collective/mediawiki_dump_tools
16
0
Fork 0
Code Issues Pull Requests Activity

Start interoperability between wikidiff2 and deltas

Browse Source 
The big challenges here (and remaining) are as follows:

1. Deltas requires changes to be given at the token level,
whereas wikidiff2 reports changes at the byte level. Thus,
it is often required to tokenize sequences of text to convert
to the desired token indices. As-is this is done inefficiently,
often requiring re-tokenization of previously-tokenized sequences.
A better implementation would incrementally tokenize, or
automatically find the referenced sequences.

2. Deltas only allows for Equal/Insert/Delete operations,
while wikidiff2 also detects paragraph moves. These paragraph
moves are NOT equivalent to Equal, as the moved paragraphs
are not guaranteed to be equivalent, just very similar.
Wikidiff2 does not report changes to moved paragraphs, so
to preserve token persistence, a difference algorithm
would need to be performed on the before/after sequences.
A stopgap (currently implemented) is to turn these
into strict deletions/insertions.

3. There appears to be a lot of memory consumption, and
sometimes this results in memory overflow. I am unsure
if this is a memory leak or simply that re-tokenizing
causes significant enough memory throughput that
my machine can't handle it.

4. Deltas expects all tokens in the before/after text to
be covered by segment ranges of Equal/Insert/Delete, but
wikidiff2 does not appear to ever emit any Equal ranges,
instead skipping them. These ranges must be computed
and inserted in sequence. As-is the code does not correctly
handle unchanged text at the end of pages.

Signed-off-by: Will Beason <willbeason@gmail.com>
This commit is contained in:
Will Beason 2025-06-26 16:08:50 -05:00
parent 1ec8bfaad4
commit bc7f186112
3 changed files with 243 additions and 43 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
test/Wikiq_Unit_Test.py
View File

@ -1,9 +1,7 @@
import shutil
import sys
import unittest
import os
import subprocess
from shutil import copyfile
import numpy as np
import pandas as pd
@ -106,8 +104,8 @@ class WikiqTestCase(unittest.TestCase):
tester = WikiqTester(IKWIKI, "noargs")
try:
tester.call_wikiq()
# tester.call_wikiq("--compute-incremental-diffs")
# tester.call_wikiq()
tester.call_wikiq("--wikidiff-url=http://localhost:8000")
except subprocess.CalledProcessError as exc:
self.fail(exc.stderr.decode("utf8"))

