Merge pull request #227 from mailgun/maxim/develop

PIP-1615: Operate on unicode data exclusively [python3]

.build/Dockerfile

@@ -0,0 +1,20 @@
+FROM python:3.9-slim-buster AS deps
+
+RUN apt-get update && \
+    apt-get install -y build-essential git curl python3-dev libatlas3-base libatlas-base-dev liblapack-dev libxml2 libxml2-dev libffi6 libffi-dev musl-dev libxslt-dev
+
+FROM deps AS testable
+ARG REPORT_PATH
+
+VOLUME ["/var/mailgun", "/etc/mailgun/ssl", ${REPORT_PATH}]
+
+ADD . /app
+WORKDIR /app
+COPY wheel/* /wheel/
+
+RUN mkdir -p ${REPORT_PATH}
+
+RUN python ./setup.py build bdist_wheel -d /wheel && \
+    pip install --no-deps /wheel/*
+
+ENTRYPOINT ["/bin/sh", "/app/run_tests.sh"]

.gitignore

@@ -54,3 +54,6 @@ _trial_temp
 
 # OSX
 .DS_Store
+
+# vim-backup
+*.bak

MANIFEST.in

@@ -5,3 +5,10 @@ include classifier
 include LICENSE
 include MANIFEST.in
 include README.rst
+include talon/signature/data/train.data
+include talon/signature/data/classifier
+include talon/signature/data/classifier_01.npy
+include talon/signature/data/classifier_02.npy
+include talon/signature/data/classifier_03.npy
+include talon/signature/data/classifier_04.npy
+include talon/signature/data/classifier_05.npy

README.rst

@@ -129,6 +129,22 @@ start using it for talon.
 .. _EDRM: http://www.edrm.net/resources/data-sets/edrm-enron-email-data-set
 .. _forge: https://github.com/mailgun/forge
 
+Training on your dataset
+------------------------
+
+talon comes with a pre-processed dataset and a pre-trained classifier. To retrain the classifier on your own dataset of raw emails, structure and annotate them in the same way the `forge`_ project does. Then do:
+
+.. code:: python
+
+    from talon.signature.learning.dataset import build_extraction_dataset
+    from talon.signature.learning import classifier as c 
+    
+    build_extraction_dataset("/path/to/your/P/folder", "/path/to/talon/signature/data/train.data")
+    c.train(c.init(), "/path/to/talon/signature/data/train.data", "/path/to/talon/signature/data/classifier")
+
+Note that for signature extraction you need just the folder with the positive samples with annotated signature lines (P folder).
+
+.. _forge: https://github.com/mailgun/forge
 
 Research
 --------

requirements.txt

@@ -0,0 +1,11 @@
+chardet>=1.0.1
+cchardet>=0.3.5
+cssselect
+html5lib
+joblib
+lxml>=2.3.3
+numpy
+regex>=1
+scikit-learn>=1.0.0
+scipy
+six>=1.10.0

run_tests.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -ex
+REPORT_PATH="${REPORT_PATH:-./}"
+nosetests --with-xunit --with-coverage --cover-xml --cover-xml-file $REPORT_PATH/coverage.xml --xunit-file=$REPORT_PATH/nosetests.xml --cover-package=talon .

setup.py

@@ -19,17 +19,17 @@ class InstallCommand(install):
         if self.no_ml:
             dist = self.distribution
             dist.packages=find_packages(exclude=[
-                'tests',
-                'tests.*',
-                'talon.signature',
-                'talon.signature.*',
+                "tests",
+                "tests.*",
+                "talon.signature",
+                "talon.signature.*",
             ])
-            for not_required in ['numpy', 'scipy', 'scikit-learn==0.16.1']:
+            for not_required in ["numpy", "scipy", "scikit-learn==0.24.1"]:
                 dist.install_requires.remove(not_required)
 
 
 setup(name='talon',
-      version='1.4.5',
+      version='1.6.0',
       description=("Mailgun library "
                    "to extract message quotations and signatures."),
       long_description=open("README.rst").read(),
@@ -44,20 +44,21 @@ setup(name='talon',
       include_package_data=True,
       zip_safe=True,
       install_requires=[
-          "lxml>=2.3.3",
-          "regex>=1",
+          "lxml",
+          "regex",
           "numpy",
           "scipy",
-          "scikit-learn>=0.16.1", # pickled versions of classifier, else rebuild
-          'chardet>=1.0.1',
-          'cchardet>=0.3.5',
-          'cssselect',
-          'six>=1.10.0',
-          'html5lib'
+          "scikit-learn>=1.0.0",
+          "chardet",
+          "cchardet",
+          "cssselect",
+          "six",
+          "html5lib",
+          "joblib",
           ],
       tests_require=[
           "mock",
-          "nose>=1.2.1",
+          "nose",
           "coverage"
           ]
       )

talon/html_quotations.py

@@ -87,23 +87,24 @@ def cut_gmail_quote(html_message):
 
