up

Makefile

@@ -0,0 +1,9 @@
+check_dirs := .
+
+quality:
+	black --check $(check_dirs)
+	ruff $(check_dirs)
+
+style:
+	black $(check_dirs)
+	ruff $(check_dirs) --fix

app.py

@@ -0,0 +1,96 @@
+import numpy as np
+import unicodedata
+import diff_match_patch as dmp_module
+from enum import Enum
+import gradio as gr
+from datasets import load_dataset
+import pandas as pd
+from jiwer import process_words, wer_default
+
+
+class Action(Enum):
+    INSERTION = 1
+    DELETION = -1
+    EQUAL = 0
+
+
+def compare_string(text1: str, text2: str) -> list:
+    text1_normalized = unicodedata.normalize("NFKC", text1)
+    text2_normalized = unicodedata.normalize("NFKC", text2)
+
+    dmp = dmp_module.diff_match_patch()
+    diff = dmp.diff_main(text1_normalized, text2_normalized)
+    dmp.diff_cleanupSemantic(diff)
+
+    return diff
+
+
+def style_text(diff):
+    fullText = ""
+    for action, text in diff:
+        if action == Action.INSERTION.value:
+            fullText += f"<span style='background-color:Lightgreen'>{text}</span>"
+        elif action == Action.DELETION.value:
+            fullText += f"<span style='background-color:#FFCCCB'><s>{text}</s></span>"
+        elif action == Action.EQUAL.value:
+            fullText += f"{text}"
+        else:
+            raise Exception("Not Implemented")
+    fullText = fullText.replace("](", "]\(").replace("~", "\~")
+    return fullText
+
+
+dataset = load_dataset("distil-whisper/tedlium-long-form", split="validation")
+
+csv = pd.read_csv("assets/large-v2.csv")
+
+norm_target = csv["Norm Target"]
+norm_pred = csv["Norm Pred"]
+
+norm_target = [norm_target[i] for i in range(len(norm_target))]
+norm_pred = [norm_pred[i] for i in range(len(norm_pred))]
+
+target_dtype = np.int16
+max_range = np.iinfo(target_dtype).max
+
+
+def get_visualisation(idx):
+    audio = dataset[idx]["audio"]
+    array = (audio["array"] * max_range).astype(np.int16)
+    sampling_rate = audio["sampling_rate"]
+
+    text1 = norm_target[idx]
+    text2 = norm_pred[idx]
+
+    wer_output = process_words(text1, text2, wer_default, wer_default)
+    wer_percentage = 100 * wer_output.wer
+    num_insertions = wer_output.insertions
+
+    rel_length = len(text2.split()) / len(text1.split())
+
+    diff = compare_string(text1, text2)
+    full_text = style_text(diff)
+
+    return (sampling_rate, array), wer_percentage, num_insertions, rel_length, full_text
+
+
+if __name__ == "__main__":
+    with gr.Blocks() as demo:
+        slider = gr.Slider(
+            minimum=0, maximum=len(norm_target), step=1, label="Dataset sample"
+        )
+        btn = gr.Button("Analyse")
+        audio_out = gr.Audio(label="Audio input")
+        with gr.Row():
+            wer = gr.Number(label="WER")
+            insertions = gr.Number(label="Insertions")
+            relative_length = gr.Number(label="Relative length of target / reference")
+        text_out = gr.Markdown(label="Text difference")
+
+        btn.click(
+            fn=get_visualisation,
+            inputs=slider,
+            outputs=[audio_out, wer, insertions, relative_length, text_out],
+        )
+
+    demo.launch()

diff_match_patch.py

@@ -0,0 +1,2019 @@
+#!/usr/bin/python3
+
+"""Diff Match and Patch
+Copyright 2018 The diff-match-patch Authors.
+https://github.com/google/diff-match-patch
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+  http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+"""
+
+"""Functions for diff, match and patch.
+
+Computes the difference between two texts to create a patch.
+Applies the patch onto another text, allowing for errors.
+"""
+
+__author__ = "fraser@google.com (Neil Fraser)"
+
+import re
+import sys
+import time
+import urllib.parse
+
+
+class diff_match_patch:
+    """Class containing the diff, match and patch methods.
+
+    Also contains the behaviour settings.
+    """
+
+    def __init__(self):
+        """Inits a diff_match_patch object with default settings.
+        Redefine these in your program to override the defaults.
+        """
+
+        # Number of seconds to map a diff before giving up (0 for infinity).
+        self.Diff_Timeout = 1.0
+        # Cost of an empty edit operation in terms of edit characters.
+        self.Diff_EditCost = 4
+        # At what point is no match declared (0.0 = perfection, 1.0 = very loose).
+        self.Match_Threshold = 0.5
+        # How far to search for a match (0 = exact location, 1000+ = broad match).
+        # A match this many characters away from the expected location will add
+        # 1.0 to the score (0.0 is a perfect match).
+        self.Match_Distance = 1000
+        # When deleting a large block of text (over ~64 characters), how close do
+        # the contents have to be to match the expected contents. (0.0 = perfection,
+        # 1.0 = very loose).  Note that Match_Threshold controls how closely the
+        # end points of a delete need to match.
+        self.Patch_DeleteThreshold = 0.5
+        # Chunk size for context length.
+        self.Patch_Margin = 4
+
+        # The number of bits in an int.
+        # Python has no maximum, thus to disable patch splitting set to 0.
+        # However to avoid long patches in certain pathological cases, use 32.
+        # Multiple short patches (using native ints) are much faster than long ones.
+        self.Match_MaxBits = 32
+
+    #  DIFF FUNCTIONS
+
+    # The data structure representing a diff is an array of tuples:
+    # [(DIFF_DELETE, "Hello"), (DIFF_INSERT, "Goodbye"), (DIFF_EQUAL, " world.")]
+    # which means: delete "Hello", add "Goodbye" and keep " world."
+    DIFF_DELETE = -1
+    DIFF_INSERT = 1
+    DIFF_EQUAL = 0
+
+    def diff_main(self, text1, text2, checklines=True, deadline=None):
+        """Find the differences between two texts.  Simplifies the problem by
+          stripping any common prefix or suffix off the texts before diffing.
+
+        Args:
+          text1: Old string to be diffed.
+          text2: New string to be diffed.
+          checklines: Optional speedup flag.  If present and false, then don't run
+            a line-level diff first to identify the changed areas.
+            Defaults to true, which does a faster, slightly less optimal diff.
+          deadline: Optional time when the diff should be complete by.  Used
+            internally for recursive calls.  Users should set DiffTimeout instead.
+
+        Returns:
+          Array of changes.
+        """
+        # Set a deadline by which time the diff must be complete.
+        if deadline is None:
+            # Unlike in most languages, Python counts time in seconds.
+            if self.Diff_Timeout <= 0:
+                deadline = sys.maxsize
+            else:
+                deadline = time.time() + self.Diff_Timeout
+
+        # Check for null inputs.
+        if text1 is None or text2 is None:
+            raise ValueError("Null inputs. (diff_main)")
+
+        # Check for equality (speedup).
+        if text1 == text2:
+            if text1:
+                return [(self.DIFF_EQUAL, text1)]
+            return []
+
+        # Trim off common prefix (speedup).
+        commonlength = self.diff_commonPrefix(text1, text2)
+        commonprefix = text1[:commonlength]
+        text1 = text1[commonlength:]
+        text2 = text2[commonlength:]
+
+        # Trim off common suffix (speedup).
+        commonlength = self.diff_commonSuffix(text1, text2)
+        if commonlength == 0:
+            commonsuffix = ""
+        else:
+            commonsuffix = text1[-commonlength:]
+            text1 = text1[:-commonlength]
+            text2 = text2[:-commonlength]
+
+        # Compute the diff on the middle block.
+        diffs = self.diff_compute(text1, text2, checklines, deadline)
+
+        # Restore the prefix and suffix.
+        if commonprefix:
+            diffs[:0] = [(self.DIFF_EQUAL, commonprefix)]
+        if commonsuffix:
+            diffs.append((self.DIFF_EQUAL, commonsuffix))
+        self.diff_cleanupMerge(diffs)
+        return diffs
+
+    def diff_compute(self, text1, text2, checklines, deadline):
+        """Find the differences between two texts.  Assumes that the texts do not
+          have any common prefix or suffix.
+
+        Args:
+          text1: Old string to be diffed.
+          text2: New string to be diffed.
+          checklines: Speedup flag.  If false, then don't run a line-level diff
+            first to identify the changed areas.
+            If true, then run a faster, slightly less optimal diff.
+          deadline: Time when the diff should be complete by.
+
+        Returns:
+          Array of changes.
+        """
+        if not text1:
+            # Just add some text (speedup).
+            return [(self.DIFF_INSERT, text2)]
+
+        if not text2:
+            # Just delete some text (speedup).
+            return [(self.DIFF_DELETE, text1)]
+
+        if len(text1) > len(text2):
+            (longtext, shorttext) = (text1, text2)
+        else:
+            (shorttext, longtext) = (text1, text2)
+        i = longtext.find(shorttext)
+        if i != -1:
+            # Shorter text is inside the longer text (speedup).
+            diffs = [
+                (self.DIFF_INSERT, longtext[:i]),
+                (self.DIFF_EQUAL, shorttext),
+                (self.DIFF_INSERT, longtext[i + len(shorttext) :]),
+            ]
+            # Swap insertions for deletions if diff is reversed.
+            if len(text1) > len(text2):
+                diffs[0] = (self.DIFF_DELETE, diffs[0][1])
+                diffs[2] = (self.DIFF_DELETE, diffs[2][1])
+            return diffs
+
+        if len(shorttext) == 1:
+            # Single character string.