223
wiki_diff_matcher.py Normal file
View File

@ -0,0 +1,223 @@
import json
import sys
import requests
from deltas import tokenizers, RegexTokenizer, DiffEngine, Equal, Insert, Delete
TOKENIZER = tokenizers.text_split
def compute_diffs(url: str, texts: list[str]) -> list:
response = None
try:
response = requests.post(url, json=texts)
response.raise_for_status()
incremental_diffs = response.json()
except requests.exceptions.ConnectionError as e:
print(
f"Connection Error: Could not connect to the server at {url}. Make sure your local server is running.")
print(e)
raise e
except requests.exceptions.HTTPError as e:
print(f"HTTP Error: {e}")
if response is not None:
print(f"Response Body: {response.text}")
raise e
except requests.exceptions.JSONDecodeError as e:
# Must come before RequestException as JSONDecodeError is
# a subclass.
print(f"JSON Decode Error: {e}", file=sys.stderr)
if response is not None:
print(f"Response Body: {response.text}", file=sys.stderr)
raise e
except requests.exceptions.RequestException as e:
print(f"An unexpected error occurred: {e}")
raise e
return incremental_diffs
def to_operations(previous_text, next_text, diff, tokenizer: RegexTokenizer) -> list:
d = json.loads(diff)
# Keep track of the last difference we saw in order to notice unaccounted-for
# tokens. Each token at the end of "to" which is skipped for the next diff
# must be represented as an "Equal()" segment.
from_last_end = 0
to_last_end = 0
result = []
# DiffState expects differences to be represented in order from the
# result's perspective ("to"), not the previous text. Thus, if a line
# is moved earlier then its insertion should appear before its deletion.
# As a rule of thumb, the "to" segments should be non-overlapping and
# strictly increasing, while the "from" segments should merely be
# non-overlapping.
#
# wikidiff2 appears to follow this same convention, but this behavior
# is not documented.
for entry in d['diff']:
from_start_line = entry['offset']['from']
to_start_line = entry['offset']['to']
# Per above, to_start_line appears to be nondecreasing, but
# from_start_line may sometimes decrease for detected paragraph moves.
from_start_tokens = len(tokenizer.tokenize(previous_text[:from_start_line]))
to_start_tokens = len(tokenizer.tokenize(next_text[:to_start_line]))
# These constant calls to tokenizer.tokenize can definitely be optimized
# as tokenization is currently a bottleneck. Ideally tokenization would
# happen incrementally where possible, or somehow be cached, but this
# would be more complex.
if entry['type'] == 0:
# wikidiff2 doesn't appear to emit diffs of this type, but cover anyway.
line_tokens = len(tokenizer.tokenize(entry['text']))
from_end_tokens = from_start_tokens + line_tokens
to_end_tokens = to_start_tokens + line_tokens
result.append(Equal(from_start_tokens, from_end_tokens,
to_start_tokens, to_end_tokens))
from_last_end = from_end_tokens
to_last_end = to_end_tokens
continue
else:
# These do not appear to be generated by wikidiff2, and so must be
# inferred.
equal_tokens = to_start_tokens - to_last_end
# If we notice that the next non-zero segment (which must be a
# change, given that its type is non-zero), begins after the end
# of the previous segment, we must add an Equal segment.
# TODO: While the "to" token ranges are correct, the "from"
# ranges are likely not, particularly in histories with paragraph
# moves.
if equal_tokens > 0:
result.append(Equal(from_last_end, from_start_line,
to_last_end, to_start_line))
if entry['type'] == 1 or entry['type'] == 4:
# TODO: Separate out type 4 to recognize this is the insertion
# part of a paragraph move. Note that for paragraph moves
# the text is not necessarily identical, just similar.
line_tokens = len(tokenizer.tokenize(entry['text']))
to_end_tokens = to_start_tokens + line_tokens
result.append(Insert(from_start_tokens, from_start_tokens,
to_start_tokens, to_end_tokens,
))
# We have now used more of the "to" tokens.
to_last_end = to_end_tokens
elif entry['type'] == 2 or entry['type'] == 5:
# TODO: Separate out type 5 to recognize this is the deletion
# part of a paragraph move. Note that for paragraph moves
# the text is not necessarily identical, just similar.
line_tokens = len(tokenizer.tokenize(entry['text']))
from_end_tokens = from_start_tokens + line_tokens
result.append(Delete(from_start_tokens, from_end_tokens,
to_start_tokens, to_start_tokens,
))
# We have not used more of the "from" tokens.
from_last_end = from_end_tokens
elif entry['type'] == 3:
# The text field is an overlapping mix of both the previous and next
# lines, and so we can't directly tokenize it.
text = entry['text']
last_end = 0
previous_line = ""
next_line = ""
# A line will have one or more highlightRanges.
# It is not guaranteed that insertions/deletions are matched,
# for instance, if a word is deleted from the middle of a line.
for highlightRange in entry['highlightRanges']:
if highlightRange['start'] > last_end:
previous_line += text[last_end:highlightRange['start']]
next_line += text[last_end:highlightRange['start']]
# Add an Equal segment.
rangeStart = highlightRange['start']
rangeEnd = rangeStart + highlightRange['length']
if highlightRange['type'] == 0:
# Insertion
next_line += text[rangeStart:rangeEnd]
# Add an Insert segment.
elif highlightRange['type'] == 1:
# Deletion
previous_line += text[rangeStart:rangeEnd]
# Add a Delete segment.
else:
raise Exception(entry)
from_tokens = len(tokenizer.tokenize(previous_line))
to_tokens = len(tokenizer.tokenize(next_line))
from_start_tokens += from_tokens
to_start_tokens += to_tokens
else:
# The 'type' isn't one of the known
raise ValueError(d)
# TODO: Handle trailing tokens
# raise Exception(result)
return result
class WikiDiffMatcher:
def __init__(self,
url: str,
texts: list[str],
tokenizer: RegexTokenizer = None,
):
# Pre-compute diffs to reduce traffic overhead.
self.diffs = compute_diffs(url, texts)
self.tokenizer = tokenizer or TOKENIZER
class Processor(DiffEngine.Processor):
def __init__(self,
diffs,
tokenizer=None
):
self.diffs = iter(diffs)
self.tokenizer = tokenizer or TOKENIZER
self.last_tokens = []
self.previous_text = ""
def update(self, last_tokens):
self.last_tokens = last_tokens
def process(self, text, token_class=None):
# IDEs will report the method signature as incorrect, but this is
# expected. The DiffEngine.Processor class must be inherited from,
# and its process definition incorrectly excludes a "self" argument.
# The diff has already been computed, but we need to incrementally
# retrieve it to recreate the behavior DiffState expects.
diff = next(self.diffs)
tokens = self.tokenizer.tokenize(text, token_class=token_class)
operations = to_operations(self.previous_text, text, diff, self.tokenizer)
a = self.last_tokens
b = tokens
self.last_tokens = tokens
self.previous_text = text
return operations, a, b
def processor(self, *args, **kwargs):
return self.Processor(self.diffs, self.tokenizer)
def process(self):
# DiffState checks for this method even though it is not called.
raise Exception("Unnecessary implementation")

