Merge pull request #183 from mailgun/sergey/date

fix text with Date: misclassified as quotations splitter

.gitignore

@@ -39,6 +39,8 @@ nosetests.xml
 /.emacs.desktop
 /.emacs.desktop.lock
 .elc
+.idea
+.cache
 auto-save-list
 tramp
 .\#*

MANIFEST.in

@@ -1,5 +1,3 @@
-recursive-include tests *
-recursive-include talon *
 recursive-exclude tests *.pyc *~
 recursive-exclude talon *.pyc *~
 include train.data

README.rst

@@ -95,7 +95,7 @@ classifiers. The core of machine learning algorithm lays in
 apply to a message (``featurespace.py``), how data sets are built
 (``dataset.py``), classifier’s interface (``classifier.py``).
 
-The data used for training is taken from our personal email
+Currently the data used for training is taken from our personal email
 conversations and from `ENRON`_ dataset. As a result of applying our set
 of features to the dataset we provide files ``classifier`` and
 ``train.data`` that don’t have any personal information but could be
@@ -116,8 +116,35 @@ or
     from talon.signature.learning.classifier import train, init
     train(init(), EXTRACTOR_DATA, EXTRACTOR_FILENAME)
 
+Open-source Dataset
+-------------------
+
+Recently we started a `forge`_ project to create an open-source, annotated dataset of raw emails. In the project we
+used a subset of `ENRON`_ data, cleansed of private, health and financial information by `EDRM`_. At the moment over 190
+emails are annotated. Any contribution and collaboration on the project are welcome. Once the dataset is ready we plan to
+start using it for talon.
+
 .. _scikit-learn: http://scikit-learn.org
 .. _ENRON: https://www.cs.cmu.edu/~enron/
+.. _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
 --------

setup.py

@@ -1,8 +1,35 @@
+from __future__ import absolute_import
 from setuptools import setup, find_packages
+from setuptools.command.install import install
+
+
+class InstallCommand(install):
+    user_options = install.user_options + [
+        ('no-ml', None, "Don't install without Machine Learning modules."),
+    ]
+
+    boolean_options = install.boolean_options + ['no-ml']
+
+    def initialize_options(self):
+        install.initialize_options(self)
+        self.no_ml = None
+
+    def finalize_options(self):
+        install.finalize_options(self)
+        if self.no_ml:
+            dist = self.distribution
+            dist.packages=find_packages(exclude=[
+                'tests',
+                'tests.*',
+                'talon.signature',
+                'talon.signature.*',
+            ])
+            for not_required in ['numpy', 'scipy', 'scikit-learn==0.16.1']:
+                dist.install_requires.remove(not_required)
 
 
 setup(name='talon',
-      version='1.2.4',
+      version='1.4.7',
       description=("Mailgun library "
                    "to extract message quotations and signatures."),
       long_description=open("README.rst").read(),
@@ -10,7 +37,10 @@ setup(name='talon',
       author_email='admin@mailgunhq.com',
       url='https://github.com/mailgun/talon',
       license='APACHE2',
-      packages=find_packages(exclude=['tests']),
+      cmdclass={
+          'install': InstallCommand,
+      },
+      packages=find_packages(exclude=['tests', 'tests.*']),
       include_package_data=True,
       zip_safe=True,
       install_requires=[
@@ -18,10 +48,12 @@ setup(name='talon',
           "regex>=1",
           "numpy",
           "scipy",
-          "scikit-learn==0.16.1", # pickled versions of classifier, else rebuild
+          "scikit-learn>=0.16.1", # pickled versions of classifier, else rebuild
           'chardet>=1.0.1',
           'cchardet>=0.3.5',
-          'cssselect'
+          'cssselect',
+          'six>=1.10.0',
+          'html5lib'
           ],
       tests_require=[
           "mock",

talon/__init__.py

@@ -1,7 +1,13 @@
+from __future__ import absolute_import
 from talon.quotations import register_xpath_extensions
+try:
     from talon import signature
+    ML_ENABLED = True
+except ImportError:
+    ML_ENABLED = False
 
 
 def init():
     register_xpath_extensions()
+    if ML_ENABLED:
         signature.initialize()

talon/constants.py

@@ -1,3 +1,4 @@
+from __future__ import absolute_import
 import regex as re
 
 

talon/html_quotations.py

@@ -3,8 +3,10 @@ The module's functions operate on message bodies trying to extract original
 messages (without quoted messages) from html
 """
 
