                            rev_data.tokens_window = PERSISTENCE_RADIUS-1

                            self.print_rev_data(rev_data)

                else:
                    self.print_rev_data(rev_data)

                rev_count += 1

            if self.persist != PersistMethod.none:
                # print out metadata for the last RADIUS revisions
                for i, item in enumerate(window):
                    # if the window was full, we've already printed item 0
                    if len(window) == PERSISTENCE_RADIUS and i == 0:
                        continue

                    rev_id, rev_data, tokens_added, tokens_removed = item
                    num_token_revs, num_tokens = calculate_persistence(tokens_added)

                    rev_data.token_revs = num_token_revs
                    rev_data.tokens_added = num_tokens
                    rev_data.tokens_removed = len(tokens_removed)
                    rev_data.tokens_window = len(window)-(i+1)
                    self.print_rev_data(rev_data)

            page_count += 1

        print("Done: %s revisions and %s pages." % (rev_count, page_count),
              file=sys.stderr)

        if self.output_parquet is True:
            self.flush_parquet_buffer()
            self.pq_writer.close()

        else:
            output_file.close()


    def write_parquet_row(self, rev_data):
        padata = rev_data.to_pyarrow()
        self.output_buffer.append(padata)

        if len(self.output_buffer) >= self.output_buffer_size:
            self.flush_parquet_buffer()


    def flush_parquet_buffer(self):
        outtable = pa.table.concat_arrays(self.output_buffer)                        
        if self.pq_writer is None:
            schema = pa.schema(self.revdata_type.pa_schema_field)
            self.pq_writer = pq.ParquetWriter(self.output_file, schema, flavor='spark')

        self.pq_writer.write_table(outtable)
        self.output_buffer = []
        
    def print_rev_data(self, rev_data):
        if self.output_parquet is False:
            printfunc = self.write_tsv_row
        else:
            printfunc = self.write_parquet_row
        
        printfunc(rev_data)

    def write_tsv_row(self, rev_data):
        
        self.output_buffer.append(rev_data.to_tsv_line())

        if len(self.output_buffer) >= self.output_buffer_size:
            self.flush_tsv_buffer()


    def flush_tsv_buffer():
        if self.output_header:

def open_input_file(input_filename):
    if re.match(r'.*\.7z$', input_filename):
        cmd = ["7za", "x", "-so", input_filename, "*.xml"] 
    elif re.match(r'.*\.gz$', input_filename):
        cmd = ["zcat", input_filename] 
    elif re.match(r'.*\.bz2$', input_filename):
        cmd = ["bzcat", "-dk", input_filename] 

    try:
        input_file = Popen(cmd, stdout=PIPE).stdout
    except NameError:
        input_file = open(input_filename, 'r')

    return input_file

def get_output_filename(input_filename, parquet = False):
    output_filename = re.sub(r'\.(7z|gz|bz2)?$', '', input_filename)
    output_filename = re.sub(r'\.xml', '', output_filename)
    if parquet is False:
        output_filename = output_filename + ".tsv"
    else:
        output_filename = output_filename + ".parquet"
    return output_filename

def open_output_file(input_filename):
    # create a regex that creates the output filename
    output_filename = get_output_filename(input_filename, parquet = False)
    output_file = open(output_filename, "w")
    return output_file

parser = argparse.ArgumentParser(description='Parse MediaWiki XML database dumps into tab delimitted data.')

# arguments for the input direction
parser.add_argument('dumpfiles', metavar="DUMPFILE", nargs="*", type=str, 
                    help="Filename of the compressed or uncompressed XML database dump. If absent, we'll look for content on stdin and output on stdout.")

parser.add_argument('-o', '--output-dir', metavar='DIR', dest='output_dir', type=str, nargs=1,
                    help="Directory for output files. If it ends with .parquet output will be in parquet format.")

parser.add_argument('-s', '--stdout', dest="stdout", action="store_true",
                    help="Write output to standard out (do not create dump file)")

parser.add_argument('--collapse-user', dest="collapse_user", action="store_true",
                    help="Operate only on the final revision made by user a user within all sequences of consecutive edits made by a user. This can be useful for addressing issues with text persistence measures.")