57
wikiq
View File

@ -17,7 +17,6 @@ from hashlib import sha1
from typing import Any, IO, TextIO, Generator, Union
import mwxml
import requests
from mwxml import Dump
from deltas.tokenizers import wikitext_split
@ -26,6 +25,7 @@ import mwreverts
import tables
from tables import RevisionTable
from wiki_diff_matcher import WikiDiffMatcher
TO_ENCODE = ('title', 'editor')
PERSISTENCE_RADIUS = 7
@ -43,6 +43,7 @@ class PersistMethod:
sequence = 1
segment = 2
legacy = 3
wikidiff = 4
def calculate_persistence(tokens_added):
@ -218,7 +219,7 @@ class WikiqParser:
revert_radius: int = 15,
output_parquet: bool = True,
parquet_buffer_size: int = 2000,
compute_incremental_diffs: bool = False,
wikidiff_url: str = "",
):
"""
@ -231,7 +232,7 @@ class WikiqParser:
self.persist: int = persist
self.namespaces = []
self.revert_radius = revert_radius
self.compute_incremental_diffs: bool = compute_incremental_diffs
self.wikidiff_url: str = wikidiff_url
if namespaces is not None:
self.namespace_filter = set(namespaces)
@ -367,9 +368,6 @@ class WikiqParser:
schema = schema.append(pa.field('tokens_removed', pa.int64(), nullable=True))
schema = schema.append(pa.field('tokens_window', pa.int64(), nullable=True))
if self.compute_incremental_diffs:
schema = schema.append(pa.field('incremental diffs', pa.string()))
if self.output_parquet:
writer = pq.ParquetWriter(self.output_file, schema, flavor='spark')
else:
@ -379,7 +377,7 @@ class WikiqParser:
# Iterate through pages
for page in dump:
payload = []
revision_texts = []
# skip namespaces not in the filter
if self.namespace_filter is not None:
@ -419,36 +417,10 @@ class WikiqParser:
regex_matches[k] = []
regex_matches[k].append(v)
if self.compute_incremental_diffs:
payload.append(rev.text)
revision_texts.append(rev.text)
# Collect the set of pages currently buffered in the table so we can run multi-page functions on them.
row_buffer = table.pop()
if self.compute_incremental_diffs:
try:
response = requests.post(DIFFS_URL, json=payload)
response.raise_for_status()
incremental_diffs = response.json()
except requests.exceptions.ConnectionError as e:
print(
f"Connection Error: Could not connect to the server at {DIFFS_URL}. Make sure your local server is running.")
print(e)
raise e
except requests.exceptions.HTTPError as e:
print(f"HTTP Error: {e}")
print(f"Response Body: {response.text}")
raise e
except requests.exceptions.JSONDecodeError as e:
# Must come before RequestException as JSONDecodeError is
# a subclass.
print(f"JSON Decode Error: {e}", file=sys.stderr)
print(f"Response Body: {response.text}", file=sys.stderr)
raise e
except requests.exceptions.RequestException as e:
print(f"An unexpected error occurred: {e}")
raise e
row_buffer['incremental diffs'] = incremental_diffs
is_revert_column: list[Union[bool, None]] = []
for r, d in zip(row_buffer['reverteds'], row_buffer['deleted']):
@ -477,6 +449,11 @@ class WikiqParser:
elif self.persist == PersistMethod.segment:
state = mwpersistence.DiffState(SegmentMatcher(tokenizer=wikitext_split),
revert_radius=PERSISTENCE_RADIUS)
elif self.persist == PersistMethod.wikidiff:
state = mwpersistence.DiffState(WikiDiffMatcher(self.wikidiff_url,
revision_texts,
tokenizer=wikitext_split),
revert_radius=PERSISTENCE_RADIUS)
else:
from mw.lib import persistence
state = persistence.State()
@ -614,20 +591,22 @@ def main():
action='store_true',
help="Whether the archive is from the fandom 2020 dumps by Wikiteam. These dumps can have multiple .xml files in their archives.")
parser.add_argument('--compute-incremental-diffs', dest="compute_incremental_diffs",
action='store_true',
help="Compute and store incremental diffs by edit session.")
parser.add_argument('--wikidiff-url', dest="wikidiff_url",
action='store',
help="The URL to a server running WikiDiff2.")
args = parser.parse_args()
# set persistence method
if args.persist is None:
if args.persist is None and not args.wikidiff_url:
persist = PersistMethod.none
elif args.persist == "segment":
persist = PersistMethod.segment
elif args.persist == "legacy":
persist = PersistMethod.legacy
elif args.wikidiff_url:
persist = PersistMethod.wikidiff
else:
persist = PersistMethod.sequence
@ -670,7 +649,7 @@ def main():
regex_match_comment=args.regex_match_comment,
regex_comment_label=args.regex_comment_label,
output_parquet=output_parquet,
compute_incremental_diffs=args.compute_incremental_diffs,
wikidiff_url=args.wikidiff_url,
)
wikiq.process()