+from __future__ import absolute_import
 import regex as re
 
+from talon.utils import cssselect 
 
 CHECKPOINT_PREFIX = '#!%!'
 CHECKPOINT_SUFFIX = '!%!#'
@@ -77,22 +79,32 @@ def delete_quotation_tags(html_note, counter, quotation_checkpoints):
 
 def cut_gmail_quote(html_message):
     ''' Cuts the outermost block element with class gmail_quote. '''
-    gmail_quote = html_message.cssselect('div.gmail_quote')
+    gmail_quote = cssselect('div.gmail_quote', html_message)
-    if gmail_quote and not RE_FWD.match(gmail_quote[0].text):
+    if gmail_quote and (gmail_quote[0].text is None or not RE_FWD.match(gmail_quote[0].text)):
         gmail_quote[0].getparent().remove(gmail_quote[0])
         return True
 
 
 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
+        #outlook 2007, 2010, 2013 (international, american)
-        "//div[@style='border:none;border-top:solid #B5C4DF 1.0pt;"
+        "//div[@style[re:match(., 'border:none; ?border-top:solid #(E1E1E1|B5C4DF) 1.0pt; ?"
-        "padding:3.0pt 0cm 0cm 0cm']|"
+        "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:
@@ -131,7 +143,7 @@ def cut_microsoft_quote(html_message):
 def cut_by_id(html_message):
     found = False
     for quote_id in QUOTE_IDS:
-        quote = html_message.cssselect('#{}'.format(quote_id))
+        quote = cssselect('#{}'.format(quote_id), html_message)
         if quote:
             found = True
             quote[0].getparent().remove(quote[0])
@@ -175,7 +187,21 @@ def cut_from_block(html_message):
                 len(maybe_body.getchildren()) == 1)
 
             if not parent_div_is_all_content:
-                block.getparent().remove(block)
+                parent = block.getparent()
+                next_sibling = block.getnext()
+
+                # remove all tags after found From block
+                # (From block and quoted message are in separate divs)
+                while next_sibling is not None:
+                    parent.remove(block)
+                    block = next_sibling
+                    next_sibling = block.getnext()
+
+                # remove the last sibling (or the
+                # From block if no siblings)
+                if block is not None:
+                    parent.remove(block)
+
                 return True
         else:
             return False

talon/quotations.py

@@ -5,20 +5,24 @@ The module's functions operate on message bodies trying to extract
 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_to_text
+from talon.utils import (get_delimiter, html_tree_to_text,
+                         html_document_fromstring)
 from talon import html_quotations
+from six.moves import range
+import six
 
 
 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(
@@ -34,10 +38,14 @@ RE_ON_DATE_SMB_WROTE = re.compile(
             'Op',
             # German
             'Am',
+            # Portuguese
+            'Em',
             # Norwegian
             u'På',
             # Swedish, Danish
             'Den',
+            # Vietnamese
+            u'Vào',
         )),
         # Date and sender separator
         u'|'.join((
@@ -58,8 +66,12 @@ RE_ON_DATE_SMB_WROTE = re.compile(
             'schreef','verzond','geschreven',
             # German
             'schrieb',
+            # Portuguese
+            'escreveu',
             # Norwegian, Swedish
             'skrev',
+            # Vietnamese
+            u'đã viết',
         ))
     ))
 # Special case for languages where text is translated like this: 'on {date} wrote {somebody}:'
@@ -127,14 +139,33 @@ 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',
+        '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(
+    u'|'.join((
+        # English
+        'wrote',
     ))), re.I)
 