parser.add_argument('-p', '--persistence', dest="persist", default=None, const='', type=str, choices = ['','segment','sequence','legacy'], nargs='?',
                    help="Compute and report measures of content persistent: (1) persistent token revisions, (2) tokens added, and (3) number of revision used in computing the first measure. This may by slow.  The defualt is -p=sequence, which uses the same algorithm as in the past, but with improvements to wikitext parsing. Use -p=legacy for old behavior used in older research projects. Use -p=segment for advanced persistence calculation method that is robust to content moves, but prone to bugs, and slower.")

parser.add_argument('-u', '--url-encode', dest="urlencode", action="store_true",
                    help="Output url encoded text strings. This works around some data issues like newlines in editor names. In the future it may be used to output other text data.")

parser.add_argument('-n', '--namespace-include', dest="namespace_filter", type=int, action='append',
                    help="Id number of namspace to include. Can be specified more than once.")

parser.add_argument('-rr',
                    '--revert-radius',
                    dest="revert_radius",
                    type=int,
                    action='store',
                    default=15,
                    help="Number of edits to check when looking for reverts (default: 15)")

parser.add_argument('-RP', '--revision-pattern', dest="regex_match_revision", default=None, type=str, action='append',
                    help="The regular expression to search for in revision text. The regex must be surrounded by quotes.")

parser.add_argument('-RPl', '--revision-pattern-label', dest="regex_revision_label", default=None, type=str, action='append',
                    help="The label for the outputted column based on matching the regex in revision text.")

parser.add_argument('-CP', '--comment-pattern', dest="regex_match_comment", default=None, type=str, action='append',
                    help="The regular expression to search for in comments of revisions.")

parser.add_argument('-CPl', '--comment-pattern-label', dest="regex_comment_label", default=None, type=str, action='append',
                    help="The label for the outputted column based on matching the regex in comments.")

args = parser.parse_args()



# set persistence method

if args.persist is None:
    persist = PersistMethod.none
elif args.persist == "segment":
    persist = PersistMethod.segment
elif args.persist == "legacy":
    persist = PersistMethod.legacy
else:
    persist = PersistMethod.sequence

if args.namespace_filter is not None:
    namespaces = args.namespace_filter
else:
    namespaces = None

if len(args.dumpfiles) > 0:
    output_parquet = False
    for filename in args.dumpfiles:
        input_file = open_input_file(filename)

        # open directory for output
        if args.output_dir:
            output_dir = args.output_dir[0]
        else:
            output_dir = "."

        if output_dir.endswith(".parquet"):
            output_parquet = True

        print("Processing file: %s" % filename, file=sys.stderr)

        if args.stdout:
            output_file = sys.stdout
        else:
            filename = os.path.join(output_dir, os.path.basename(filename))
            output_file = get_output_filename(filename, parquet = output_parquet)

        wikiq = WikiqParser(input_file,
                            output_file,
                            collapse_user=args.collapse_user,
                            persist=persist,
                            urlencode=args.urlencode,
                            namespaces=namespaces,
                            revert_radius=args.revert_radius,
                            regex_match_revision = args.regex_match_revision,
                            regex_revision_label = args.regex_revision_label,
                            regex_match_comment = args.regex_match_comment,
                            regex_comment_label = args.regex_comment_label,
                            output_parquet=output_parquet)

        print(wikiq.output_parquet)
        wikiq.process()

        # close things 
        input_file.close()

else:
    wikiq = WikiqParser(sys.stdin,
                        sys.stdout,
                        collapse_user=args.collapse_user,
                        persist=persist,
                        #persist_legacy=args.persist_legacy,
                        urlencode=args.urlencode,
                        namespaces=namespaces,
                        revert_radius=args.revert_radius,
                        regex_match_revision = args.regex_match_revision,
                        regex_revision_label = args.regex_revision_label,
                        regex_match_comment = args.regex_match_comment,
                        regex_comment_label = args.regex_comment_label)

    wikiq.process() 

# stop_words = "a,able,about,across,after,all,almost,also,am,among,an,and,any,are,as,at,be,because,been,but,by,can,cannot,could,dear,did,do,does,either,else,ever,every,for,from,get,got,had,has,have,he,her,hers,him,his,how,however,i,if,in,into,is,it,its,just,least,let,like,likely,may,me,might,most,must,my,neither,no,nor,not,of,off,often,on,only,or,other,our,own,rather,said,say,says,she,should,since,so,some,than,that,the,their,them,then,there,these,they,this,tis,to,too,twas,us,wants,was,we,were,what,when,where,which,while,who,whom,why,will,with,would,yet,you,your"
# stop_words = stop_words.split(",")