adding pyproject.toml

[mediawiki_dump_tools.git] / wikiq

diff --git a/wikiq b/wikiq
index 8c51979ca5b697c73a78525a2b1c4fb14a016229..75c1af8045ad016eb4c574ba0a8ab3db6f96198b 100755 (executable)
--- a/wikiq
+++ b/wikiq
@@ -8,7 +8,7 @@ import argparse
 import sys
 import os, os.path
 import re
 import sys
 import os, os.path
 import re

-from datetime import datetime
+from datetime import datetime,timezone

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


@@ -26,7 +26,7 @@ from deltas import SequenceMatcher
 from deltas import SegmentMatcher
 
 import dataclasses as dc
 from deltas import SegmentMatcher
 
 import dataclasses as dc

-from dataclasses import dataclass, make_dataclass
+from dataclasses import dataclass

 import pyarrow as pa
 import pyarrow.parquet as pq
 
 import pyarrow as pa
 import pyarrow.parquet as pq
 


@@ -133,6 +133,11 @@ class WikiqPage():
         return next(self.__revisions)
 
 
         return next(self.__revisions)
 
 

+"""
+A RegexPair is defined by a regular expression (pattern) and a label.
+The pattern can include capture groups.  If it does then each capture group will have a resulting column in the output.
+If the pattern does not include a capture group, then only one output column will result.
+"""

 class RegexPair(object):
     def __init__(self, pattern, label):
         self.pattern = re.compile(pattern)
 class RegexPair(object):
     def __init__(self, pattern, label):
         self.pattern = re.compile(pattern)


@@ -141,6 +146,15 @@ class RegexPair(object):
         if self.has_groups:
             self.capture_groups = list(self.pattern.groupindex.keys())
             
         if self.has_groups:
             self.capture_groups = list(self.pattern.groupindex.keys())
             

+    def get_pyarrow_fields(self):
+        if self.has_groups:
+            fields = [pa.field(self._make_key(cap_group),pa.list_(pa.string()))
+                      for cap_group in self.capture_groups]
+        else:
+            fields = [pa.field(self.label, pa.list_(pa.string()))]
+
+        return fields
+

     def _make_key(self, cap_group):
         return ("{}_{}".format(self.label, cap_group))
 
     def _make_key(self, cap_group):
         return ("{}_{}".format(self.label, cap_group))
 


@@ -192,6 +206,22 @@ class RegexPair(object):
 
         return rev_data
 
 
         return rev_data
 

+"""
+
+We used to use a dictionary to collect fields for the output. 
+Now we use dataclasses. Compared to a dictionary, this should help:
+- prevent some bugs
+- make it easier to output parquet data. 
+- use class attribute '.' syntax instead of dictionary syntax. 
+- improve support for tooling (autocomplete, type hints)
+- use type information to define formatting rules
+
+Depending on the parameters passed into Wikiq, the output schema can be different. 
+Therefore, we need to end up constructing a dataclass with the correct output schema. 
+It also needs to have the correct pyarrow schema so we can write parquet files.
+
+The RevDataBase type has all the fields that will be output no matter how wikiq is invoked.
+"""

 @dataclass()
 class RevDataBase():
     revid: int 
 @dataclass()
 class RevDataBase():
     revid: int 


@@ -209,30 +239,39 @@ class RevDataBase():
     editor: str = None
     anon: bool = None
 
     editor: str = None
     anon: bool = None
 

+    # toggles url encoding. this isn't a dataclass field since it doesn't have a type annotation
+    urlencode = False
+
+    # defines pyarrow schema.
+    # each field in the data class needs an entry in this array.
+    # the names should match and be in the same order.
+    # this isn't a dataclass field since it doesn't have a type annotation

     pa_schema_fields = [
         pa.field("revid", pa.int64()),
     pa_schema_fields = [
         pa.field("revid", pa.int64()),

-        pa.field("date_time",pa.timestamp('ms')),
+        pa.field("date_time", pa.timestamp('ms')),

         pa.field("articleid",pa.int64()),
         pa.field("articleid",pa.int64()),

-        pa.field("editorid",pa.int64()),
+        pa.field("editorid",pa.int64(), nullable=True),

         pa.field("title",pa.string()),
         pa.field("namespace",pa.int32()),
         pa.field("deleted",pa.bool_()),
         pa.field("title",pa.string()),
         pa.field("namespace",pa.int32()),
         pa.field("deleted",pa.bool_()),