+# Support polymail.io reply format
+# On Tue, Apr 11, 2017 at 10:07 PM John Smith
+#
+# <
+# mailto:John Smith <johnsmith@gmail.com>
+# > wrote:
+RE_POLYMAIL = re.compile('On.*\s{2}<\smailto:.*\s> wrote:', re.I)
+
 SPLITTER_PATTERNS = [
     RE_ORIGINAL_MESSAGE,
     RE_ON_DATE_SMB_WROTE,
@@ -142,27 +173,36 @@ SPLITTER_PATTERNS = [
     RE_FROM_COLON_OR_DATE_COLON,
     # 02.04.2012 14:20 пользователь "bob@example.com" <
     # bob@xxx.mailgun.org> написал:
-    re.compile("(\d+/\d+/\d+|\d+\.\d+\.\d+).*@", re.S),
+    re.compile("(\d+/\d+/\d+|\d+\.\d+\.\d+).*\s\S+@\S+", re.S),
     # 2014-10-17 11:28 GMT+03:00 Bob <
     # bob@example.com>:
-    re.compile("\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+GMT.*@", re.S),
+    re.compile("\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+GMT.*\s\S+@\S+", re.S),
     # Thu, 26 Jun 2014 14:00:51 +0400 Bob <bob@example.com>:
     re.compile('\S{3,10}, \d\d? \S{3,10} 20\d\d,? \d\d?:\d\d(:\d\d)?'
-               '( \S+){3,6}@\S+:')
+               '( \S+){3,6}@\S+:'),
+    # Sent from Samsung MobileName <address@example.com> wrote:
+    re.compile('Sent from Samsung.* \S+@\S+> wrote'),
+    RE_ANDROID_WROTE,
+    RE_POLYMAIL
     ]
 
-
 RE_LINK = re.compile('<(http://[^>]*)>')
 RE_NORMALIZED_LINK = re.compile('@@(http://[^>@]*)@@')
 
 RE_PARENTHESIS_LINK = re.compile("\(https?://")
 
-SPLITTER_MAX_LINES = 4
+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].*')
 
+# Regular expression to identify if a line is a header.
+RE_HEADER = re.compile(": ")
+
 
 def extract_from(msg_body, content_type='text/plain'):
     try:
@@ -176,6 +216,19 @@ def extract_from(msg_body, content_type='text/plain'):
     return msg_body
 
 
+def remove_initial_spaces_and_mark_message_lines(lines):
+    """
+    Removes the initial spaces in each line before marking message lines.
+
+    This ensures headers can be identified if they are indented with spaces.
+    """
+    i = 0
+    while i < len(lines):
+        lines[i] = lines[i].lstrip(' ')
+        i += 1
+    return mark_message_lines(lines)
+
+
 def mark_message_lines(lines):
     """Mark message lines with markers to distinguish quotation lines.
 
@@ -189,7 +242,7 @@ def mark_message_lines(lines):
     >>> mark_message_lines(['answer', 'From: foo@bar.com', '', '> question'])
     'tsem'
     """
-    markers = bytearray(len(lines))
+    markers = ['e' for _ in lines]
     i = 0
     while i < len(lines):
         if not lines[i].strip():
@@ -205,7 +258,7 @@ def mark_message_lines(lines):
             if splitter:
                 # append as many splitter markers as lines in splitter
                 splitter_lines = splitter.group().splitlines()
-                for j in xrange(len(splitter_lines)):
+                for j in range(len(splitter_lines)):
                     markers[i + j] = 's'
 
                 # skip splitter lines
@@ -215,7 +268,7 @@ def mark_message_lines(lines):
                 markers[i] = 't'
         i += 1
 