+            # After the previous speedup, the character can't be an equality.
+            return [(self.DIFF_DELETE, text1), (self.DIFF_INSERT, text2)]
+
+        # Check to see if the problem can be split in two.
+        hm = self.diff_halfMatch(text1, text2)
+        if hm:
+            # A half-match was found, sort out the return data.
+            (text1_a, text1_b, text2_a, text2_b, mid_common) = hm
+            # Send both pairs off for separate processing.
+            diffs_a = self.diff_main(text1_a, text2_a, checklines, deadline)
+            diffs_b = self.diff_main(text1_b, text2_b, checklines, deadline)
+            # Merge the results.
+            return diffs_a + [(self.DIFF_EQUAL, mid_common)] + diffs_b
+
+        if checklines and len(text1) > 100 and len(text2) > 100:
+            return self.diff_lineMode(text1, text2, deadline)
+
+        return self.diff_bisect(text1, text2, deadline)
+
+    def diff_lineMode(self, text1, text2, deadline):
+        """Do a quick line-level diff on both strings, then rediff the parts for
+          greater accuracy.
+          This speedup can produce non-minimal diffs.
+
+        Args:
+          text1: Old string to be diffed.
+          text2: New string to be diffed.
+          deadline: Time when the diff should be complete by.
+
+        Returns:
+          Array of changes.
+        """
+
+        # Scan the text on a line-by-line basis first.
+        (text1, text2, linearray) = self.diff_linesToChars(text1, text2)
+
+        diffs = self.diff_main(text1, text2, False, deadline)
+
+        # Convert the diff back to original text.
+        self.diff_charsToLines(diffs, linearray)
+        # Eliminate freak matches (e.g. blank lines)
+        self.diff_cleanupSemantic(diffs)
+
+        # Rediff any replacement blocks, this time character-by-character.
+        # Add a dummy entry at the end.
+        diffs.append((self.DIFF_EQUAL, ""))
+        pointer = 0
+        count_delete = 0
+        count_insert = 0
+        text_delete = ""
+        text_insert = ""
+        while pointer < len(diffs):
+            if diffs[pointer][0] == self.DIFF_INSERT:
+                count_insert += 1
+                text_insert += diffs[pointer][1]
+            elif diffs[pointer][0] == self.DIFF_DELETE:
+                count_delete += 1
+                text_delete += diffs[pointer][1]
+            elif diffs[pointer][0] == self.DIFF_EQUAL:
+                # Upon reaching an equality, check for prior redundancies.
+                if count_delete >= 1 and count_insert >= 1:
+                    # Delete the offending records and add the merged ones.
+                    subDiff = self.diff_main(text_delete, text_insert, False, deadline)
+                    diffs[pointer - count_delete - count_insert : pointer] = subDiff
+                    pointer = pointer - count_delete - count_insert + len(subDiff)
+                count_insert = 0
+                count_delete = 0
+                text_delete = ""
+                text_insert = ""
+
+            pointer += 1
+
+        diffs.pop()  # Remove the dummy entry at the end.
+
+        return diffs
+
+    def diff_bisect(self, text1, text2, deadline):
+        """Find the 'middle snake' of a diff, split the problem in two
+          and return the recursively constructed diff.
+          See Myers 1986 paper: An O(ND) Difference Algorithm and Its Variations.
+
+        Args:
+          text1: Old string to be diffed.
+          text2: New string to be diffed.
+          deadline: Time at which to bail if not yet complete.
+
+        Returns:
+          Array of diff tuples.
+        """
+
+        # Cache the text lengths to prevent multiple calls.
+        text1_length = len(text1)
+        text2_length = len(text2)
+        max_d = (text1_length + text2_length + 1) // 2
+        v_offset = max_d
+        v_length = 2 * max_d
+        v1 = [-1] * v_length
+        v1[v_offset + 1] = 0
+        v2 = v1[:]
+        delta = text1_length - text2_length
+        # If the total number of characters is odd, then the front path will
+        # collide with the reverse path.
+        front = delta % 2 != 0
+        # Offsets for start and end of k loop.
+        # Prevents mapping of space beyond the grid.
+        k1start = 0
+        k1end = 0
+        k2start = 0
+        k2end = 0
+        for d in range(max_d):
+            # Bail out if deadline is reached.
+            if time.time() > deadline:
+                break
+
+            # Walk the front path one step.
+            for k1 in range(-d + k1start, d + 1 - k1end, 2):
+                k1_offset = v_offset + k1
+                if k1 == -d or (k1 != d and v1[k1_offset - 1] < v1[k1_offset + 1]):
+                    x1 = v1[k1_offset + 1]
+                else:
+                    x1 = v1[k1_offset - 1] + 1
+                y1 = x1 - k1
+                while (
+                    x1 < text1_length and y1 < text2_length and text1[x1] == text2[y1]
+                ):
+                    x1 += 1
+                    y1 += 1
+                v1[k1_offset] = x1
+                if x1 > text1_length:
+                    # Ran off the right of the graph.
+                    k1end += 2
+                elif y1 > text2_length:
+                    # Ran off the bottom of the graph.
+                    k1start += 2
+                elif front:
+                    k2_offset = v_offset + delta - k1
+                    if k2_offset >= 0 and k2_offset < v_length and v2[k2_offset] != -1:
+                        # Mirror x2 onto top-left coordinate system.
+                        x2 = text1_length - v2[k2_offset]
+                        if x1 >= x2:
+                            # Overlap detected.
+                            return self.diff_bisectSplit(text1, text2, x1, y1, deadline)
+
+            # Walk the reverse path one step.
+            for k2 in range(-d + k2start, d + 1 - k2end, 2):
+                k2_offset = v_offset + k2
+                if k2 == -d or (k2 != d and v2[k2_offset - 1] < v2[k2_offset + 1]):
+                    x2 = v2[k2_offset + 1]
+                else:
+                    x2 = v2[k2_offset - 1] + 1
+                y2 = x2 - k2
+                while (
+                    x2 < text1_length
+                    and y2 < text2_length
+                    and text1[-x2 - 1] == text2[-y2 - 1]
+                ):
+                    x2 += 1
+                    y2 += 1
+                v2[k2_offset] = x2
+                if x2 > text1_length:
+                    # Ran off the left of the graph.
+                    k2end += 2
+                elif y2 > text2_length:
+                    # Ran off the top of the graph.
+                    k2start += 2
+                elif not front:
+                    k1_offset = v_offset + delta - k2
+                    if k1_offset >= 0 and k1_offset < v_length and v1[k1_offset] != -1:
+                        x1 = v1[k1_offset]
+                        y1 = v_offset + x1 - k1_offset
+                        # Mirror x2 onto top-left coordinate system.
+                        x2 = text1_length - x2
+                        if x1 >= x2:
+                            # Overlap detected.
+                            return self.diff_bisectSplit(text1, text2, x1, y1, deadline)
+
+        # Diff took too long and hit the deadline or
+        # number of diffs equals number of characters, no commonality at all.
+        return [(self.DIFF_DELETE, text1), (self.DIFF_INSERT, text2)]
+
+    def diff_bisectSplit(self, text1, text2, x, y, deadline):
+        """Given the location of the 'middle snake', split the diff in two parts
+        and recurse.
+
+        Args:
+          text1: Old string to be diffed.
+          text2: New string to be diffed.
+          x: Index of split point in text1.
+          y: Index of split point in text2.
+          deadline: Time at which to bail if not yet complete.
+
+        Returns:
+          Array of diff tuples.
+        """
+        text1a = text1[:x]
+        text2a = text2[:y]
+        text1b = text1[x:]
+        text2b = text2[y:]
+
+        # Compute both diffs serially.
+        diffs = self.diff_main(text1a, text2a, False, deadline)
+        diffsb = self.diff_main(text1b, text2b, False, deadline)
+
+        return diffs + diffsb
+
+    def diff_linesToChars(self, text1, text2):
+        """Split two texts into an array of strings.  Reduce the texts to a string
+        of hashes where each Unicode character represents one line.
+
+        Args:
+          text1: First string.
+          text2: Second string.
+
+        Returns:
+          Three element tuple, containing the encoded text1, the encoded text2 and
+          the array of unique strings.  The zeroth element of the array of unique
+          strings is intentionally blank.
+        """
+        lineArray = []  # e.g. lineArray[4] == "Hello\n"
+        lineHash = {}  # e.g. lineHash["Hello\n"] == 4
+
+        # "\x00" is a valid character, but various debuggers don't like it.
+        # So we'll insert a junk entry to avoid generating a null character.
+        lineArray.append("")
+
+        def diff_linesToCharsMunge(text):
+            """Split a text into an array of strings.  Reduce the texts to a string
+            of hashes where each Unicode character represents one line.
+            Modifies linearray and linehash through being a closure.
+
+            Args:
+              text: String to encode.
+
+            Returns:
+              Encoded string.
+            """
+            chars = []
+            # Walk the text, pulling out a substring for each line.
+            # text.split('\n') would would temporarily double our memory footprint.
+            # Modifying text would create many large strings to garbage collect.
+            lineStart = 0
+            lineEnd = -1
+            while lineEnd < len(text) - 1:
+                lineEnd = text.find("\n", lineStart)
+                if lineEnd == -1:
+                    lineEnd = len(text) - 1
+                line = text[lineStart : lineEnd + 1]
+
+                if line in lineHash:
+                    chars.append(chr(lineHash[line]))
+                else:
+                    if len(lineArray) == maxLines:
+                        # Bail out at 1114111 because chr(1114112) throws.