 def cut_microsoft_quote(html_message):
     ''' Cuts splitter block and all following blocks. '''
+    #use EXSLT extensions to have a regex match() function with lxml
+    ns = {"re": "http://exslt.org/regular-expressions"}
+
+    #general pattern: @style='border:none;border-top:solid <color> 1.0pt;padding:3.0pt 0<unit> 0<unit> 0<unit>'
+    #outlook 2007, 2010 (international) <color=#B5C4DF> <unit=cm>
+    #outlook 2007, 2010 (american)      <color=#B5C4DF> <unit=pt>
+    #outlook 2013       (international) <color=#E1E1E1> <unit=cm>
+    #outlook 2013       (american)      <color=#E1E1E1> <unit=pt>
+    #also handles a variant with a space after the semicolon
     splitter = html_message.xpath(
-        #outlook 2007, 2010 (international)
-        "//div[@style='border:none;border-top:solid #B5C4DF 1.0pt;"
-        "padding:3.0pt 0cm 0cm 0cm']|"
-        #outlook 2007, 2010 (american)
-        "//div[@style='border:none;border-top:solid #B5C4DF 1.0pt;"
-        "padding:3.0pt 0in 0in 0in']|"
-        #outlook 2013 (international)
-        "//div[@style='border:none;border-top:solid #E1E1E1 1.0pt;"
-        "padding:3.0pt 0cm 0cm 0cm']|"
-        #outlook 2013 (american)
-        "//div[@style='border:none;border-top:solid #E1E1E1 1.0pt;"
-        "padding:3.0pt 0in 0in 0in']|"
+        #outlook 2007, 2010, 2013 (international, american)
+        "//div[@style[re:match(., 'border:none; ?border-top:solid #(E1E1E1|B5C4DF) 1.0pt; ?"
+        "padding:3.0pt 0(in|cm) 0(in|cm) 0(in|cm)')]]|"
         #windows mail
         "//div[@style='padding-top: 5px; "
         "border-top-color: rgb(229, 229, 229); "
         "border-top-width: 1px; border-top-style: solid;']"
+        , namespaces=ns
     )
 
     if splitter:

talon/quotations.py

@@ -6,23 +6,22 @@ original messages (without quoted messages)
 """
 
 from __future__ import absolute_import
-import regex as re
+
 import logging
 from copy import deepcopy
 
-from lxml import html, etree
-
-from talon.utils import (get_delimiter, html_tree_to_text,
-                         html_document_fromstring)
-from talon import html_quotations
+import regex as re
+from lxml import etree, html
 from six.moves import range
-import six
 
+from talon import html_quotations
+from talon.utils import (get_delimiter, html_document_fromstring,
+                         html_tree_to_text)
 
 log = logging.getLogger(__name__)
 
 
-RE_FWD = re.compile("^[-]+[ ]*Forwarded message[ ]*[-]+$", re.I | re.M)
+RE_FWD = re.compile("^[-]+[ ]*Forwarded message[ ]*[-]+\s*$", re.I | re.M)
 
 RE_ON_DATE_SMB_WROTE = re.compile(
     u'(-*[>]?[ ]?({0})[ ].*({1})(.*\n){{0,2}}.*({2}):?-*)'.format(
@@ -38,6 +37,8 @@ RE_ON_DATE_SMB_WROTE = re.compile(
             'Op',
             # German
             'Am',
+            # Portuguese
+            'Em',
             # Norwegian
             u'På',
             # Swedish, Danish
@@ -64,6 +65,8 @@ RE_ON_DATE_SMB_WROTE = re.compile(
             'schreef','verzond','geschreven',
             # German
             'schrieb',
+            # Portuguese
+            'escreveu',
             # Norwegian, Swedish
             'skrev',
             # Vietnamese
@@ -90,7 +93,7 @@ RE_ON_DATE_WROTE_SMB = re.compile(
     )
 
 RE_QUOTATION = re.compile(
-    r'''
+    r"""
     (
         # quotation border: splitter line or a number of quotation marker lines
         (?:
@@ -108,10 +111,10 @@ RE_QUOTATION = re.compile(
 
     # after quotations should be text only or nothing at all
     [te]*$
-    ''', re.VERBOSE)
+    """, re.VERBOSE)
 
 RE_EMPTY_QUOTATION = re.compile(
-    r'''
+    r"""
     (
         # quotation border: splitter line or a number of quotation marker lines
         (?:
@@ -121,7 +124,7 @@ RE_EMPTY_QUOTATION = re.compile(
         )
     )
     e*
-    ''', re.VERBOSE)
+    """, re.VERBOSE)
 
 # ------Original Message------ or ---- Reply Message ----
 # With variations in other languages.
@@ -135,13 +138,17 @@ RE_ORIGINAL_MESSAGE = re.compile(u'[\s]*[-]+[ ]*({})[ ]*[-]+'.format(
         'Oprindelig meddelelse',
     ))), re.I)
 
-RE_FROM_COLON_OR_DATE_COLON = re.compile(u'(_+\r?\n)?[\s]*(:?[*]?{})[\s]?:[*]?.*'.format(
+RE_FROM_COLON_OR_DATE_COLON = re.compile(u'((_+\r?\n)?[\s]*:?[*]?({})[\s]?:([^\n$]+\n){{1,2}}){{2,}}'.format(
     u'|'.join((
         # "From" in different languages.
         'From', 'Van', 'De', 'Von', 'Fra', u'Från',
         # "Date" in different languages.
-        'Date', 'Datum', u'Envoyé', 'Skickat', 'Sendt',
-    ))), re.I)
+        'Date', '[S]ent', 'Datum', u'Envoyé', 'Skickat', 'Sendt', 'Gesendet',
+        # "Subject" in different languages.
+        'Subject', 'Betreff', 'Objet', 'Emne', u'Ämne',
+        # "To" in different languages.
+        'To', 'An', 'Til', u'À', 'Till'
+    ))), re.I | re.M)
 
 # ---- John Smith wrote ----
 RE_ANDROID_WROTE = re.compile(u'[\s]*[-]+.*({})[ ]*[-]+'.format(
@@ -185,9 +192,6 @@ RE_PARENTHESIS_LINK = re.compile("\(https?://")
 
 SPLITTER_MAX_LINES = 6
 MAX_LINES_COUNT = 1000
-# an extensive research shows that exceeding this limit
-# leads to excessive processing time
-MAX_HTML_LEN = 2794202
 
 QUOT_PATTERN = re.compile('^>+ ?')
 NO_QUOT_LINE = re.compile('^[^>].*[\S].*')
@@ -286,7 +290,7 @@ def process_marked_lines(lines, markers, return_flags=[False, -1, -1]):
     # inlined reply
     # use lookbehind assertions to find overlapping entries e.g. for 'mtmtm'
     # both 't' entries should be found
-    for inline_reply in re.finditer('(?<=m)e*((?:t+e*)+)m', markers):
+    for inline_reply in re.finditer('(?<=m)e*(t[te]*)m', markers):
         # long links could break sequence of quotation lines but they shouldn't
         # be considered an inline reply
         links = (
@@ -338,9 +342,6 @@ def _replace_link_brackets(msg_body):
 