-    return markers
+    return ''.join(markers)
 
 
 def process_marked_lines(lines, markers, return_flags=[False, -1, -1]):
@@ -229,6 +282,7 @@ def process_marked_lines(lines, markers, return_flags=[False, -1, -1]):
     return_flags = [were_lines_deleted, first_deleted_line,
                     last_deleted_line]
     """
+    markers = ''.join(markers)
     # if there are no splitter there should be no markers
     if 's' not in markers and not re.search('(me*){3}', markers):
         markers = markers.replace('m', 't')
@@ -240,7 +294,7 @@ 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 = (
@@ -274,10 +328,27 @@ def preprocess(msg_body, delimiter, content_type='text/plain'):
     Replaces link brackets so that they couldn't be taken for quotation marker.
     Splits line in two if splitter pattern preceded by some text on the same
     line (done only for 'On <date> <person> wrote:' pattern).
+
+    Converts msg_body into a unicode.
     """
-    # normalize links i.e. replace '<', '>' wrapping the link with some symbols
+    msg_body = _replace_link_brackets(msg_body)
-    # so that '>' closing the link couldn't be mistakenly taken for quotation
+
-    # marker.
+    msg_body = _wrap_splitter_with_newline(msg_body, delimiter, content_type)
+
+    return msg_body
+
+
+def _replace_link_brackets(msg_body):
+    """
+    Normalize links i.e. replace '<', '>' wrapping the link with some symbols
+    so that '>' closing the link couldn't be mistakenly taken for quotation
+    marker.
+
+    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] == ">":
@@ -286,7 +357,14 @@ def preprocess(msg_body, delimiter, content_type='text/plain'):
             return "@@%s@@" % link.group(1)
 
     msg_body = re.sub(RE_LINK, link_wrapper, msg_body)
+    return msg_body
 
+
+def _wrap_splitter_with_newline(msg_body, delimiter, content_type='text/plain'):
+    """
+    Splits line in two if splitter pattern preceded by some text on the same
+    line (done only for 'On <date> <person> wrote:' pattern.
+    """
     def splitter_wrapper(splitter):
         """Wraps splitter with new line"""
         if splitter.start() and msg_body[splitter.start() - 1] != '\n':
@@ -340,15 +418,52 @@ def extract_from_html(msg_body):
     then extracting quotations from text,
     then checking deleted checkpoints,
     then deleting necessary tags.
+
+    Returns a unicode string.
     """
-    if msg_body.strip() == '':
+    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')
+
+    result = _extract_from_html(msg_body)
+    if isinstance(result, bytes):
+        result = result.decode('utf8')
+
+    return result
+
+
+def _extract_from_html(msg_body):
+    """
+    Extract not quoted message from provided html message body
+    using tags and plain text algorithm.
+
+    Cut out first some encoding html tags such as xml and doctype
+    for avoiding conflict with unicode decoding
+
+    Cut out the 'blockquote', 'gmail_quote' tags.
+    Cut Microsoft quotations.
+
+    Then use plain text algorithm to cut out splitter or
+    leftover quotation.
+    This works by adding checkpoint text to all html tags,
+    then converting html to text,
+    then extracting quotations from text,
+    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')
+
+    msg_body = re.sub(r"\<\?xml.+\?\>|\<\!DOCTYPE.+]\>", "", msg_body)
+
+    html_tree = html_document_fromstring(msg_body)
+
+    if html_tree is None:
         return msg_body
 
-    msg_body = msg_body.replace('\r\n', '').replace('\n', '')
-    html_tree = html.document_fromstring(
-        msg_body,
-        parser=html.HTMLParser(encoding="utf-8")
-    )
     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
@@ -360,8 +475,7 @@ def extract_from_html(msg_body):
 
     number_of_checkpoints = html_quotations.add_checkpoint(html_tree, 0)
     quotation_checkpoints = [False] * number_of_checkpoints
-    msg_with_checkpoints = html.tostring(html_tree)
+    plain_text = html_tree_to_text(html_tree)
-    plain_text = html_to_text(msg_with_checkpoints)
     plain_text = preprocess(plain_text, '\n', content_type='text/html')
     lines = plain_text.splitlines()
 