+                        line = text[lineStart:]
+                        lineEnd = len(text)
+                    lineArray.append(line)
+                    lineHash[line] = len(lineArray) - 1
+                    chars.append(chr(len(lineArray) - 1))
+                lineStart = lineEnd + 1
+            return "".join(chars)
+
+        # Allocate 2/3rds of the space for text1, the rest for text2.
+        maxLines = 666666
+        chars1 = diff_linesToCharsMunge(text1)
+        maxLines = 1114111
+        chars2 = diff_linesToCharsMunge(text2)
+        return (chars1, chars2, lineArray)
+
+    def diff_charsToLines(self, diffs, lineArray):
+        """Rehydrate the text in a diff from a string of line hashes to real lines
+        of text.
+
+        Args:
+          diffs: Array of diff tuples.
+          lineArray: Array of unique strings.
+        """
+        for i in range(len(diffs)):
+            text = []
+            for char in diffs[i][1]:
+                text.append(lineArray[ord(char)])
+            diffs[i] = (diffs[i][0], "".join(text))
+
+    def diff_commonPrefix(self, text1, text2):
+        """Determine the common prefix of two strings.
+
+        Args:
+          text1: First string.
+          text2: Second string.
+
+        Returns:
+          The number of characters common to the start of each string.
+        """
+        # Quick check for common null cases.
+        if not text1 or not text2 or text1[0] != text2[0]:
+            return 0
+        # Binary search.
+        # Performance analysis: https://neil.fraser.name/news/2007/10/09/
+        pointermin = 0
+        pointermax = min(len(text1), len(text2))
+        pointermid = pointermax
+        pointerstart = 0
+        while pointermin < pointermid:
+            if text1[pointerstart:pointermid] == text2[pointerstart:pointermid]:
+                pointermin = pointermid
+                pointerstart = pointermin
+            else:
+                pointermax = pointermid
+            pointermid = (pointermax - pointermin) // 2 + pointermin
+        return pointermid
+
+    def diff_commonSuffix(self, text1, text2):
+        """Determine the common suffix of two strings.
+
+        Args:
+          text1: First string.
+          text2: Second string.
+
+        Returns:
+          The number of characters common to the end of each string.
+        """
+        # Quick check for common null cases.
+        if not text1 or not text2 or text1[-1] != text2[-1]:
+            return 0
+        # Binary search.
+        # Performance analysis: https://neil.fraser.name/news/2007/10/09/
+        pointermin = 0
+        pointermax = min(len(text1), len(text2))
+        pointermid = pointermax
+        pointerend = 0
+        while pointermin < pointermid:
+            if (
+                text1[-pointermid : len(text1) - pointerend]
+                == text2[-pointermid : len(text2) - pointerend]
+            ):
+                pointermin = pointermid
+                pointerend = pointermin
+            else:
+                pointermax = pointermid
+            pointermid = (pointermax - pointermin) // 2 + pointermin
+        return pointermid
+
+    def diff_commonOverlap(self, text1, text2):
+        """Determine if the suffix of one string is the prefix of another.
+
+        Args:
+          text1 First string.
+          text2 Second string.
+
+        Returns:
+          The number of characters common to the end of the first
+          string and the start of the second string.
+        """
+        # Cache the text lengths to prevent multiple calls.
+        text1_length = len(text1)
+        text2_length = len(text2)
+        # Eliminate the null case.
+        if text1_length == 0 or text2_length == 0:
+            return 0
+        # Truncate the longer string.
+        if text1_length > text2_length:
+            text1 = text1[-text2_length:]
+        elif text1_length < text2_length:
+            text2 = text2[:text1_length]
+        text_length = min(text1_length, text2_length)
+        # Quick check for the worst case.
+        if text1 == text2:
+            return text_length
+
+        # Start by looking for a single character match
+        # and increase length until no match is found.
+        # Performance analysis: https://neil.fraser.name/news/2010/11/04/
+        best = 0
+        length = 1
+        while True:
+            pattern = text1[-length:]
+            found = text2.find(pattern)
+            if found == -1:
+                return best
+            length += found
+            if found == 0 or text1[-length:] == text2[:length]:
+                best = length
+                length += 1
+
+    def diff_halfMatch(self, text1, text2):
+        """Do the two texts share a substring which is at least half the length of
+        the longer text?
+        This speedup can produce non-minimal diffs.
+
+        Args:
+          text1: First string.
+          text2: Second string.
+
+        Returns:
+          Five element Array, containing the prefix of text1, the suffix of text1,
+          the prefix of text2, the suffix of text2 and the common middle.  Or None
+          if there was no match.
+        """
+        if self.Diff_Timeout <= 0:
+            # Don't risk returning a non-optimal diff if we have unlimited time.
+            return None
+        if len(text1) > len(text2):
+            (longtext, shorttext) = (text1, text2)
+        else:
+            (shorttext, longtext) = (text1, text2)
+        if len(longtext) < 4 or len(shorttext) * 2 < len(longtext):
+            return None  # Pointless.
+
+        def diff_halfMatchI(longtext, shorttext, i):
+            """Does a substring of shorttext exist within longtext such that the
+            substring is at least half the length of longtext?
+            Closure, but does not reference any external variables.
+
+            Args:
+              longtext: Longer string.
+              shorttext: Shorter string.
+              i: Start index of quarter length substring within longtext.
+
+            Returns:
+              Five element Array, containing the prefix of longtext, the suffix of
+              longtext, the prefix of shorttext, the suffix of shorttext and the
+              common middle.  Or None if there was no match.
+            """
+            seed = longtext[i : i + len(longtext) // 4]
+            best_common = ""
+            j = shorttext.find(seed)
+            while j != -1:
+                prefixLength = self.diff_commonPrefix(longtext[i:], shorttext[j:])
+                suffixLength = self.diff_commonSuffix(longtext[:i], shorttext[:j])
+                if len(best_common) < suffixLength + prefixLength:
+                    best_common = (
+                        shorttext[j - suffixLength : j]
+                        + shorttext[j : j + prefixLength]
+                    )
+                    best_longtext_a = longtext[: i - suffixLength]
+                    best_longtext_b = longtext[i + prefixLength :]
+                    best_shorttext_a = shorttext[: j - suffixLength]
+                    best_shorttext_b = shorttext[j + prefixLength :]
+                j = shorttext.find(seed, j + 1)
+
+            if len(best_common) * 2 >= len(longtext):
+                return (
+                    best_longtext_a,
+                    best_longtext_b,
+                    best_shorttext_a,
+                    best_shorttext_b,
+                    best_common,
+                )
+            else:
+                return None
+
+        # First check if the second quarter is the seed for a half-match.
+        hm1 = diff_halfMatchI(longtext, shorttext, (len(longtext) + 3) // 4)
+        # Check again based on the third quarter.
+        hm2 = diff_halfMatchI(longtext, shorttext, (len(longtext) + 1) // 2)
+        if not hm1 and not hm2:
+            return None
+        elif not hm2:
+            hm = hm1
+        elif not hm1:
+            hm = hm2
+        else:
+            # Both matched.  Select the longest.
+            if len(hm1[4]) > len(hm2[4]):
+                hm = hm1
+            else:
+                hm = hm2
+
+        # A half-match was found, sort out the return data.
+        if len(text1) > len(text2):
+            (text1_a, text1_b, text2_a, text2_b, mid_common) = hm
+        else:
+            (text2_a, text2_b, text1_a, text1_b, mid_common) = hm
+        return (text1_a, text1_b, text2_a, text2_b, mid_common)
+
+    def diff_cleanupSemantic(self, diffs):
+        """Reduce the number of edits by eliminating semantically trivial
+        equalities.
+
+        Args:
+          diffs: Array of diff tuples.
+        """
+        changes = False
+        equalities = []  # Stack of indices where equalities are found.
+        lastEquality = None  # Always equal to diffs[equalities[-1]][1]
+        pointer = 0  # Index of current position.
+        # Number of chars that changed prior to the equality.
+        length_insertions1, length_deletions1 = 0, 0
+        # Number of chars that changed after the equality.
+        length_insertions2, length_deletions2 = 0, 0
+        while pointer < len(diffs):
+            if diffs[pointer][0] == self.DIFF_EQUAL:  # Equality found.
+                equalities.append(pointer)
+                length_insertions1, length_insertions2 = length_insertions2, 0
+                length_deletions1, length_deletions2 = length_deletions2, 0
+                lastEquality = diffs[pointer][1]
+            else:  # An insertion or deletion.
+                if diffs[pointer][0] == self.DIFF_INSERT:
+                    length_insertions2 += len(diffs[pointer][1])
+                else:
+                    length_deletions2 += len(diffs[pointer][1])
+                # Eliminate an equality that is smaller or equal to the edits on both
+                # sides of it.
+                if (
+                    lastEquality
+                    and (
+                        len(lastEquality) <= max(length_insertions1, length_deletions1)
+                    )
+                    and (
+                        len(lastEquality) <= max(length_insertions2, length_deletions2)
+                    )
+                ):
+                    # Duplicate record.
+                    diffs.insert(equalities[-1], (self.DIFF_DELETE, lastEquality))
+                    # Change second copy to insert.
+                    diffs[equalities[-1] + 1] = (
+                        self.DIFF_INSERT,
+                        diffs[equalities[-1] + 1][1],
+                    )
+                    # Throw away the equality we just deleted.
+                    equalities.pop()
+                    # Throw away the previous equality (it needs to be reevaluated).