     Converts msg_body into a unicode
     """
-    if isinstance(msg_body, bytes):
-        msg_body = msg_body.decode('utf8')
-
     def link_wrapper(link):
         newline_index = msg_body[:link.start()].rfind("\n")
         if msg_body[newline_index + 1] == ">":
@@ -380,8 +381,6 @@ def postprocess(msg_body):
 
 def extract_from_plain(msg_body):
     """Extracts a non quoted message from provided plain text."""
-    stripped_text = msg_body
-
     delimiter = get_delimiter(msg_body)
     msg_body = preprocess(msg_body, delimiter)
     # don't process too long messages
@@ -413,22 +412,27 @@ def extract_from_html(msg_body):
 
     Returns a unicode string.
     """
-    if isinstance(msg_body, six.text_type):
-        msg_body = msg_body.encode('utf8')
-    elif not isinstance(msg_body, bytes):
-        msg_body = msg_body.encode('ascii')
+    if msg_body.strip() == "":
+        return msg_body
 
-    result = _extract_from_html(msg_body)
-    if isinstance(result, bytes):
-        result = result.decode('utf8')
+    msg_body = msg_body.replace("\r\n", "\n")
+    # Cut out xml and doctype tags to avoid conflict with unicode decoding.
+    msg_body = re.sub(r"\<\?xml.+\?\>|\<\!DOCTYPE.+]\>", "", msg_body)
+    html_tree = html_document_fromstring(msg_body)
+    if html_tree is None:
+        return msg_body
+
+    result = extract_from_html_tree(html_tree)
+    if not result:
+        return msg_body
 
     return result
 
 
-def _extract_from_html(msg_body):
+def extract_from_html_tree(html_tree):
     """
-    Extract not quoted message from provided html message body
-    using tags and plain text algorithm.
+    Extract not quoted message from provided parsed html tree using tags and
+    plain text algorithm.
 
     Cut out the 'blockquote', 'gmail_quote' tags.
     Cut Microsoft quotations.
@@ -441,15 +445,6 @@ def _extract_from_html(msg_body):
     then checking deleted checkpoints,
     then deleting necessary tags.
     """
-    if msg_body.strip() == b'':
-        return msg_body
-
-    msg_body = msg_body.replace(b'\r\n', b'\n')
-    html_tree = html_document_fromstring(msg_body)
-
-    if html_tree is None:
-        return msg_body
-
     cut_quotations = (html_quotations.cut_gmail_quote(html_tree) or
                       html_quotations.cut_zimbra_quote(html_tree) or
                       html_quotations.cut_blockquote(html_tree) or
@@ -467,7 +462,7 @@ def _extract_from_html(msg_body):
 
     # Don't process too long messages
     if len(lines) > MAX_LINES_COUNT:
-        return msg_body
+        return None
 
     # Collect checkpoints on each line
     line_checkpoints = [
@@ -486,7 +481,7 @@ def _extract_from_html(msg_body):
     lines_were_deleted, first_deleted, last_deleted = return_flags
 
     if not lines_were_deleted and not cut_quotations:
-        return msg_body
+        return None
 
     if lines_were_deleted:
         #collect checkpoints from deleted lines
@@ -500,9 +495,73 @@ def _extract_from_html(msg_body):
         )
 
     if _readable_text_empty(html_tree_copy):
-        return msg_body
+        return None
 
-    return html.tostring(html_tree_copy)
+    # NOTE: We remove_namespaces() because we are using an HTML5 Parser, HTML
+    # parsers do not recognize namespaces in HTML tags. As such the rendered
+    # HTML tags are no longer recognizable HTML tags. Example: <o:p> becomes
+    # <oU0003Ap>. When we port this to golang we should look into using an
+    # XML Parser NOT and HTML5 Parser since we do not know what input a
+    # customer will send us. Switching to a common XML parser in python
+    # opens us up to a host of vulnerabilities.
+    # See https://docs.python.org/3/library/xml.html#xml-vulnerabilities
+    #
+    # The down sides to removing the namespaces is that customers might
+    # judge the XML namespaces important. If that is the case then support
+    # should encourage customers to preform XML parsing of the un-stripped
+    # body to get the full unmodified XML payload.
+    #
+    # Alternatives to this approach are
+    # 1. Ignore the U0003A in tag names and let the customer deal with it.
+    #    This is not ideal, as most customers use stripped-html for viewing
+    #    emails sent from a recipient, as such they cannot control the HTML
+    #    provided by a recipient.
+    # 2. Preform a string replace of 'U0003A' to ':' on the rendered HTML
+    #    string. While this would solve the issue simply, it runs the risk
+    #    of replacing data outside the <tag> which might be essential to
+    #    the customer.
+    remove_namespaces(html_tree_copy)
+    s = html.tostring(html_tree_copy, encoding="ascii")
+    if not s:
+        return None
+
+    return s.decode("ascii")
+
+
+def remove_namespaces(root):
+    """
+    Given the root of an HTML document iterate through all the elements
+    and remove any namespaces that might have been provided and remove
+    any attributes that contain a namespace
+
+    <html xmlns:o="urn:schemas-microsoft-com:office:office">
+    becomes
+    <html>
+
+    <o:p>Hi</o:p>
+    becomes
+    <p>Hi</p>
+
+    Start tags do NOT have a namespace; COLON characters have no special meaning.
+    if we don't remove the namespace the parser translates the tag name into a
+    unicode representation. For example <o:p> becomes <oU0003Ap>
+
+    See https://www.w3.org/TR/2011/WD-html5-20110525/syntax.html#start-tags
+
+
+    """
+    for child in root.iter():
+        for key, value in child.attrib.items():
+            # If the attribute includes a colon
+            if key.rfind("U0003A") != -1:
+                child.attrib.pop(key)
+
+        # If the tag includes a colon
+        idx = child.tag.rfind("U0003A")
+        if idx != -1:
+            child.tag = child.tag[idx+6:]
+
+    return root
 