-        pa.field("test_chars",pa.int32()),
-        pa.field("revert",pa.bool_()),
-        pa.field("reverteds",pa.list_(pa.int64())),
+        pa.field("text_chars",pa.int32()),
+        pa.field("revert",pa.bool_(), nullable=True),
+        pa.field("reverteds",pa.list_(pa.int64()), nullable=True),

         pa.field("sha1",pa.string()),
         pa.field("minor",pa.bool_()),
         pa.field("editor",pa.string()),
         pa.field("anon",pa.bool_())
     ]
 
         pa.field("sha1",pa.string()),
         pa.field("minor",pa.bool_()),
         pa.field("editor",pa.string()),
         pa.field("anon",pa.bool_())
     ]
 

+    # pyarrow is a columnar format, so most of the work happens in the flush_parquet_buffer function

     def to_pyarrow(self):
         return dc.astuple(self)
 
     def to_pyarrow(self):
         return dc.astuple(self)
 

+    # logic to convert each field into the wikiq tsv format goes here.

     def to_tsv_row(self):
         
         row = []
     def to_tsv_row(self):
         
         row = []

-        for f in self.fields():
+        for f in dc.fields(self):

             val = getattr(self, f.name)
             if getattr(self, f.name) is None:
                 row.append("")
             val = getattr(self, f.name)
             if getattr(self, f.name) is None:
                 row.append("")


@@ -244,43 +283,47 @@ class RevDataBase():
 
             elif f.name in {'editor','title'}:
                 s = '"' + val + '"'
 
             elif f.name in {'editor','title'}:
                 s = '"' + val + '"'

-                if f.name in TO_ENCODE:
-                    row.append(quote(str(val)))
+                if self.urlencode and f.name in TO_ENCODE:
+                    row.append(quote(str(s)))
+                else:
+                    row.append(s)

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

-                if f.name in TO_ENCODE:
+                if self.urlencode and f.name in TO_ENCODE:

                     row.append(quote(str(val)))
                     row.append(quote(str(val)))

+                else:
+                    row.append(val)

             else:
                 row.append(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):
+        return '\t'.join(map(str,row))

 
 

-        
+    def header_row(self):
+        return '\t'.join(map(lambda f: f.name, dc.fields(self)))
+
+"""

 
 

+If collapse=True we'll use a RevDataCollapse dataclass.
+This class inherits from RevDataBase. This means that it has all the same fields and functions. 
+
+It just adds a new field and updates the pyarrow schema.
+
+"""

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

+"""
+
+If persistence data is to be computed we'll need the fields added by RevDataPersistence. 
+
+"""

 @dataclass()
 class RevDataPersistence(RevDataBase):
     token_revs:int = None
 @dataclass()
 class RevDataPersistence(RevDataBase):
     token_revs:int = None


@@ -296,6 +339,10 @@ class RevDataPersistence(RevDataBase):
         
     pa_schema_fields = RevDataBase.pa_schema_fields  + pa_persistence_schema_fields
 
         
     pa_schema_fields = RevDataBase.pa_schema_fields  + pa_persistence_schema_fields
 

+"""
+class RevDataCollapsePersistence uses multiple inheritence to make a class that has both persistence and collapse fields.
+
+"""

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


@@ -323,6 +370,9 @@ class WikiqParser():
         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)
 
         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)
 

+
+        # This is where we set the type for revdata.
+        

         if self.collapse_user is True:
             if self.persist == PersistMethod.none:
                 revdata_type = RevDataCollapse
         if self.collapse_user is True:
             if self.persist == PersistMethod.none:
                 revdata_type = RevDataCollapse


@@ -333,18 +383,23 @@ class WikiqParser():
         else:
             revdata_type = RevDataBase
 
         else:
             revdata_type = RevDataBase
 

+        # if there are regex fields, we need to add them to the revdata type.

         regex_fields = [(field.name, list[str], dc.field(default=None)) for field in self.regex_schemas]
 
         regex_fields = [(field.name, list[str], dc.field(default=None)) for field in self.regex_schemas]
 

-        self.revdata_type = make_dataclass('RevData_Parser',
-                                           fields=regex_fields,
-                                           bases=(revdata_type,))
+        # make_dataclass is a function that defines a new dataclass type.
+        # here we extend the type we have already chosen and add the regular expression types 
+        self.revdata_type = dc.make_dataclass('RevData_Parser',
+                                              fields=regex_fields,
+                                              bases=(revdata_type,))

         
         

+        # we also need to make sure that we have the right pyarrow schema

         self.revdata_type.pa_schema_fields = revdata_type.pa_schema_fields + self.regex_schemas
                         
         self.revdata_type.pa_schema_fields = revdata_type.pa_schema_fields + self.regex_schemas
                         