+                    if len(equalities):
+                        equalities.pop()
+                    if len(equalities):
+                        pointer = equalities[-1]
+                    else:
+                        pointer = -1
+                    # Reset the counters.
+                    length_insertions1, length_deletions1 = 0, 0
+                    length_insertions2, length_deletions2 = 0, 0
+                    lastEquality = None
+                    changes = True
+            pointer += 1
+
+        # Normalize the diff.
+        if changes:
+            self.diff_cleanupMerge(diffs)
+        self.diff_cleanupSemanticLossless(diffs)
+
+        # Find any overlaps between deletions and insertions.
+        # e.g: <del>abcxxx</del><ins>xxxdef</ins>
+        #   -> <del>abc</del>xxx<ins>def</ins>
+        # e.g: <del>xxxabc</del><ins>defxxx</ins>
+        #   -> <ins>def</ins>xxx<del>abc</del>
+        # Only extract an overlap if it is as big as the edit ahead or behind it.
+        pointer = 1
+        while pointer < len(diffs):
+            if (
+                diffs[pointer - 1][0] == self.DIFF_DELETE
+                and diffs[pointer][0] == self.DIFF_INSERT
+            ):
+                deletion = diffs[pointer - 1][1]
+                insertion = diffs[pointer][1]
+                overlap_length1 = self.diff_commonOverlap(deletion, insertion)
+                overlap_length2 = self.diff_commonOverlap(insertion, deletion)
+                if overlap_length1 >= overlap_length2:
+                    if (
+                        overlap_length1 >= len(deletion) / 2.0
+                        or overlap_length1 >= len(insertion) / 2.0
+                    ):
+                        # Overlap found.  Insert an equality and trim the surrounding edits.
+                        diffs.insert(
+                            pointer, (self.DIFF_EQUAL, insertion[:overlap_length1])
+                        )
+                        diffs[pointer - 1] = (
+                            self.DIFF_DELETE,
+                            deletion[: len(deletion) - overlap_length1],
+                        )
+                        diffs[pointer + 1] = (
+                            self.DIFF_INSERT,
+                            insertion[overlap_length1:],
+                        )
+                        pointer += 1
+                else:
+                    if (
+                        overlap_length2 >= len(deletion) / 2.0
+                        or overlap_length2 >= len(insertion) / 2.0
+                    ):
+                        # Reverse overlap found.
+                        # Insert an equality and swap and trim the surrounding edits.
+                        diffs.insert(
+                            pointer, (self.DIFF_EQUAL, deletion[:overlap_length2])
+                        )
+                        diffs[pointer - 1] = (
+                            self.DIFF_INSERT,
+                            insertion[: len(insertion) - overlap_length2],
+                        )
+                        diffs[pointer + 1] = (
+                            self.DIFF_DELETE,
+                            deletion[overlap_length2:],
+                        )
+                        pointer += 1
+                pointer += 1
+            pointer += 1
+
+    def diff_cleanupSemanticLossless(self, diffs):
+        """Look for single edits surrounded on both sides by equalities
+        which can be shifted sideways to align the edit to a word boundary.
+        e.g: The c<ins>at c</ins>ame. -> The <ins>cat </ins>came.
+
+        Args:
+          diffs: Array of diff tuples.
+        """
+
+        def diff_cleanupSemanticScore(one, two):
+            """Given two strings, compute a score representing whether the
+            internal boundary falls on logical boundaries.
+            Scores range from 6 (best) to 0 (worst).
+            Closure, but does not reference any external variables.
+
+            Args:
+              one: First string.
+              two: Second string.
+
+            Returns:
+              The score.
+            """
+            if not one or not two:
+                # Edges are the best.
+                return 6
+
+            # Each port of this function behaves slightly differently due to
+            # subtle differences in each language's definition of things like
+            # 'whitespace'.  Since this function's purpose is largely cosmetic,
+            # the choice has been made to use each language's native features
+            # rather than force total conformity.
+            char1 = one[-1]
+            char2 = two[0]
+            nonAlphaNumeric1 = not char1.isalnum()
+            nonAlphaNumeric2 = not char2.isalnum()
+            whitespace1 = nonAlphaNumeric1 and char1.isspace()
+            whitespace2 = nonAlphaNumeric2 and char2.isspace()
+            lineBreak1 = whitespace1 and (char1 == "\r" or char1 == "\n")
+            lineBreak2 = whitespace2 and (char2 == "\r" or char2 == "\n")
+            blankLine1 = lineBreak1 and self.BLANKLINEEND.search(one)
+            blankLine2 = lineBreak2 and self.BLANKLINESTART.match(two)
+
+            if blankLine1 or blankLine2:
+                # Five points for blank lines.
+                return 5
+            elif lineBreak1 or lineBreak2:
+                # Four points for line breaks.
+                return 4
+            elif nonAlphaNumeric1 and not whitespace1 and whitespace2:
+                # Three points for end of sentences.
+                return 3
+            elif whitespace1 or whitespace2:
+                # Two points for whitespace.
+                return 2
+            elif nonAlphaNumeric1 or nonAlphaNumeric2:
+                # One point for non-alphanumeric.
+                return 1
+            return 0
+
+        pointer = 1
+        # Intentionally ignore the first and last element (don't need checking).
+        while pointer < len(diffs) - 1:
+            if (
+                diffs[pointer - 1][0] == self.DIFF_EQUAL
+                and diffs[pointer + 1][0] == self.DIFF_EQUAL
+            ):
+                # This is a single edit surrounded by equalities.
+                equality1 = diffs[pointer - 1][1]
+                edit = diffs[pointer][1]
+                equality2 = diffs[pointer + 1][1]
+
+                # First, shift the edit as far left as possible.
+                commonOffset = self.diff_commonSuffix(equality1, edit)
+                if commonOffset:
+                    commonString = edit[-commonOffset:]
+                    equality1 = equality1[:-commonOffset]
+                    edit = commonString + edit[:-commonOffset]
+                    equality2 = commonString + equality2
+
+                # Second, step character by character right, looking for the best fit.
+                bestEquality1 = equality1
+                bestEdit = edit
+                bestEquality2 = equality2
+                bestScore = diff_cleanupSemanticScore(
+                    equality1, edit
+                ) + diff_cleanupSemanticScore(edit, equality2)
+                while edit and equality2 and edit[0] == equality2[0]:
+                    equality1 += edit[0]
+                    edit = edit[1:] + equality2[0]
+                    equality2 = equality2[1:]
+                    score = diff_cleanupSemanticScore(
+                        equality1, edit
+                    ) + diff_cleanupSemanticScore(edit, equality2)
+                    # The >= encourages trailing rather than leading whitespace on edits.
+                    if score >= bestScore:
+                        bestScore = score
+                        bestEquality1 = equality1
+                        bestEdit = edit
+                        bestEquality2 = equality2
+
+                if diffs[pointer - 1][1] != bestEquality1:
+                    # We have an improvement, save it back to the diff.
+                    if bestEquality1:
+                        diffs[pointer - 1] = (diffs[pointer - 1][0], bestEquality1)
+                    else:
+                        del diffs[pointer - 1]
+                        pointer -= 1
+                    diffs[pointer] = (diffs[pointer][0], bestEdit)
+                    if bestEquality2:
+                        diffs[pointer + 1] = (diffs[pointer + 1][0], bestEquality2)
+                    else:
+                        del diffs[pointer + 1]
+                        pointer -= 1
+            pointer += 1
+
+    # Define some regex patterns for matching boundaries.
+    BLANKLINEEND = re.compile(r"\n\r?\n$")
+    BLANKLINESTART = re.compile(r"^\r?\n\r?\n")
+
+    def diff_cleanupEfficiency(self, diffs):
+        """Reduce the number of edits by eliminating operationally trivial
+        equalities.
+
+        Args:
+          diffs: Array of diff tuples.
+        """
+        changes = False
+        equalities = []  # Stack of indices where equalities are found.
+        lastEquality = None  # Always equal to diffs[equalities[-1]][1]
+        pointer = 0  # Index of current position.
+        pre_ins = False  # Is there an insertion operation before the last equality.
+        pre_del = False  # Is there a deletion operation before the last equality.
+        post_ins = False  # Is there an insertion operation after the last equality.
+        post_del = False  # Is there a deletion operation after the last equality.
+        while pointer < len(diffs):
+            if diffs[pointer][0] == self.DIFF_EQUAL:  # Equality found.
+                if len(diffs[pointer][1]) < self.Diff_EditCost and (
+                    post_ins or post_del
+                ):
+                    # Candidate found.
+                    equalities.append(pointer)
+                    pre_ins = post_ins
+                    pre_del = post_del
+                    lastEquality = diffs[pointer][1]
+                else:
+                    # Not a candidate, and can never become one.
+                    equalities = []
+                    lastEquality = None
+
+                post_ins = post_del = False
+            else:  # An insertion or deletion.
+                if diffs[pointer][0] == self.DIFF_DELETE:
+                    post_del = True
+                else:
+                    post_ins = True
+
+                # Five types to be split:
+                # <ins>A</ins><del>B</del>XY<ins>C</ins><del>D</del>
+                # <ins>A</ins>X<ins>C</ins><del>D</del>
+                # <ins>A</ins><del>B</del>X<ins>C</ins>
+                # <ins>A</del>X<ins>C</ins><del>D</del>
+                # <ins>A</ins><del>B</del>X<del>C</del>
+
+                if lastEquality and (
+                    (pre_ins and pre_del and post_ins and post_del)
+                    or (
+                        (len(lastEquality) < self.Diff_EditCost / 2)
+                        and (pre_ins + pre_del + post_ins + post_del) == 3
+                    )
+                ):
+                    # Duplicate record.