 
 def split_emails(msg):
@@ -557,7 +616,6 @@ def _correct_splitlines_in_headers(markers, lines):
     updated_markers = ""
     i = 0
     in_header_block = False
-
     for m in markers:
         # Only set in_header_block flag when we hit an 's' and line is a header
         if m == 's':
@@ -586,10 +644,10 @@ def _readable_text_empty(html_tree):
 
 
 def is_splitter(line):
-    '''
+    """
     Returns Matcher object if provided string is a splitter and
     None otherwise.
-    '''
+    """
     for pattern in SPLITTER_PATTERNS:
         matcher = re.match(pattern, line)
         if matcher:
@@ -597,12 +655,12 @@ def is_splitter(line):
 
 
 def text_content(context):
-    '''XPath Extension function to return a node text content.'''
+    """XPath Extension function to return a node text content."""
     return context.context_node.xpath("string()").strip()
 
 
 def tail(context):
-    '''XPath Extension function to return a node tail text.'''
+    """XPath Extension function to return a node tail text."""
     return context.context_node.tail or ''
 
 

talon/signature/__init__.py

@@ -23,17 +23,14 @@ trained against, don't forget to regenerate:
 from __future__ import absolute_import
 import os
 
-from . import extraction
-from . extraction import extract  #noqa
-from . learning import classifier
-
-
-DATA_DIR = os.path.join(os.path.dirname(__file__), 'data')
-
-EXTRACTOR_FILENAME = os.path.join(DATA_DIR, 'classifier')
-EXTRACTOR_DATA = os.path.join(DATA_DIR, 'train.data')
+from talon.signature import extraction
+from talon.signature.extraction import extract
+from talon.signature.learning import classifier
 
 
 def initialize():
-    extraction.EXTRACTOR = classifier.load(EXTRACTOR_FILENAME,
-                                           EXTRACTOR_DATA)
+    data_dir = os.path.join(os.path.dirname(__file__), 'data')
+    extractor_filename = os.path.join(data_dir, 'classifier')
+    extractor_data_filename = os.path.join(data_dir, 'train.data')
+    extraction.EXTRACTOR = classifier.load(extractor_filename,
+                                           extractor_data_filename)

talon/signature/data/__init__.py

@@ -0,0 +1 @@
+

talon/signature/learning/classifier.py

@@ -8,7 +8,7 @@ body belongs to the signature.
 from __future__ import absolute_import
 
 from numpy import genfromtxt
-from sklearn.externals import joblib
+import joblib
 from sklearn.svm import LinearSVC
 
 

talon/signature/learning/helpers.py

@@ -5,21 +5,17 @@
 * regexp's constants used when evaluating signature's features
 
 """
-
-from __future__ import absolute_import
 import unicodedata
+
 import regex as re
 
-from talon.utils import to_unicode
-
 from talon.signature.constants import SIGNATURE_MAX_LINES
 
-
 rc = re.compile
 
 RE_EMAIL = rc('\S@\S')
 RE_RELAX_PHONE = rc('(\(? ?[\d]{2,3} ?\)?.{,3}?){2,}')
-RE_URL = rc(r'''https?://|www\.[\S]+\.[\S]''')
+RE_URL = rc(r"""https?://|www\.[\S]+\.[\S]""")
 
 # Taken from:
 # http://www.cs.cmu.edu/~vitor/papers/sigFilePaper_finalversion.pdf
@@ -55,7 +51,7 @@ BAD_SENDER_NAMES = [
 
 
 def binary_regex_search(prog):
-    '''Returns a function that returns 1 or 0 depending on regex search result.
+    """Returns a function that returns 1 or 0 depending on regex search result.
 
     If regular expression compiled into prog is present in a string
     the result of calling the returned function with the string will be 1
@@ -66,12 +62,12 @@ def binary_regex_search(prog):
     1
     >>> binary_regex_search(re.compile("12"))("34")
     0
-    '''
+    """
     return lambda s: 1 if prog.search(s) else 0
 
 
 def binary_regex_match(prog):
-    '''Returns a function that returns 1 or 0 depending on regex match result.
+    """Returns a function that returns 1 or 0 depending on regex match result.
 
     If a string matches regular expression compiled into prog
     the result of calling the returned function with the string will be 1
@@ -82,7 +78,7 @@ def binary_regex_match(prog):
     1
     >>> binary_regex_match(re.compile("12"))("3 12")
     0
-    '''
+    """
     return lambda s: 1 if prog.match(s) else 0
 
 
@@ -102,7 +98,7 @@ def flatten_list(list_to_flatten):
 
 
 def contains_sender_names(sender):
-    '''Returns a functions to search sender\'s name or it\'s part.
+    """Returns a functions to search sender\'s name or it\'s part.
 
     >>> feature = contains_sender_names("Sergey N.  Obukhov <xxx@example.com>")
     >>> feature("Sergey Obukhov")
@@ -115,7 +111,7 @@ def contains_sender_names(sender):
     1
     >>> contains_sender_names("<serobnic@mail.ru>")("serobnic")
     1
-    '''
+    """
     names = '( |$)|'.join(flatten_list([[e, e.capitalize()]
                                         for e in extract_names(sender)]))
     names = names or sender
@@ -135,20 +131,25 @@ def extract_names(sender):
     >>> extract_names('')
     []
     """
-    sender = to_unicode(sender, precise=True)
     # Remove non-alphabetical characters
     sender = "".join([char if char.isalpha() else ' ' for char in sender])
     # Remove too short words and words from "black" list i.e.
     # words like `ru`, `gmail`, `com`, `org`, etc.
-    sender = [word for word in sender.split() if len(word) > 1 and
-              not word in BAD_SENDER_NAMES]
-    # Remove duplicates
-    names = list(set(sender))
+    names = list()
+    for word in sender.split():
+        if len(word) < 2:
+            continue
+        if word in BAD_SENDER_NAMES:
+            continue
+        if word in names:
+            continue
+        names.append(word)
+
     return names
 
 
 def categories_percent(s, categories):
-    '''Returns category characters percent.
+    """Returns category characters percent.
 