@@ -384,25 +498,106 @@ def extract_from_html(msg_body):
     return_flags = []
     process_marked_lines(lines, markers, return_flags)
     lines_were_deleted, first_deleted, last_deleted = return_flags
+
+    if not lines_were_deleted and not cut_quotations:
+        return msg_body
+
     if lines_were_deleted:
         #collect checkpoints from deleted lines
-        for i in xrange(first_deleted, last_deleted):
+        for i in range(first_deleted, last_deleted):
             for checkpoint in line_checkpoints[i]:
                 quotation_checkpoints[checkpoint] = True
-    else:
-        if cut_quotations:
-            return html.tostring(html_tree_copy)
-        else:
-            return msg_body
 
         # Remove tags with quotation checkpoints
         html_quotations.delete_quotation_tags(
             html_tree_copy, 0, quotation_checkpoints
         )
 
+    if _readable_text_empty(html_tree_copy):
+        return msg_body
+
     return html.tostring(html_tree_copy)
 
 
+def split_emails(msg):
+    """
+    Given a message (which may consist of an email conversation thread with
+    multiple emails), mark the lines to identify split lines, content lines and
+    empty lines.
+
+    Correct the split line markers inside header blocks. Header blocks are
+    identified by the regular expression RE_HEADER.
+
+    Return the corrected markers
+    """
+    msg_body = _replace_link_brackets(msg)
+
+    # don't process too long messages
+    lines = msg_body.splitlines()[:MAX_LINES_COUNT]
+    markers = remove_initial_spaces_and_mark_message_lines(lines)
+
+    markers = _mark_quoted_email_splitlines(markers, lines)
+
+    # we don't want splitlines in header blocks
+    markers = _correct_splitlines_in_headers(markers, lines)
+
+    return markers
+
+
+def _mark_quoted_email_splitlines(markers, lines):
+    """
+    When there are headers indented with '>' characters, this method will
+    attempt to identify if the header is a splitline header. If it is, then we
+    mark it with 's' instead of leaving it as 'm' and return the new markers.
+    """
+    # Create a list of markers to easily alter specific characters
+    markerlist = list(markers)
+    for i, line in enumerate(lines):
+        if markerlist[i] != 'm':
+            continue
+        for pattern in SPLITTER_PATTERNS:
+            matcher = re.search(pattern, line)
+            if matcher:
+                markerlist[i] = 's'
+                break
+
+    return "".join(markerlist)
+
+
+def _correct_splitlines_in_headers(markers, lines):
+    """
+    Corrects markers by removing splitlines deemed to be inside header blocks.
+    """
+    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':
+            if not in_header_block:
+                if bool(re.search(RE_HEADER, lines[i])):
+                    in_header_block = True
+            else:
+                if QUOT_PATTERN.match(lines[i]):
+                    m = 'm'
+                else:
+                    m = 't'
+
+        # If the line is not a header line, set in_header_block false.
+        if not bool(re.search(RE_HEADER, lines[i])):
+            in_header_block = False
+
+        # Add the marker to the new updated markers string.
+        updated_markers += m
+        i += 1
+
+    return updated_markers
+
+
+def _readable_text_empty(html_tree):
+    return not bool(html_tree_to_text(html_tree).strip())
+
+
 def is_splitter(line):
     '''
     Returns Matcher object if provided string is a splitter and
@@ -416,7 +611,7 @@ def is_splitter(line):
 
 def text_content(context):
     '''XPath Extension function to return a node text content.'''
-    return context.context_node.text_content().strip()
+    return context.context_node.xpath("string()").strip()
 
 
 def tail(context):

talon/signature/__init__.py

@@ -20,6 +20,7 @@ trained against, don't forget to regenerate:
 * signature/data/classifier
 """
 