+                    diffs.insert(equalities[-1], (self.DIFF_DELETE, lastEquality))
+                    # Change second copy to insert.
+                    diffs[equalities[-1] + 1] = (
+                        self.DIFF_INSERT,
+                        diffs[equalities[-1] + 1][1],
+                    )
+                    equalities.pop()  # Throw away the equality we just deleted.
+                    lastEquality = None
+                    if pre_ins and pre_del:
+                        # No changes made which could affect previous entry, keep going.
+                        post_ins = post_del = True
+                        equalities = []
+                    else:
+                        if len(equalities):
+                            equalities.pop()  # Throw away the previous equality.
+                        if len(equalities):
+                            pointer = equalities[-1]
+                        else:
+                            pointer = -1
+                        post_ins = post_del = False
+                    changes = True
+            pointer += 1
+
+        if changes:
+            self.diff_cleanupMerge(diffs)
+
+    def diff_cleanupMerge(self, diffs):
+        """Reorder and merge like edit sections.  Merge equalities.
+        Any edit section can move as long as it doesn't cross an equality.
+
+        Args:
+          diffs: Array of diff tuples.
+        """
+        diffs.append((self.DIFF_EQUAL, ""))  # Add a dummy entry at the end.
+        pointer = 0
+        count_delete = 0
+        count_insert = 0
+        text_delete = ""
+        text_insert = ""
+        while pointer < len(diffs):
+            if diffs[pointer][0] == self.DIFF_INSERT:
+                count_insert += 1
+                text_insert += diffs[pointer][1]
+                pointer += 1
+            elif diffs[pointer][0] == self.DIFF_DELETE:
+                count_delete += 1
+                text_delete += diffs[pointer][1]
+                pointer += 1
+            elif diffs[pointer][0] == self.DIFF_EQUAL:
+                # Upon reaching an equality, check for prior redundancies.
+                if count_delete + count_insert > 1:
+                    if count_delete != 0 and count_insert != 0:
+                        # Factor out any common prefixies.
+                        commonlength = self.diff_commonPrefix(text_insert, text_delete)
+                        if commonlength != 0:
+                            x = pointer - count_delete - count_insert - 1
+                            if x >= 0 and diffs[x][0] == self.DIFF_EQUAL:
+                                diffs[x] = (
+                                    diffs[x][0],
+                                    diffs[x][1] + text_insert[:commonlength],
+                                )
+                            else:
+                                diffs.insert(
+                                    0, (self.DIFF_EQUAL, text_insert[:commonlength])
+                                )
+                                pointer += 1
+                            text_insert = text_insert[commonlength:]
+                            text_delete = text_delete[commonlength:]
+                        # Factor out any common suffixies.
+                        commonlength = self.diff_commonSuffix(text_insert, text_delete)
+                        if commonlength != 0:
+                            diffs[pointer] = (
+                                diffs[pointer][0],
+                                text_insert[-commonlength:] + diffs[pointer][1],
+                            )
+                            text_insert = text_insert[:-commonlength]
+                            text_delete = text_delete[:-commonlength]
+                    # Delete the offending records and add the merged ones.
+                    new_ops = []
+                    if len(text_delete) != 0:
+                        new_ops.append((self.DIFF_DELETE, text_delete))
+                    if len(text_insert) != 0:
+                        new_ops.append((self.DIFF_INSERT, text_insert))
+                    pointer -= count_delete + count_insert
+                    diffs[pointer : pointer + count_delete + count_insert] = new_ops
+                    pointer += len(new_ops) + 1
+                elif pointer != 0 and diffs[pointer - 1][0] == self.DIFF_EQUAL:
+                    # Merge this equality with the previous one.
+                    diffs[pointer - 1] = (
+                        diffs[pointer - 1][0],
+                        diffs[pointer - 1][1] + diffs[pointer][1],
+                    )
+                    del diffs[pointer]
+                else:
+                    pointer += 1
+
+                count_insert = 0
+                count_delete = 0
+                text_delete = ""
+                text_insert = ""
+
+        if diffs[-1][1] == "":
+            diffs.pop()  # Remove the dummy entry at the end.
+
+        # Second pass: look for single edits surrounded on both sides by equalities
+        # which can be shifted sideways to eliminate an equality.
+        # e.g: A<ins>BA</ins>C -> <ins>AB</ins>AC
+        changes = False
+        pointer = 1
+        # Intentionally ignore the first and last element (don't need checking).
+        while pointer < len(diffs) - 1:
+            if (
+                diffs[pointer - 1][0] == self.DIFF_EQUAL
+                and diffs[pointer + 1][0] == self.DIFF_EQUAL
+            ):
+                # This is a single edit surrounded by equalities.
+                if diffs[pointer][1].endswith(diffs[pointer - 1][1]):
+                    # Shift the edit over the previous equality.
+                    if diffs[pointer - 1][1] != "":
+                        diffs[pointer] = (
+                            diffs[pointer][0],
+                            diffs[pointer - 1][1]
+                            + diffs[pointer][1][: -len(diffs[pointer - 1][1])],
+                        )
+                        diffs[pointer + 1] = (
+                            diffs[pointer + 1][0],
+                            diffs[pointer - 1][1] + diffs[pointer + 1][1],
+                        )
+                    del diffs[pointer - 1]
+                    changes = True
+                elif diffs[pointer][1].startswith(diffs[pointer + 1][1]):
+                    # Shift the edit over the next equality.
+                    diffs[pointer - 1] = (
+                        diffs[pointer - 1][0],
+                        diffs[pointer - 1][1] + diffs[pointer + 1][1],
+                    )
+                    diffs[pointer] = (
+                        diffs[pointer][0],
+                        diffs[pointer][1][len(diffs[pointer + 1][1]) :]
+                        + diffs[pointer + 1][1],
+                    )
+                    del diffs[pointer + 1]
+                    changes = True
+            pointer += 1
+
+        # If shifts were made, the diff needs reordering and another shift sweep.
+        if changes:
+            self.diff_cleanupMerge(diffs)
+
+    def diff_xIndex(self, diffs, loc):
+        """loc is a location in text1, compute and return the equivalent location
+        in text2.  e.g. "The cat" vs "The big cat", 1->1, 5->8
+
+        Args:
+          diffs: Array of diff tuples.
+          loc: Location within text1.
+
+        Returns:
+          Location within text2.
+        """
+        chars1 = 0
+        chars2 = 0
+        last_chars1 = 0
+        last_chars2 = 0
+        for x in range(len(diffs)):
+            (op, text) = diffs[x]
+            if op != self.DIFF_INSERT:  # Equality or deletion.
+                chars1 += len(text)
+            if op != self.DIFF_DELETE:  # Equality or insertion.
+                chars2 += len(text)
+            if chars1 > loc:  # Overshot the location.
+                break
+            last_chars1 = chars1
+            last_chars2 = chars2
+
+        if len(diffs) != x and diffs[x][0] == self.DIFF_DELETE:
+            # The location was deleted.
+            return last_chars2
+        # Add the remaining len(character).
+        return last_chars2 + (loc - last_chars1)
+
+    def diff_prettyHtml(self, diffs):
+        """Convert a diff array into a pretty HTML report.
+
+        Args:
+          diffs: Array of diff tuples.
+
+        Returns:
+          HTML representation.
+        """
+        html = []
+        for op, data in diffs:
+            text = (
+                data.replace("&", "&amp;")
+                .replace("<", "&lt;")
+                .replace(">", "&gt;")
+                .replace("\n", "&para;<br>")
+            )
+            if op == self.DIFF_INSERT:
+                html.append('<ins Makefile="background:#e6ffe6;">%s</ins>' % text)
+            elif op == self.DIFF_DELETE:
+                html.append('<del Makefile="background:#ffe6e6;">%s</del>' % text)
+            elif op == self.DIFF_EQUAL:
+                html.append("<span>%s</span>" % text)
+        return "".join(html)
+
+    def diff_text1(self, diffs):
+        """Compute and return the source text (all equalities and deletions).
+
+        Args:
+          diffs: Array of diff tuples.
+
+        Returns:
+          Source text.
+        """
+        text = []
+        for op, data in diffs:
+            if op != self.DIFF_INSERT:
+                text.append(data)
+        return "".join(text)
+
+    def diff_text2(self, diffs):
+        """Compute and return the destination text (all equalities and insertions).
+
+        Args:
+          diffs: Array of diff tuples.
+
+        Returns:
+          Destination text.
+        """
+        text = []
+        for op, data in diffs:
+            if op != self.DIFF_DELETE:
+                text.append(data)
+        return "".join(text)
+
+    def diff_levenshtein(self, diffs):
+        """Compute the Levenshtein distance; the number of inserted, deleted or
+        substituted characters.
+
+        Args:
+          diffs: Array of diff tuples.
+
+        Returns:
+          Number of changes.
+        """
+        levenshtein = 0
+        insertions = 0
+        deletions = 0
+        for op, data in diffs:
+            if op == self.DIFF_INSERT:
+                insertions += len(data)
+            elif op == self.DIFF_DELETE:
+                deletions += len(data)
+            elif op == self.DIFF_EQUAL:
+                # A deletion and an insertion is one substitution.
+                levenshtein += max(insertions, deletions)
+                insertions = 0
+                deletions = 0
+        levenshtein += max(insertions, deletions)
+        return levenshtein
+
+    def diff_toDelta(self, diffs):
+        """Crush the diff into an encoded string which describes the operations
+        required to transform text1 into text2.