     >>> categories_percent("qqq ggg hhh", ["Po"])
     0.0
@@ -160,9 +161,8 @@ def categories_percent(s, categories):
     50.0
     >>> categories_percent("s.s,5s", ["Po", "Nd"])
     50.0
-    '''
+    """
     count = 0
-    s = to_unicode(s, precise=True)
     for c in s:
         if unicodedata.category(c) in categories:
             count += 1
@@ -170,19 +170,18 @@ def categories_percent(s, categories):
 
 
 def punctuation_percent(s):
-    '''Returns punctuation percent.
+    """Returns punctuation percent.
 
     >>> punctuation_percent("qqq ggg hhh")
     0.0
     >>> punctuation_percent("q,w.")
     50.0
-    '''
+    """
     return categories_percent(s, ['Po'])
 
 
 def capitalized_words_percent(s):
-    '''Returns capitalized words percent.'''
-    s = to_unicode(s, precise=True)
+    """Returns capitalized words percent."""
     words = re.split('\s', s)
     words = [w for w in words if w.strip()]
     words = [w for w in words if len(w) > 2]    
@@ -208,20 +207,26 @@ def many_capitalized_words(s):
 
 
 def has_signature(body, sender):
-    '''Checks if the body has signature. Returns True or False.'''
+    """Checks if the body has signature. Returns True or False."""
     non_empty = [line for line in body.splitlines() if line.strip()]
     candidate = non_empty[-SIGNATURE_MAX_LINES:]
     upvotes = 0
+    sender_check = contains_sender_names(sender)
     for line in candidate:
         # we check lines for sender's name, phone, email and url,
         # those signature lines don't take more then 27 lines
         if len(line.strip()) > 27:
             continue
-        elif contains_sender_names(sender)(line):
+
+        if sender_check(line):
             return True
-        elif (binary_regex_search(RE_RELAX_PHONE)(line) +
-              binary_regex_search(RE_EMAIL)(line) +
-              binary_regex_search(RE_URL)(line) == 1):
+
+        if (binary_regex_search(RE_RELAX_PHONE)(line) +
+                binary_regex_search(RE_EMAIL)(line) +
+                binary_regex_search(RE_URL)(line) == 1):
             upvotes += 1
+
     if upvotes > 1:
         return True
+
+    return False

talon/utils.py

@@ -1,110 +1,17 @@
 # coding:utf-8
+from __future__ import annotations
 
-from __future__ import absolute_import
-
-from random import shuffle
-
-import cchardet
-import chardet
 import html5lib
 import regex as re
-import six
+from html5lib import HTMLParser
 from lxml.cssselect import CSSSelector
+from lxml.etree import _Element
 from lxml.html import html5parser
 
 from talon.constants import RE_DELIMITER
 
 
-def safe_format(format_string, *args, **kwargs):
-    """
-    Helper: formats string with any combination of bytestrings/unicode
-    strings without raising exceptions
-    """
-    try:
-        if not args and not kwargs:
-            return format_string
-        else:
-            return format_string.format(*args, **kwargs)
-
-    # catch encoding errors and transform everything into utf-8 string
-    # before logging:
-    except (UnicodeEncodeError, UnicodeDecodeError):
-        format_string = to_utf8(format_string)
-        args = [to_utf8(p) for p in args]
-        kwargs = {k: to_utf8(v) for k, v in six.iteritems(kwargs)}
-        return format_string.format(*args, **kwargs)
-
-    # ignore other errors
-    except:
-        return u''
-
-
-def to_unicode(str_or_unicode, precise=False):
-    """
-    Safely returns a unicode version of a given string
-    >>> utils.to_unicode('привет')
-        u'привет'
-    >>> utils.to_unicode(u'привет')
-        u'привет'
-    If `precise` flag is True, tries to guess the correct encoding first.
-    """
-    if not isinstance(str_or_unicode, six.text_type):
-        encoding = quick_detect_encoding(str_or_unicode) if precise else 'utf-8'
-        return six.text_type(str_or_unicode, encoding, 'replace')
-    return str_or_unicode
-
-
-def detect_encoding(string):
-    """
-    Tries to detect the encoding of the passed string.
-
-    Defaults to UTF-8.
-    """
-    assert isinstance(string, bytes)
-    try:
-        detected = chardet.detect(string)
-        if detected:
-            return detected.get('encoding') or 'utf-8'
-    except Exception as e:
-        pass
-    return 'utf-8'
-
-
-def quick_detect_encoding(string):
-    """
-    Tries to detect the encoding of the passed string.
-
-    Uses cchardet. Fallbacks to detect_encoding.
-    """
-    assert isinstance(string, bytes)
-    try:
-        detected = cchardet.detect(string)
-        if detected:
-            return detected.get('encoding') or detect_encoding(string)
-    except Exception as e:
-        pass
-    return detect_encoding(string)
-
-
-def to_utf8(str_or_unicode):
-    """
-    Safely returns a UTF-8 version of a given string
-    >>> utils.to_utf8(u'hi')
-        'hi'
-    """
-    if not isinstance(str_or_unicode, six.text_type):
-        return str_or_unicode.encode("utf-8", "ignore")
-    return str(str_or_unicode)
-
-
-def random_token(length=7):
-    vals = ("a b c d e f g h i j k l m n o p q r s t u v w x y z "
-            "0 1 2 3 4 5 6 7 8 9").split(' ')
-    shuffle(vals)
-    return ''.join(vals[:length])
-
-
-def get_delimiter(msg_body):
+def get_delimiter(msg_body: str) -> str:
     delimiter = RE_DELIMITER.search(msg_body)
     if delimiter:
         delimiter = delimiter.group()
@@ -114,7 +21,7 @@ def get_delimiter(msg_body):
     return delimiter
 