+from __future__ import absolute_import
 import os
 
 from . import extraction

talon/signature/bruteforce.py

@@ -1,14 +1,15 @@
+from __future__ import absolute_import
+
 import logging
 
 import regex as re
 
-from talon.utils import get_delimiter
 from talon.signature.constants import (SIGNATURE_MAX_LINES,
                                        TOO_LONG_SIGNATURE_LINE)
+from talon.utils import get_delimiter
 
 log = logging.getLogger(__name__)
 
-
 # regex to fetch signature based on common signature words
 RE_SIGNATURE = re.compile(r'''
                (
@@ -27,7 +28,6 @@ RE_SIGNATURE = re.compile(r'''
                )
                ''', re.I | re.X | re.M | re.S)
 
-
 # signatures appended by phone email clients
 RE_PHONE_SIGNATURE = re.compile(r'''
                (
@@ -44,7 +44,6 @@ RE_PHONE_SIGNATURE = re.compile(r'''
                )
                ''', re.I | re.X | re.M | re.S)
 
-
 # see _mark_candidate_indexes() for details
 # c - could be signature line
 # d - line starts with dashes (could be signature or list item)
@@ -111,7 +110,7 @@ def extract_signature(msg_body):
 
             return (stripped_body.strip(),
                     signature.strip())
-    except Exception, e:
+    except Exception:
         log.exception('ERROR extracting signature')
         return (msg_body, None)
 
@@ -162,7 +161,7 @@ def _mark_candidate_indexes(lines, candidate):
     'cdc'
     """
     # at first consider everything to be potential signature lines
-    markers = bytearray('c'*len(candidate))
+    markers = list('c' * len(candidate))
 
     # mark lines starting from bottom up
     for i, line_idx in reversed(list(enumerate(candidate))):
@@ -173,7 +172,7 @@ def _mark_candidate_indexes(lines, candidate):
             if line.startswith('-') and line.strip("-"):
                 markers[i] = 'd'
 
-    return markers
+    return "".join(markers)
 
 
 def _process_marked_candidate_indexes(candidate, markers):

talon/signature/extraction.py

@@ -1,15 +1,15 @@
 # -*- coding: utf-8 -*-
 
+from __future__ import absolute_import
+
 import logging
 
-import regex as re
 import numpy
-
+import regex as re
-from talon.signature.learning.featurespace import features, build_pattern
-from talon.utils import get_delimiter
 from talon.signature.bruteforce import get_signature_candidate
+from talon.signature.learning.featurespace import features, build_pattern
 from talon.signature.learning.helpers import has_signature
-
+from talon.utils import get_delimiter
 
 log = logging.getLogger(__name__)
 
@@ -32,7 +32,7 @@ RE_REVERSE_SIGNATURE = re.compile(r'''
 
 def is_signature_line(line, sender, classifier):
     '''Checks if the line belongs to signature. Returns True or False.'''
-    data = numpy.array(build_pattern(line, features(sender)))
+    data = numpy.array(build_pattern(line, features(sender))).reshape(1, -1)
     return classifier.predict(data) > 0
 
 
@@ -57,7 +57,7 @@ def extract(body, sender):
                 text = delimiter.join(text)
                 if text.strip():
                     return (text, delimiter.join(signature))
-    except Exception:
+    except Exception as e:
         log.exception('ERROR when extracting signature with classifiers')
 
     return (body, None)
@@ -80,7 +80,7 @@ def _mark_lines(lines, sender):
     candidate = get_signature_candidate(lines)
 
     # at first consider everything to be text no signature
-    markers = bytearray('t'*len(lines))
+    markers = list('t' * len(lines))
 
     # mark lines starting from bottom up
     # mark only lines that belong to candidate
@@ -95,7 +95,7 @@ def _mark_lines(lines, sender):
         elif is_signature_line(line, sender, EXTRACTOR):
             markers[j] = 's'
 