+        E.g. =3\t-2\t+ing  -> Keep 3 chars, delete 2 chars, insert 'ing'.
+        Operations are tab-separated.  Inserted text is escaped using %xx notation.
+
+        Args:
+          diffs: Array of diff tuples.
+
+        Returns:
+          Delta text.
+        """
+        text = []
+        for op, data in diffs:
+            if op == self.DIFF_INSERT:
+                # High ascii will raise UnicodeDecodeError.  Use Unicode instead.
+                data = data.encode("utf-8")
+                text.append("+" + urllib.parse.quote(data, "!~*'();/?:@&=+$,# "))
+            elif op == self.DIFF_DELETE:
+                text.append("-%d" % len(data))
+            elif op == self.DIFF_EQUAL:
+                text.append("=%d" % len(data))
+        return "\t".join(text)
+
+    def diff_fromDelta(self, text1, delta):
+        """Given the original text1, and an encoded string which describes the
+        operations required to transform text1 into text2, compute the full diff.
+
+        Args:
+          text1: Source string for the diff.
+          delta: Delta text.
+
+        Returns:
+          Array of diff tuples.
+
+        Raises:
+          ValueError: If invalid input.
+        """
+        diffs = []
+        pointer = 0  # Cursor in text1
+        tokens = delta.split("\t")
+        for token in tokens:
+            if token == "":
+                # Blank tokens are ok (from a trailing \t).
+                continue
+            # Each token begins with a one character parameter which specifies the
+            # operation of this token (delete, insert, equality).
+            param = token[1:]
+            if token[0] == "+":
+                param = urllib.parse.unquote(param)
+                diffs.append((self.DIFF_INSERT, param))
+            elif token[0] == "-" or token[0] == "=":
+                try:
+                    n = int(param)
+                except ValueError:
+                    raise ValueError("Invalid number in diff_fromDelta: " + param)
+                if n < 0:
+                    raise ValueError("Negative number in diff_fromDelta: " + param)
+                text = text1[pointer : pointer + n]
+                pointer += n
+                if token[0] == "=":
+                    diffs.append((self.DIFF_EQUAL, text))
+                else:
+                    diffs.append((self.DIFF_DELETE, text))
+            else:
+                # Anything else is an error.
+                raise ValueError(
+                    "Invalid diff operation in diff_fromDelta: " + token[0]
+                )
+        if pointer != len(text1):
+            raise ValueError(
+                "Delta length (%d) does not equal source text length (%d)."
+                % (pointer, len(text1))
+            )
+        return diffs
+
+    #  MATCH FUNCTIONS
+
+    def match_main(self, text, pattern, loc):
+        """Locate the best instance of 'pattern' in 'text' near 'loc'.
+
+        Args:
+          text: The text to search.
+          pattern: The pattern to search for.
+          loc: The location to search around.
+
+        Returns:
+          Best match index or -1.
+        """
+        # Check for null inputs.
+        if text is None or pattern is None:
+            raise ValueError("Null inputs. (match_main)")
+
+        loc = max(0, min(loc, len(text)))
+        if text == pattern:
+            # Shortcut (potentially not guaranteed by the algorithm)
+            return 0
+        elif not text:
+            # Nothing to match.
+            return -1
+        elif text[loc : loc + len(pattern)] == pattern:
+            # Perfect match at the perfect spot!  (Includes case of null pattern)
+            return loc
+        else:
+            # Do a fuzzy compare.
+            match = self.match_bitap(text, pattern, loc)
+            return match
+
+    def match_bitap(self, text, pattern, loc):
+        """Locate the best instance of 'pattern' in 'text' near 'loc' using the
+        Bitap algorithm.
+
+        Args:
+          text: The text to search.
+          pattern: The pattern to search for.
+          loc: The location to search around.
+
+        Returns:
+          Best match index or -1.
+        """
+        # Python doesn't have a maxint limit, so ignore this check.
+        # if self.Match_MaxBits != 0 and len(pattern) > self.Match_MaxBits:
+        #  raise ValueError("Pattern too long for this application.")
+
+        # Initialise the alphabet.
+        s = self.match_alphabet(pattern)
+
+        def match_bitapScore(e, x):
+            """Compute and return the score for a match with e errors and x location.
+            Accesses loc and pattern through being a closure.
+
+            Args:
+              e: Number of errors in match.
+              x: Location of match.
+
+            Returns:
+              Overall score for match (0.0 = good, 1.0 = bad).
+            """
+            accuracy = float(e) / len(pattern)
+            proximity = abs(loc - x)
+            if not self.Match_Distance:
+                # Dodge divide by zero error.
+                return proximity and 1.0 or accuracy
+            return accuracy + (proximity / float(self.Match_Distance))
+
+        # Highest score beyond which we give up.
+        score_threshold = self.Match_Threshold
+        # Is there a nearby exact match? (speedup)
+        best_loc = text.find(pattern, loc)
+        if best_loc != -1:
+            score_threshold = min(match_bitapScore(0, best_loc), score_threshold)
+            # What about in the other direction? (speedup)
+            best_loc = text.rfind(pattern, loc + len(pattern))
+            if best_loc != -1:
+                score_threshold = min(match_bitapScore(0, best_loc), score_threshold)
+
+        # Initialise the bit arrays.
+        matchmask = 1 << (len(pattern) - 1)
+        best_loc = -1
+
+        bin_max = len(pattern) + len(text)
+        # Empty initialization added to appease pychecker.
+        last_rd = None
+        for d in range(len(pattern)):
+            # Scan for the best match each iteration allows for one more error.
+            # Run a binary search to determine how far from 'loc' we can stray at
+            # this error level.
+            bin_min = 0
+            bin_mid = bin_max
+            while bin_min < bin_mid:
+                if match_bitapScore(d, loc + bin_mid) <= score_threshold:
+                    bin_min = bin_mid
+                else:
+                    bin_max = bin_mid
+                bin_mid = (bin_max - bin_min) // 2 + bin_min
+
+            # Use the result from this iteration as the maximum for the next.
+            bin_max = bin_mid
+            start = max(1, loc - bin_mid + 1)
+            finish = min(loc + bin_mid, len(text)) + len(pattern)
+
+            rd = [0] * (finish + 2)
+            rd[finish + 1] = (1 << d) - 1
+            for j in range(finish, start - 1, -1):
+                if len(text) <= j - 1:
+                    # Out of range.
+                    charMatch = 0
+                else:
+                    charMatch = s.get(text[j - 1], 0)
+                if d == 0:  # First pass: exact match.
+                    rd[j] = ((rd[j + 1] << 1) | 1) & charMatch
+                else:  # Subsequent passes: fuzzy match.
+                    rd[j] = (
+                        (((rd[j + 1] << 1) | 1) & charMatch)
+                        | (((last_rd[j + 1] | last_rd[j]) << 1) | 1)
+                        | last_rd[j + 1]
+                    )
+                if rd[j] & matchmask:
+                    score = match_bitapScore(d, j - 1)
+                    # This match will almost certainly be better than any existing match.
+                    # But check anyway.
+                    if score <= score_threshold:
+                        # Told you so.
+                        score_threshold = score
+                        best_loc = j - 1
+                        if best_loc > loc:
+                            # When passing loc, don't exceed our current distance from loc.
+                            start = max(1, 2 * loc - best_loc)
+                        else:
+                            # Already passed loc, downhill from here on in.
+                            break
+            # No hope for a (better) match at greater error levels.
+            if match_bitapScore(d + 1, loc) > score_threshold:
+                break
+            last_rd = rd
+        return best_loc
+
+    def match_alphabet(self, pattern):
+        """Initialise the alphabet for the Bitap algorithm.
+
+        Args:
+          pattern: The text to encode.
+
+        Returns:
+          Hash of character locations.
+        """
+        s = {}
+        for char in pattern:
+            s[char] = 0
+        for i in range(len(pattern)):
+            s[pattern[i]] |= 1 << (len(pattern) - i - 1)
+        return s
+
+    #  PATCH FUNCTIONS
+
+    def patch_addContext(self, patch, text):
+        """Increase the context until it is unique,
+        but don't let the pattern expand beyond Match_MaxBits.
+
+        Args:
+          patch: The patch to grow.
+          text: Source text.
+        """
+        if len(text) == 0:
+            return
+        pattern = text[patch.start2 : patch.start2 + patch.length1]
+        padding = 0
+
+        # Look for the first and last matches of pattern in text.  If two different
+        # matches are found, increase the pattern length.
+        while text.find(pattern) != text.rfind(pattern) and (
+            self.Match_MaxBits == 0
+            or len(pattern) < self.Match_MaxBits - self.Patch_Margin - self.Patch_Margin
+        ):
+            padding += self.Patch_Margin
+            pattern = text[
+                max(0, patch.start2 - padding) : patch.start2 + patch.length1 + padding
+            ]
+        # Add one chunk for good luck.
+        padding += self.Patch_Margin
+
+        # Add the prefix.
+        prefix = text[max(0, patch.start2 - padding) : patch.start2]
+        if prefix:
+            patch.diffs[:0] = [(self.DIFF_EQUAL, prefix)]
+        # Add the suffix.
+        suffix = text[
+            patch.start2 + patch.length1 : patch.start2 + patch.length1 + padding
+        ]
+        if suffix:
+            patch.diffs.append((self.DIFF_EQUAL, suffix))
+
+        # Roll back the start points.
+        patch.start1 -= len(prefix)
+        patch.start2 -= len(prefix)
+        # Extend lengths.