 
-def html_tree_to_text(tree):
+def html_tree_to_text(tree: _Element) -> str:
     for style in CSSSelector('style')(tree):
         style.getparent().remove(style)
 
@@ -131,7 +38,7 @@ def html_tree_to_text(tree):
     for el in tree.iter():
         el_text = (el.text or '') + (el.tail or '')
         if len(el_text) > 1:
-            if el.tag in _BLOCKTAGS:
+            if el.tag in _BLOCKTAGS + _HARDBREAKS:
                 text += "\n"
             if el.tag == 'li':
                 text += "  * "
@@ -142,29 +49,26 @@ def html_tree_to_text(tree):
             if href:
                 text += "(%s) " % href
 
-        if el.tag in _HARDBREAKS and text and not text.endswith("\n"):
+        if (el.tag in _HARDBREAKS and text and
+            not text.endswith("\n") and not el_text):
             text += "\n"
 
-    retval = _rm_excessive_newlines(text)
-    return _encode_utf8(retval)
+    text = _rm_excessive_newlines(text)
+    return text
 
 
-def html_to_text(string):
+def html_to_text(s: str) -> str | None:
     """
     Dead-simple HTML-to-text converter:
         >>> html_to_text("one<br>two<br>three")
-        >>> "one\ntwo\nthree"
+        <<< "one\ntwo\nthree"
 
     NOTES:
         1. the string is expected to contain UTF-8 encoded HTML!
-        2. returns utf-8 encoded str (not unicode)
         3. if html can't be parsed returns None
     """
-    if isinstance(string, six.text_type):
-        string = string.encode('utf8')
-
-    s = _prepend_utf8_declaration(string)
-    s = s.replace(b"\n", b"")
+    s = _prepend_utf8_declaration(s)
+    s = s.replace("\n", "")
     tree = html_fromstring(s)
 
     if tree is None:
@@ -173,74 +77,46 @@ def html_to_text(string):
     return html_tree_to_text(tree)
 
 
-def html_fromstring(s):
+def html_fromstring(s: str) -> _Element:
     """Parse html tree from string. Return None if the string can't be parsed.
     """
-    if isinstance(s, six.text_type):
-        s = s.encode('utf8')
-    try:
-        if html_too_big(s):
-            return None
-
-        return html5parser.fromstring(s, parser=_html5lib_parser())
-    except Exception:
-        pass
+    return html5parser.fromstring(s, parser=_html5lib_parser())
 
 
-def html_document_fromstring(s):
+def html_document_fromstring(s: str) -> _Element:
     """Parse html tree from string. Return None if the string can't be parsed.
     """
-    if isinstance(s, six.text_type):
-        s = s.encode('utf8')
-    try:
-        if html_too_big(s):
-            return None
-
-        return html5parser.document_fromstring(s, parser=_html5lib_parser())
-    except Exception:
-        pass
+    return html5parser.document_fromstring(s, parser=_html5lib_parser())
 
 
-def cssselect(expr, tree):
+def cssselect(expr: str, tree: str) -> list[_Element]:
     return CSSSelector(expr)(tree)
 
 
-def html_too_big(s):
-    if isinstance(s, six.text_type):
-        s = s.encode('utf8')
-    return s.count(b'<') > _MAX_TAGS_COUNT
-
-
-def _contains_charset_spec(s):
+def _contains_charset_spec(s: str) -> str:
     """Return True if the first 4KB contain charset spec
     """
-    return s.lower().find(b'html; charset=', 0, 4096) != -1
+    return s.lower().find('html; charset=', 0, 4096) != -1
 
 
-def _prepend_utf8_declaration(s):
+def _prepend_utf8_declaration(s: str) -> str:
     """Prepend 'utf-8' encoding declaration if the first 4KB don't have any
     """
     return s if _contains_charset_spec(s) else _UTF8_DECLARATION + s
 
 
-def _rm_excessive_newlines(s):
+def _rm_excessive_newlines(s: str) -> str:
     """Remove excessive newlines that often happen due to tons of divs
     """
     return _RE_EXCESSIVE_NEWLINES.sub("\n\n", s).strip()
 
 
-def _encode_utf8(s):
-    """Encode in 'utf-8' if unicode
-    """
-    return s.encode('utf-8') if isinstance(s, six.text_type) else s
-
-
-def _html5lib_parser():
+def _html5lib_parser() -> HTMLParser:
     """
     html5lib is a pure-python library that conforms to the WHATWG HTML spec
     and is not vulnarable to certain attacks common for XML libraries
     """
-    return html5lib.HTMLParser(
+    return HTMLParser(
         # build lxml tree
         html5lib.treebuilders.getTreeBuilder("lxml"),
         # remove namespace value from inside lxml.html.html5paser element tag
@@ -250,14 +126,10 @@ def _html5lib_parser():
     )
 
 
-_UTF8_DECLARATION = (b'<meta http-equiv="Content-Type" content="text/html;'
-                     b'charset=utf-8">')
+_UTF8_DECLARATION = ('<meta http-equiv="Content-Type" content="text/html;'
+                     'charset=utf-8">')
 
 _BLOCKTAGS = ['div', 'p', 'ul', 'li', 'h1', 'h2', 'h3']
 _HARDBREAKS = ['br', 'hr', 'tr']
 