+        self.revdata_type.urlencode = self.urlencode

 
 

-        # print(list(map(lambda d: d.name, dc.fields(self.revdata_type))))
-        # print(self.revdata_type.pa_schema_fields)
+        self.schema = pa.schema(self.revdata_type.pa_schema_fields)

 
 

+        # here we initialize the variables we need for output.

         if output_parquet is True:
             self.output_parquet = True
             self.pq_writer = None
         if output_parquet is True:
             self.output_parquet = True
             self.pq_writer = None


@@ -352,29 +407,34 @@ class WikiqParser():
             self.parquet_buffer = []
             self.parquet_buffer_size = parquet_buffer_size
         else:
             self.parquet_buffer = []
             self.parquet_buffer_size = parquet_buffer_size
         else:

-            self.output_file = open(output_file,'w')
-
+            self.print_header = True
+            if output_file == sys.stdout:
+                
+                self.output_file = output_file
+            else:
+                self.output_file = open(output_file,'w')
+            self.output_parquet = False

 
     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):
 
     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())))
-
+                rp = RegexPair(pattern, label)
+                result.append(rp)
+                self.regex_schemas = self.regex_schemas + rp.get_pyarrow_fields()

             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.')
 
             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)
+    def matchmake_revision(self, rev, rev_data):
+        rev_data = self.matchmake_text(rev.text, rev_data)

         rev_data = self.matchmake_comment(rev.comment, rev_data)
         return rev_data
 
         rev_data = self.matchmake_comment(rev.comment, rev_data)
         return rev_data
 

-    def matchmake_revision(self, text, rev_data):
+    def matchmake_text(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(text, rev_data, self.regex_revision_pairs)
 
     def matchmake_comment(self, comment, rev_data):


@@ -431,7 +491,7 @@ class WikiqParser():
 
             if self.persist != PersistMethod.none:
                 window = deque(maxlen=PERSISTENCE_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)
                 if self.persist == PersistMethod.sequence:
                     state = mwpersistence.DiffState(SequenceMatcher(tokenizer = wikitext_split),
                                                     revert_radius=PERSISTENCE_RADIUS)


@@ -448,8 +508,9 @@ class WikiqParser():
             # Iterate through a page's revisions
             for rev in page:
                 
             # Iterate through a page's revisions
             for rev in page:
                 

+                # create a new data object instead of a dictionary. 

                 rev_data = self.revdata_type(revid = rev.id,
                 rev_data = self.revdata_type(revid = rev.id,

-                                             date_time = datetime.fromtimestamp(rev.timestamp.unix()),
+                                             date_time = datetime.fromtimestamp(rev.timestamp.unix(), tz=timezone.utc),

                                              articleid = page.id,
                                              editorid = "" if rev.deleted.user == True or rev.user.id is None else rev.user.id,
                                              title =  page.title,
                                              articleid = page.id,
                                              editorid = "" if rev.deleted.user == True or rev.user.id is None else rev.user.id,
                                              title =  page.title,


@@ -457,7 +518,7 @@ class WikiqParser():
                                              namespace = namespace
                                              )
 
                                              namespace = namespace
                                              )
 

-                rev_data = self.matchmake(rev, rev_data)
+                rev_data = self.matchmake_revision(rev, rev_data)

 
                 if not rev.deleted.text:
                     # rev.text can be None if the page has no text
 
                 if not rev.deleted.text:
                     # rev.text can be None if the page has no text


@@ -468,7 +529,6 @@ class WikiqParser():
                     if rev.sha1:
                         text_sha1 = rev.sha1
                     else:
                     if rev.sha1:
                         text_sha1 = rev.sha1
                     else:

-

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


@@ -491,8 +551,8 @@ class WikiqParser():
                 if not rev.deleted.user:
                     # wrap user-defined editors in quotes for fread
                     rev_data.editor = rev.user.text 
                 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
-                    
+                    rev_data.anon = rev.user.id is None
+                

                 #if re.match(r'^#redirect \[\[.*\]\]', rev.text, re.I):
                 #    redirect = True
                 #else:
                 #if re.match(r'^#redirect \[\[.*\]\]', rev.text, re.I):
                 #    redirect = True
                 #else:


@@ -504,8 +564,8 @@ class WikiqParser():
                 if self.collapse_user:
                     rev_data.collapsed_revs = rev.collapsed_revs
 
                 if self.collapse_user:
                     rev_data.collapsed_revs = rev.collapsed_revs
 