+        patch.length1 += len(prefix) + len(suffix)
+        patch.length2 += len(prefix) + len(suffix)
+
+    def patch_make(self, a, b=None, c=None):
+        """Compute a list of patches to turn text1 into text2.
+        Use diffs if provided, otherwise compute it ourselves.
+        There are four ways to call this function, depending on what data is
+        available to the caller:
+        Method 1:
+        a = text1, b = text2
+        Method 2:
+        a = diffs
+        Method 3 (optimal):
+        a = text1, b = diffs
+        Method 4 (deprecated, use method 3):
+        a = text1, b = text2, c = diffs
+
+        Args:
+          a: text1 (methods 1,3,4) or Array of diff tuples for text1 to
+              text2 (method 2).
+          b: text2 (methods 1,4) or Array of diff tuples for text1 to
+              text2 (method 3) or undefined (method 2).
+          c: Array of diff tuples for text1 to text2 (method 4) or
+              undefined (methods 1,2,3).
+
+        Returns:
+          Array of Patch objects.
+        """
+        text1 = None
+        diffs = None
+        if isinstance(a, str) and isinstance(b, str) and c is None:
+            # Method 1: text1, text2
+            # Compute diffs from text1 and text2.
+            text1 = a
+            diffs = self.diff_main(text1, b, True)
+            if len(diffs) > 2:
+                self.diff_cleanupSemantic(diffs)
+                self.diff_cleanupEfficiency(diffs)
+        elif isinstance(a, list) and b is None and c is None:
+            # Method 2: diffs
+            # Compute text1 from diffs.
+            diffs = a
+            text1 = self.diff_text1(diffs)
+        elif isinstance(a, str) and isinstance(b, list) and c is None:
+            # Method 3: text1, diffs
+            text1 = a
+            diffs = b
+        elif isinstance(a, str) and isinstance(b, str) and isinstance(c, list):
+            # Method 4: text1, text2, diffs
+            # text2 is not used.
+            text1 = a
+            diffs = c
+        else:
+            raise ValueError("Unknown call format to patch_make.")
+
+        if not diffs:
+            return []  # Get rid of the None case.
+        patches = []
+        patch = patch_obj()
+        char_count1 = 0  # Number of characters into the text1 string.
+        char_count2 = 0  # Number of characters into the text2 string.
+        prepatch_text = text1  # Recreate the patches to determine context info.
+        postpatch_text = text1
+        for x in range(len(diffs)):
+            (diff_type, diff_text) = diffs[x]
+            if len(patch.diffs) == 0 and diff_type != self.DIFF_EQUAL:
+                # A new patch starts here.
+                patch.start1 = char_count1
+                patch.start2 = char_count2
+            if diff_type == self.DIFF_INSERT:
+                # Insertion
+                patch.diffs.append(diffs[x])
+                patch.length2 += len(diff_text)
+                postpatch_text = (
+                    postpatch_text[:char_count2]
+                    + diff_text
+                    + postpatch_text[char_count2:]
+                )
+            elif diff_type == self.DIFF_DELETE:
+                # Deletion.
+                patch.length1 += len(diff_text)
+                patch.diffs.append(diffs[x])
+                postpatch_text = (
+                    postpatch_text[:char_count2]
+                    + postpatch_text[char_count2 + len(diff_text) :]
+                )
+            elif (
+                diff_type == self.DIFF_EQUAL
+                and len(diff_text) <= 2 * self.Patch_Margin
+                and len(patch.diffs) != 0
+                and len(diffs) != x + 1
+            ):
+                # Small equality inside a patch.
+                patch.diffs.append(diffs[x])
+                patch.length1 += len(diff_text)
+                patch.length2 += len(diff_text)
+
+            if diff_type == self.DIFF_EQUAL and len(diff_text) >= 2 * self.Patch_Margin:
+                # Time for a new patch.
+                if len(patch.diffs) != 0:
+                    self.patch_addContext(patch, prepatch_text)
+                    patches.append(patch)
+                    patch = patch_obj()
+                    # Unlike Unidiff, our patch lists have a rolling context.
+                    # https://github.com/google/diff-match-patch/wiki/Unidiff
+                    # Update prepatch text & pos to reflect the application of the
+                    # just completed patch.
+                    prepatch_text = postpatch_text
+                    char_count1 = char_count2
+
+            # Update the current character count.
+            if diff_type != self.DIFF_INSERT:
+                char_count1 += len(diff_text)
+            if diff_type != self.DIFF_DELETE:
+                char_count2 += len(diff_text)
+
+        # Pick up the leftover patch if not empty.
+        if len(patch.diffs) != 0:
+            self.patch_addContext(patch, prepatch_text)
+            patches.append(patch)
+        return patches
+
+    def patch_deepCopy(self, patches):
+        """Given an array of patches, return another array that is identical.
+
+        Args:
+          patches: Array of Patch objects.
+
+        Returns:
+          Array of Patch objects.
+        """
+        patchesCopy = []
+        for patch in patches:
+            patchCopy = patch_obj()
+            # No need to deep copy the tuples since they are immutable.
+            patchCopy.diffs = patch.diffs[:]
+            patchCopy.start1 = patch.start1
+            patchCopy.start2 = patch.start2
+            patchCopy.length1 = patch.length1
+            patchCopy.length2 = patch.length2
+            patchesCopy.append(patchCopy)
+        return patchesCopy
+
+    def patch_apply(self, patches, text):
+        """Merge a set of patches onto the text.  Return a patched text, as well
+        as a list of true/false values indicating which patches were applied.
+
+        Args:
+          patches: Array of Patch objects.
+          text: Old text.
+
+        Returns:
+          Two element Array, containing the new text and an array of boolean values.
+        """
+        if not patches:
+            return (text, [])
+
+        # Deep copy the patches so that no changes are made to originals.
+        patches = self.patch_deepCopy(patches)
+
+        nullPadding = self.patch_addPadding(patches)
+        text = nullPadding + text + nullPadding
+        self.patch_splitMax(patches)
+
+        # delta keeps track of the offset between the expected and actual location
+        # of the previous patch.  If there are patches expected at positions 10 and
+        # 20, but the first patch was found at 12, delta is 2 and the second patch
+        # has an effective expected position of 22.
+        delta = 0
+        results = []
+        for patch in patches:
+            expected_loc = patch.start2 + delta
+            text1 = self.diff_text1(patch.diffs)
+            end_loc = -1
+            if len(text1) > self.Match_MaxBits:
+                # patch_splitMax will only provide an oversized pattern in the case of
+                # a monster delete.
+                start_loc = self.match_main(
+                    text, text1[: self.Match_MaxBits], expected_loc
+                )
+                if start_loc != -1:
+                    end_loc = self.match_main(
+                        text,
+                        text1[-self.Match_MaxBits :],
+                        expected_loc + len(text1) - self.Match_MaxBits,
+                    )
+                    if end_loc == -1 or start_loc >= end_loc:
+                        # Can't find valid trailing context.  Drop this patch.
+                        start_loc = -1
+            else:
+                start_loc = self.match_main(text, text1, expected_loc)
+            if start_loc == -1:
+                # No match found.  :(
+                results.append(False)
+                # Subtract the delta for this failed patch from subsequent patches.
+                delta -= patch.length2 - patch.length1
+            else:
+                # Found a match.  :)
+                results.append(True)
+                delta = start_loc - expected_loc
+                if end_loc == -1:
+                    text2 = text[start_loc : start_loc + len(text1)]
+                else:
+                    text2 = text[start_loc : end_loc + self.Match_MaxBits]
+                if text1 == text2:
+                    # Perfect match, just shove the replacement text in.
+                    text = (
+                        text[:start_loc]
+                        + self.diff_text2(patch.diffs)
+                        + text[start_loc + len(text1) :]
+                    )
+                else:
+                    # Imperfect match.
+                    # Run a diff to get a framework of equivalent indices.
+                    diffs = self.diff_main(text1, text2, False)
+                    if (
+                        len(text1) > self.Match_MaxBits
+                        and self.diff_levenshtein(diffs) / float(len(text1))
+                        > self.Patch_DeleteThreshold
+                    ):
+                        # The end points match, but the content is unacceptably bad.
+                        results[-1] = False
+                    else:
+                        self.diff_cleanupSemanticLossless(diffs)
+                        index1 = 0
+                        for op, data in patch.diffs:
+                            if op != self.DIFF_EQUAL:
+                                index2 = self.diff_xIndex(diffs, index1)
+                            if op == self.DIFF_INSERT:  # Insertion
+                                text = (
+                                    text[: start_loc + index2]
+                                    + data
+                                    + text[start_loc + index2 :]
+                                )
+                            elif op == self.DIFF_DELETE:  # Deletion
+                                text = (
+                                    text[: start_loc + index2]
+                                    + text[
+                                        start_loc
+                                        + self.diff_xIndex(diffs, index1 + len(data)) :
+                                    ]
+                                )
+                            if op != self.DIFF_DELETE:
+                                index1 += len(data)
+        # Strip the padding off.
+        text = text[len(nullPadding) : -len(nullPadding)]
+        return (text, results)
+
+    def patch_addPadding(self, patches):
+        """Add some padding on text start and end so that edges can match
+        something.  Intended to be called only from within patch_apply.
+
+        Args:
+          patches: Array of Patch objects.
+
+        Returns:
+          The padding string added to each side.
+        """
+        paddingLength = self.Patch_Margin
+        nullPadding = ""
+        for x in range(1, paddingLength + 1):
+            nullPadding += chr(x)
+
+        # Bump all the patches forward.
+        for patch in patches:
+            patch.start1 += paddingLength
+            patch.start2 += paddingLength
+
+        # Add some padding on start of first diff.