 _RE_EXCESSIVE_NEWLINES = re.compile("\n{2,10}")
-
-# an extensive research shows that exceeding this limit
-# might lead to excessive processing time
-_MAX_TAGS_COUNT = 419

test-requirements.txt

@@ -0,0 +1,3 @@
+coverage
+mock
+nose>=1.2.1

tests/html_quotations_test.py

@@ -4,13 +4,17 @@ from __future__ import absolute_import
 
 # noinspection PyUnresolvedReferences
 import re
+from unittest.mock import Mock, patch
 
+from nose.tools import assert_false, assert_true, eq_, ok_
+
+from tests.fixtures import (OLK_SRC_BODY_SECTION,
+                            REPLY_QUOTATIONS_SHARE_BLOCK,
+                            REPLY_SEPARATED_BY_HR)
 from talon import quotations, utils as u
-from . import *
-from .fixtures import *
 
-RE_WHITESPACE = re.compile("\s")
-RE_DOUBLE_WHITESPACE = re.compile("\s")
+RE_WHITESPACE = re.compile(r"\s")
+RE_DOUBLE_WHITESPACE = re.compile(r"\s")
 
 
 def test_quotation_splitter_inside_blockquote():
@@ -165,7 +169,7 @@ def test_unicode_in_reply():
 
 <blockquote>
   Quote
-</blockquote>""".encode("utf-8")
+</blockquote>"""
 
     eq_("<html><head></head><body>Reply&#160;&#160;Text<br><div><br></div>"
         "</body></html>",
@@ -313,7 +317,6 @@ def extract_reply_and_check(filename):
     msg_body = f.read()
     reply = quotations.extract_from_html(msg_body)
     plain_reply = u.html_to_text(reply)
-    plain_reply = plain_reply.decode('utf8')
 
     eq_(RE_WHITESPACE.sub('', "Hi. I am fine.\n\nThanks,\nAlex"),
         RE_WHITESPACE.sub('', plain_reply))
@@ -390,18 +393,6 @@ def test_gmail_forwarded_msg():
     eq_(RE_WHITESPACE.sub('', msg_body), RE_WHITESPACE.sub('', extracted))
 
 
-@patch.object(u, '_MAX_TAGS_COUNT', 4)
-def test_too_large_html():
-    msg_body = 'Reply' \
-               '<div class="gmail_quote">' \
-               '<div class="gmail_quote">On 11-Apr-2011, at 6:54 PM, Bob &lt;bob@example.com&gt; wrote:' \
-               '<div>Test</div>' \
-               '</div>' \
-               '</div>'
-    eq_(RE_WHITESPACE.sub('', msg_body),
-        RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
-
-
 def test_readable_html_empty():
     msg_body = """
 <blockquote>
@@ -424,3 +415,23 @@ def test_readable_html_empty():
 def test_bad_html():
     bad_html = "<html></html>"
     eq_(bad_html, quotations.extract_from_html(bad_html))
+
+
+def test_remove_namespaces():
+    msg_body = """
+    <html xmlns:o="urn:schemas-microsoft-com:office:office" xmlns="http://www.w3.org/TR/REC-html40">
+        <body>
+            <o:p>Dear Sir,</o:p>
+            <o:p>Thank you for the email.</o:p>
+            <blockquote>thing</blockquote>
+        </body>
+    </html>
+    """
+
+    rendered = quotations.extract_from_html(msg_body)
+
+    assert_true("<p>" in rendered)
+    assert_true("xmlns" in rendered)
+
+    assert_true("<o:p>" not in rendered)
+    assert_true("<xmlns:o>" not in rendered)

tests/text_quotations_test.py

@@ -453,7 +453,8 @@ def test_link_closed_with_quotation_marker_on_new_line():
     msg_body = '''8.45am-1pm
 
 From: somebody@example.com
-
+Date: Wed, 16 May 2012 00:15:02 -0600
+ 
 <http://email.example.com/c/dHJhY2tpbmdfY29kZT1mMDdjYzBmNzM1ZjYzMGIxNT
 >  <bob@example.com <mailto:bob@example.com> >
 
@@ -494,7 +495,9 @@ def test_from_block_starts_with_date():
     msg_body = """Blah
 
 Date: Wed, 16 May 2012 00:15:02 -0600
-To: klizhentas@example.com"""
+To: klizhentas@example.com
+
+"""
     eq_('Blah', quotations.extract_from_plain(msg_body))
 
 
@@ -564,11 +567,12 @@ def test_mark_message_lines():
              # next line should be marked as splitter
              '_____________',
              'From: foo@bar.com',
+             'Date: Wed, 16 May 2012 00:15:02 -0600',
              '',
              '> Hi',
              '',
              'Signature']
-    eq_('tessemet', quotations.mark_message_lines(lines))
+    eq_('tesssemet', quotations.mark_message_lines(lines))
 
     lines = ['Just testing the email reply',
              '',
@@ -807,7 +811,7 @@ def test_split_email():
         >
         >
 """
-    expected_markers = "stttttsttttetesetesmmmmmmssmmmmmmsmmmmmmmm"
+    expected_markers = "stttttsttttetesetesmmmmmmsmmmmmmmmmmmmmmmm"
     markers = quotations.split_emails(msg)
     eq_(markers, expected_markers)
 
@@ -822,4 +826,16 @@ The user experience was unparallelled. Please continue production. I'm sending p
 that this line is intact."""
 
     parsed = quotations.extract_from_plain(msg_body)
-    eq_(msg_body, parsed.decode('utf8'))
+    eq_(msg_body, parsed)
+
+
+def test_appointment_2():
+    msg_body = """Invitation for an interview:
+
+Date: Wednesday 3, October 2011 
+Time: 7 : 00am 
+Address: 130 Fox St
+
+Please bring in your ID."""
+    parsed = quotations.extract_from_plain(msg_body)
+    eq_(msg_body, parsed)

tests/utils_test.py

@@ -2,9 +2,6 @@
 
 from __future__ import absolute_import
 
-import cchardet
-import six
-
 from talon import utils as u
 from . import *
 
@@ -15,58 +12,6 @@ def test_get_delimiter():
     eq_('\n', u.get_delimiter('abc'))
 