-    return markers
+    return "".join(markers)
 
 
 def _process_marked_lines(lines, markers):
@@ -110,3 +110,4 @@ def _process_marked_lines(lines, markers):
         return (lines[:-signature.end()], lines[-signature.end():])
 
     return (lines, None)
+

talon/signature/learning/classifier.py

@@ -5,9 +5,11 @@ The classifier could be used to detect if a certain line of the message
 body belongs to the signature.
 """
 
+from __future__ import absolute_import
+
 from numpy import genfromtxt
-from sklearn.svm import LinearSVC
 from sklearn.externals import joblib
+from sklearn.svm import LinearSVC
 
 
 def init():
@@ -28,4 +30,40 @@ def train(classifier, train_data_filename, save_classifier_filename=None):
 
 def load(saved_classifier_filename, train_data_filename):
     """Loads saved classifier. """
+    try:
         return joblib.load(saved_classifier_filename)
+    except Exception:
+        import sys
+        if sys.version_info > (3, 0):
+            return load_compat(saved_classifier_filename)
+
+        raise
+
+
+def load_compat(saved_classifier_filename):
+    import os
+    import pickle
+    import tempfile
+
+    # we need to switch to the data path to properly load the related _xx.npy files
+    cwd = os.getcwd()
+    os.chdir(os.path.dirname(saved_classifier_filename))
+
+    # convert encoding using pick.load and write to temp file which we'll tell joblib to use
+    pickle_file = open(saved_classifier_filename, 'rb')
+    classifier = pickle.load(pickle_file, encoding='latin1')
+
+    try:
+        # save our conversion if permissions allow
+        joblib.dump(classifier, saved_classifier_filename)
+    except Exception:
+        # can't write to classifier, use a temp file
+        tmp = tempfile.SpooledTemporaryFile()
+        joblib.dump(classifier, tmp)
+        saved_classifier_filename = tmp
+
+    # important, use joblib.load before switching back to original cwd
+    jb_classifier = joblib.load(saved_classifier_filename)
+    os.chdir(cwd)
+
+    return jb_classifier

talon/signature/learning/dataset.py

@@ -16,13 +16,16 @@ suffix and the corresponding sender file has the same name except for the
 suffix which should be `_sender`.
 """
 
+from __future__ import absolute_import
+
 import os
+
 import regex as re
+from six.moves import range
 
 from talon.signature.constants import SIGNATURE_MAX_LINES
 from talon.signature.learning.featurespace import build_pattern, features
 
-
 SENDER_SUFFIX = '_sender'
 BODY_SUFFIX = '_body'
 
@@ -55,9 +58,14 @@ def parse_msg_sender(filename, sender_known=True):
     algorithm:
     >>> parse_msg_sender(filename, False)
     """
+    import sys
+    kwargs = {}
+    if sys.version_info > (3, 0):
+        kwargs["encoding"] = "utf8"
+
     sender, msg = None, None
     if os.path.isfile(filename) and not is_sender_filename(filename):
-        with open(filename) as f:
+        with open(filename, **kwargs) as f:
             msg = f.read()
             sender = u''
             if sender_known:
@@ -144,7 +152,7 @@ def build_extraction_dataset(folder, dataset_filename,
             if not sender or not msg:
                 continue
             lines = msg.splitlines()
-            for i in xrange(1, min(SIGNATURE_MAX_LINES,
+            for i in range(1, min(SIGNATURE_MAX_LINES,
                                   len(lines)) + 1):
                 line = lines[-i]
                 label = -1

talon/signature/learning/featurespace.py

@@ -7,9 +7,12 @@ The body and the message sender string are converted into unicode before
 applying features to them.
 """
 