+        patch = patches[0]
+        diffs = patch.diffs
+        if not diffs or diffs[0][0] != self.DIFF_EQUAL:
+            # Add nullPadding equality.
+            diffs.insert(0, (self.DIFF_EQUAL, nullPadding))
+            patch.start1 -= paddingLength  # Should be 0.
+            patch.start2 -= paddingLength  # Should be 0.
+            patch.length1 += paddingLength
+            patch.length2 += paddingLength
+        elif paddingLength > len(diffs[0][1]):
+            # Grow first equality.
+            extraLength = paddingLength - len(diffs[0][1])
+            newText = nullPadding[len(diffs[0][1]) :] + diffs[0][1]
+            diffs[0] = (diffs[0][0], newText)
+            patch.start1 -= extraLength
+            patch.start2 -= extraLength
+            patch.length1 += extraLength
+            patch.length2 += extraLength
+
+        # Add some padding on end of last diff.
+        patch = patches[-1]
+        diffs = patch.diffs
+        if not diffs or diffs[-1][0] != self.DIFF_EQUAL:
+            # Add nullPadding equality.
+            diffs.append((self.DIFF_EQUAL, nullPadding))
+            patch.length1 += paddingLength
+            patch.length2 += paddingLength
+        elif paddingLength > len(diffs[-1][1]):
+            # Grow last equality.
+            extraLength = paddingLength - len(diffs[-1][1])
+            newText = diffs[-1][1] + nullPadding[:extraLength]
+            diffs[-1] = (diffs[-1][0], newText)
+            patch.length1 += extraLength
+            patch.length2 += extraLength
+
+        return nullPadding
+
+    def patch_splitMax(self, patches):
+        """Look through the patches and break up any which are longer than the
+        maximum limit of the match algorithm.
+        Intended to be called only from within patch_apply.
+
+        Args:
+          patches: Array of Patch objects.
+        """
+        patch_size = self.Match_MaxBits
+        if patch_size == 0:
+            # Python has the option of not splitting strings due to its ability
+            # to handle integers of arbitrary precision.
+            return
+        for x in range(len(patches)):
+            if patches[x].length1 <= patch_size:
+                continue
+            bigpatch = patches[x]
+            # Remove the big old patch.
+            del patches[x]
+            x -= 1
+            start1 = bigpatch.start1
+            start2 = bigpatch.start2
+            precontext = ""
+            while len(bigpatch.diffs) != 0:
+                # Create one of several smaller patches.
+                patch = patch_obj()
+                empty = True
+                patch.start1 = start1 - len(precontext)
+                patch.start2 = start2 - len(precontext)
+                if precontext:
+                    patch.length1 = patch.length2 = len(precontext)
+                    patch.diffs.append((self.DIFF_EQUAL, precontext))
+
+                while (
+                    len(bigpatch.diffs) != 0
+                    and patch.length1 < patch_size - self.Patch_Margin
+                ):
+                    (diff_type, diff_text) = bigpatch.diffs[0]
+                    if diff_type == self.DIFF_INSERT:
+                        # Insertions are harmless.
+                        patch.length2 += len(diff_text)
+                        start2 += len(diff_text)
+                        patch.diffs.append(bigpatch.diffs.pop(0))
+                        empty = False
+                    elif (
+                        diff_type == self.DIFF_DELETE
+                        and len(patch.diffs) == 1
+                        and patch.diffs[0][0] == self.DIFF_EQUAL
+                        and len(diff_text) > 2 * patch_size
+                    ):
+                        # This is a large deletion.  Let it pass in one chunk.
+                        patch.length1 += len(diff_text)
+                        start1 += len(diff_text)
+                        empty = False
+                        patch.diffs.append((diff_type, diff_text))
+                        del bigpatch.diffs[0]
+                    else:
+                        # Deletion or equality.  Only take as much as we can stomach.
+                        diff_text = diff_text[
+                            : patch_size - patch.length1 - self.Patch_Margin
+                        ]
+                        patch.length1 += len(diff_text)
+                        start1 += len(diff_text)
+                        if diff_type == self.DIFF_EQUAL:
+                            patch.length2 += len(diff_text)
+                            start2 += len(diff_text)
+                        else:
+                            empty = False
+
+                        patch.diffs.append((diff_type, diff_text))
+                        if diff_text == bigpatch.diffs[0][1]:
+                            del bigpatch.diffs[0]
+                        else:
+                            bigpatch.diffs[0] = (
+                                bigpatch.diffs[0][0],
+                                bigpatch.diffs[0][1][len(diff_text) :],
+                            )
+
+                # Compute the head context for the next patch.
+                precontext = self.diff_text2(patch.diffs)
+                precontext = precontext[-self.Patch_Margin :]
+                # Append the end context for this patch.
+                postcontext = self.diff_text1(bigpatch.diffs)[: self.Patch_Margin]
+                if postcontext:
+                    patch.length1 += len(postcontext)
+                    patch.length2 += len(postcontext)
+                    if len(patch.diffs) != 0 and patch.diffs[-1][0] == self.DIFF_EQUAL:
+                        patch.diffs[-1] = (
+                            self.DIFF_EQUAL,
+                            patch.diffs[-1][1] + postcontext,
+                        )
+                    else:
+                        patch.diffs.append((self.DIFF_EQUAL, postcontext))
+
+                if not empty:
+                    x += 1
+                    patches.insert(x, patch)
+
+    def patch_toText(self, patches):
+        """Take a list of patches and return a textual representation.
+
+        Args:
+          patches: Array of Patch objects.
+
+        Returns:
+          Text representation of patches.
+        """
+        text = []
+        for patch in patches:
+            text.append(str(patch))
+        return "".join(text)
+
+    def patch_fromText(self, textline):
+        """Parse a textual representation of patches and return a list of patch
+        objects.
+
+        Args:
+          textline: Text representation of patches.
+
+        Returns:
+          Array of Patch objects.
+
+        Raises:
+          ValueError: If invalid input.
+        """
+        patches = []
+        if not textline:
+            return patches
+        text = textline.split("\n")
+        while len(text) != 0:
+            m = re.match(r"^@@ -(\d+),?(\d*) \+(\d+),?(\d*) @@$", text[0])
+            if not m:
+                raise ValueError("Invalid patch string: " + text[0])
+            patch = patch_obj()
+            patches.append(patch)
+            patch.start1 = int(m.group(1))
+            if m.group(2) == "":
+                patch.start1 -= 1
+                patch.length1 = 1
+            elif m.group(2) == "0":
+                patch.length1 = 0
+            else:
+                patch.start1 -= 1
+                patch.length1 = int(m.group(2))
+
+            patch.start2 = int(m.group(3))
+            if m.group(4) == "":
+                patch.start2 -= 1
+                patch.length2 = 1
+            elif m.group(4) == "0":
+                patch.length2 = 0
+            else:
+                patch.start2 -= 1
+                patch.length2 = int(m.group(4))
+
+            del text[0]
+
+            while len(text) != 0:
+                if text[0]:
+                    sign = text[0][0]
+                else:
+                    sign = ""
+                line = urllib.parse.unquote(text[0][1:])
+                if sign == "+":
+                    # Insertion.
+                    patch.diffs.append((self.DIFF_INSERT, line))
+                elif sign == "-":
+                    # Deletion.
+                    patch.diffs.append((self.DIFF_DELETE, line))
+                elif sign == " ":
+                    # Minor equality.
+                    patch.diffs.append((self.DIFF_EQUAL, line))
+                elif sign == "@":
+                    # Start of next patch.
+                    break
+                elif sign == "":
+                    # Blank line?  Whatever.
+                    pass
+                else:
+                    # WTF?
+                    raise ValueError("Invalid patch mode: '%s'\n%s" % (sign, line))
+                del text[0]
+        return patches
+
+
+class patch_obj:
+    """Class representing one patch operation."""
+
+    def __init__(self):
+        """Initializes with an empty list of diffs."""
+        self.diffs = []
+        self.start1 = None
+        self.start2 = None
+        self.length1 = 0
+        self.length2 = 0
+
+    def __str__(self):
+        """Emulate GNU diff's format.
+        Header: @@ -382,8 +481,9 @@
+        Indices are printed as 1-based, not 0-based.
+
+        Returns:
+          The GNU diff string.
+        """
+        if self.length1 == 0:
+            coords1 = str(self.start1) + ",0"
+        elif self.length1 == 1:
+            coords1 = str(self.start1 + 1)
+        else:
+            coords1 = str(self.start1 + 1) + "," + str(self.length1)
+        if self.length2 == 0:
+            coords2 = str(self.start2) + ",0"
+        elif self.length2 == 1:
+            coords2 = str(self.start2 + 1)
+        else:
+            coords2 = str(self.start2 + 1) + "," + str(self.length2)
+        text = ["@@ -", coords1, " +", coords2, " @@\n"]
+        # Escape the body of the patch with %xx notation.
+        for op, data in self.diffs:
+            if op == diff_match_patch.DIFF_INSERT:
+                text.append("+")
+            elif op == diff_match_patch.DIFF_DELETE:
+                text.append("-")
+            elif op == diff_match_patch.DIFF_EQUAL:
+                text.append(" ")
+            # High ascii will raise UnicodeDecodeError.  Use Unicode instead.
+            data = data.encode("utf-8")
+            text.append(urllib.parse.quote(data, "!~*'();/?:@&=+$,# ") + "\n")
+        return "".join(text)

requirements.txt

@@ -0,0 +1,3 @@
+pandas
+datasets[audio]
+jiwer