+                # get the 

                 if self.persist != PersistMethod.none:
                 if self.persist != PersistMethod.none:

-

                     if not rev.deleted.text:
 
                         if self.persist != PersistMethod.legacy:
                     if not rev.deleted.text:
 
                         if self.persist != PersistMethod.legacy:


@@ -521,12 +581,12 @@ class WikiqParser():
                             
                             num_token_revs, num_tokens = calculate_persistence(old_tokens_added)
 
                             
                             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
+                            old_rev_data.token_revs = num_token_revs
+                            old_rev_data.tokens_added = num_tokens
+                            old_rev_data.tokens_removed = len(old_tokens_removed)
+                            old_rev_data.tokens_window = PERSISTENCE_RADIUS-1

 
 

-                            self.print_rev_data(rev_data)
+                            self.print_rev_data(old_rev_data)

 
                 else:
                     self.print_rev_data(rev_data)
 
                 else:
                     self.print_rev_data(rev_data)


@@ -554,14 +614,19 @@ class WikiqParser():
         print("Done: %s revisions and %s pages." % (rev_count, page_count),
               file=sys.stderr)
 
         print("Done: %s revisions and %s pages." % (rev_count, page_count),
               file=sys.stderr)
 

+        # remember to flush the parquet_buffer if we're done

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

-            output_file.close()
+            self.output_file.close()

 
 
 
 

+    """
+    For performance reasons it's better to write parquet in batches instead of one row at a time.
+    So this function just puts the data on a buffer. If the buffer is full, then it gets flushed (written).
+    """

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


@@ -570,10 +635,16 @@ class WikiqParser():
             self.flush_parquet_buffer()
 
 
             self.flush_parquet_buffer()
 
 

+    """
+    Function that actually writes data to the parquet file. 
+    It needs to transpose the data from row-by-row to column-by-column
+    """

     def flush_parquet_buffer(self):
     def flush_parquet_buffer(self):

-        schema = pa.schema(self.revdata_type.pa_schema_fields)

 
 

-        def row_to_col(rg, types):
+        """
+        Returns the pyarrow table that we'll write
+        """
+        def rows_to_table(rg, schema):

             cols = []
             first = rg[0]
             for col in first:
             cols = []
             first = rg[0]
             for col in first:


@@ -584,17 +655,18 @@ class WikiqParser():
                     cols[j].append(row[j])
 
             arrays = []
                     cols[j].append(row[j])
 
             arrays = []

-            for col, typ in zip(cols, types):
+            for col, typ in zip(cols, schema.types):

                 arrays.append(pa.array(col, typ))
                 arrays.append(pa.array(col, typ))

-            return arrays
+            return pa.Table.from_arrays(arrays, schema=schema)

 
 

-        outtable = pa.Table.from_arrays(row_to_col(self.parquet_buffer, schema.types), schema=schema)
+        outtable = rows_to_table(self.parquet_buffer, self.schema)

         if self.pq_writer is None:
             self.pq_writer = pq.ParquetWriter(self.output_file, schema, flavor='spark')
 
         self.pq_writer.write_table(outtable)
         self.parquet_buffer = []
         
         if self.pq_writer is None:
             self.pq_writer = pq.ParquetWriter(self.output_file, schema, flavor='spark')
 
         self.pq_writer.write_table(outtable)
         self.parquet_buffer = []
         

+    # depending on if we are configured to write tsv or parquet, we'll call a different function.

     def print_rev_data(self, rev_data):
         if self.output_parquet is False:
             printfunc = self.write_tsv_row
     def print_rev_data(self, rev_data):
         if self.output_parquet is False:
             printfunc = self.write_tsv_row


@@ -604,7 +676,11 @@ class WikiqParser():
         printfunc(rev_data)
 
     def write_tsv_row(self, rev_data):
         printfunc(rev_data)
 
     def write_tsv_row(self, rev_data):

-        line = rev_data.to_tsv_line()
+        if self.print_header:
+            print(rev_data.header_row(), file=self.output_file)
+            self.print_header = False
+
+        line = rev_data.to_tsv_row()

         print(line, file=self.output_file)
 
 
         print(line, file=self.output_file)
 
 


@@ -737,7 +813,6 @@ if len(args.dumpfiles) > 0:
                             regex_comment_label = args.regex_comment_label,
                             output_parquet=output_parquet)
 
                             regex_comment_label = args.regex_comment_label,
                             output_parquet=output_parquet)
 

-        print(wikiq.output_parquet)

         wikiq.process()
 
         # close things 
         wikiq.process()
 
         # close things