 
-def test_unicode():
-    eq_(u'hi', u.to_unicode('hi'))
-    eq_(type(u.to_unicode('hi')), six.text_type)
-    eq_(type(u.to_unicode(u'hi')), six.text_type)
-    eq_(type(u.to_unicode('привет')), six.text_type)
-    eq_(type(u.to_unicode(u'привет')), six.text_type)
-    eq_(u"привет", u.to_unicode('привет'))
-    eq_(u"привет", u.to_unicode(u'привет'))
-    # some latin1 stuff
-    eq_(u"Versión", u.to_unicode(u'Versi\xf3n'.encode('iso-8859-2'), precise=True))
-
-
-def test_detect_encoding():
-    eq_('ascii', u.detect_encoding(b'qwe').lower())
-    ok_(u.detect_encoding(
-        u'Versi\xf3n'.encode('iso-8859-2')).lower() in [
-            'iso-8859-1', 'iso-8859-2'])
-    eq_('utf-8', u.detect_encoding(u'привет'.encode('utf8')).lower())
-    # fallback to utf-8
-    with patch.object(u.chardet, 'detect') as detect:
-        detect.side_effect = Exception
-        eq_('utf-8', u.detect_encoding('qwe'.encode('utf8')).lower())
-
-
-def test_quick_detect_encoding():
-    eq_('ascii', u.quick_detect_encoding(b'qwe').lower())
-    ok_(u.quick_detect_encoding(
-        u'Versi\xf3n'.encode('windows-1252')).lower() in [
-            'windows-1252', 'windows-1250'])
-    eq_('utf-8', u.quick_detect_encoding(u'привет'.encode('utf8')).lower())
-
-
-@patch.object(cchardet, 'detect')
-@patch.object(u, 'detect_encoding')
-def test_quick_detect_encoding_edge_cases(detect_encoding, cchardet_detect):
-    cchardet_detect.return_value = {'encoding': 'ascii'}
-    eq_('ascii', u.quick_detect_encoding(b"qwe"))
-    cchardet_detect.assert_called_once_with(b"qwe")
-
-    # fallback to detect_encoding
-    cchardet_detect.return_value = {}
-    detect_encoding.return_value = 'utf-8'
-    eq_('utf-8', u.quick_detect_encoding(b"qwe"))
-
-    # exception
-    detect_encoding.reset_mock()
-    cchardet_detect.side_effect = Exception()
-    detect_encoding.return_value = 'utf-8'
-    eq_('utf-8', u.quick_detect_encoding(b"qwe"))
-    ok_(detect_encoding.called)
-
-
 def test_html_to_text():
     html = """<body>
 <p>Hello world!</p>
@@ -80,11 +25,11 @@ Haha
 </p>
 </body>"""
     text = u.html_to_text(html)
-    eq_(b"Hello world! \n\n  * One! \n  * Two \nHaha", text)
-    eq_(u"привет!", u.html_to_text("<b>привет!</b>").decode('utf8'))
+    eq_("Hello world! \n\n  * One! \n  * Two \nHaha", text)
+    eq_(u"привет!", u.html_to_text("<b>привет!</b>"))
 
     html = '<body><br/><br/>Hi</body>'
-    eq_(b'Hi', u.html_to_text(html))
+    eq_('Hi', u.html_to_text(html))
 
     html = """Hi
 <style type="text/css">
@@ -104,60 +49,23 @@ font: 13px 'Lucida Grande', Arial, sans-serif;
 
 }
 </style>"""
-    eq_(b'Hi', u.html_to_text(html))
+    eq_('Hi', u.html_to_text(html))
 
     html = """<div>
 <!-- COMMENT 1 -->
 <span>TEXT 1</span>
 <p>TEXT 2 <!-- COMMENT 2 --></p>
 </div>"""
-    eq_(b'TEXT 1 \nTEXT 2', u.html_to_text(html))
+    eq_('TEXT 1 \nTEXT 2', u.html_to_text(html))
 
 
 def test_comment_no_parent():
-    s = b'<!-- COMMENT 1 --> no comment'
+    s = '<!-- COMMENT 1 --> no comment'
     d = u.html_document_fromstring(s)
-    eq_(b"no comment", u.html_tree_to_text(d))
-
-
-@patch.object(u.html5parser, 'fromstring', Mock(side_effect=Exception()))
-def test_html_fromstring_exception():
-    eq_(None, u.html_fromstring("<html></html>"))
-
-
-@patch.object(u, 'html_too_big', Mock())
-@patch.object(u.html5parser, 'fromstring')
-def test_html_fromstring_too_big(fromstring):
-    eq_(None, u.html_fromstring("<html></html>"))
-    assert_false(fromstring.called)
-
-
-@patch.object(u.html5parser, 'document_fromstring')
-def test_html_document_fromstring_exception(document_fromstring):
-    document_fromstring.side_effect = Exception()
-    eq_(None, u.html_document_fromstring("<html></html>"))
-
-
-@patch.object(u, 'html_too_big', Mock())
-@patch.object(u.html5parser, 'document_fromstring')
-def test_html_document_fromstring_too_big(document_fromstring):
-    eq_(None, u.html_document_fromstring("<html></html>"))
-    assert_false(document_fromstring.called)
+    eq_("no comment", u.html_tree_to_text(d))
 
 
 @patch.object(u, 'html_fromstring', Mock(return_value=None))
 def test_bad_html_to_text():
     bad_html = "one<br>two<br>three"
     eq_(None, u.html_to_text(bad_html))
-
-
-@patch.object(u, '_MAX_TAGS_COUNT', 3)
-def test_html_too_big():
-    eq_(False, u.html_too_big("<div></div>"))
-    eq_(True, u.html_too_big("<div><span>Hi</span></div>"))
-
-
-@patch.object(u, '_MAX_TAGS_COUNT', 3)
-def test_html_to_text():
-    eq_(b"Hello", u.html_to_text("<div>Hello</div>"))
-    eq_(None, u.html_to_text("<div><span>Hi</span></div>"))