+from __future__ import absolute_import
 from talon.signature.constants import (SIGNATURE_MAX_LINES,
                                        TOO_LONG_SIGNATURE_LINE)
 from talon.signature.learning.helpers import *
+from six.moves import zip
+from functools import reduce
 
 
 def features(sender=''):

talon/signature/learning/helpers.py

@@ -6,6 +6,7 @@
 
 """
 
+from __future__ import absolute_import
 import unicodedata
 import regex as re
 
@@ -184,12 +185,13 @@ def capitalized_words_percent(s):
     s = to_unicode(s, precise=True)
     words = re.split('\s', s)
     words = [w for w in words if w.strip()]
+    words = [w for w in words if len(w) > 2]    
     capitalized_words_counter = 0
     valid_words_counter = 0
     for word in words:
         if not INVALID_WORD_START.match(word):
             valid_words_counter += 1
-            if word[0].isupper():
+            if word[0].isupper() and not word[1].isupper():
                 capitalized_words_counter += 1
     if valid_words_counter > 0 and len(words) > 1:
         return 100 * float(capitalized_words_counter) / valid_words_counter

talon/utils.py

@@ -1,13 +1,16 @@
 # coding:utf-8
 
-import logging
+from __future__ import absolute_import
-from random import shuffle
-import chardet
-import cchardet
-import regex as re
 
-from lxml import html
+from random import shuffle
+
+import cchardet
+import chardet
+import html5lib
+import regex as re
+import six
 from lxml.cssselect import CSSSelector
+from lxml.html import html5parser
 
 from talon.constants import RE_DELIMITER
 
@@ -28,7 +31,7 @@ def safe_format(format_string, *args, **kwargs):
     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 kwargs.iteritems()}
+        kwargs = {k: to_utf8(v) for k, v in six.iteritems(kwargs)}
         return format_string.format(*args, **kwargs)
 
     # ignore other errors
@@ -45,9 +48,9 @@ def to_unicode(str_or_unicode, precise=False):
         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'
-    if isinstance(str_or_unicode, str):
+        return six.text_type(str_or_unicode, encoding, 'replace')
-        return unicode(str_or_unicode, encoding, 'replace')
     return str_or_unicode
 
 
@@ -57,11 +60,12 @@ def detect_encoding(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, e:
+    except Exception as e:
         pass
     return 'utf-8'
 
@@ -72,11 +76,12 @@ def quick_detect_encoding(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, e:
+    except Exception as e:
         pass
     return detect_encoding(string)
 
@@ -87,7 +92,7 @@ def to_utf8(str_or_unicode):
     >>> utils.to_utf8(u'hi')
         'hi'
     """
-    if isinstance(str_or_unicode, unicode):
+    if not isinstance(str_or_unicode, six.text_type):
         return str_or_unicode.encode("utf-8", "ignore")
     return str(str_or_unicode)
 
@@ -109,32 +114,24 @@ def get_delimiter(msg_body):
     return delimiter
 
 
-def html_to_text(string):
+def html_tree_to_text(tree):
-    """
-    Dead-simple HTML-to-text converter:
-        >>> html_to_text("one<br>two<br>three")
-        >>> "one\ntwo\nthree"
-
-    NOTES:
-        1. the string is expected to contain UTF-8 encoded HTML!
-        2. returns utf-8 encoded str (not unicode)
-    """
-    s = _prepend_utf8_declaration(string)
-    s = s.replace("\n", "")
-
-    tree = html.fromstring(s)
-
     for style in CSSSelector('style')(tree):
         style.getparent().remove(style)
 
     for c in tree.xpath('//comment()'):
-        c.getparent().remove(c)
+        parent = c.getparent()
+
+        # comment with no parent does not impact produced text
+        if parent is None:
+            continue
+
+        parent.remove(c)
 
     text = ""
     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 += "  * "
@@ -145,17 +142,80 @@ def html_to_text(string):
             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)
 
 
+def html_to_text(string):
+    """
+    Dead-simple HTML-to-text converter:
+        >>> html_to_text("one<br>two<br>three")
+        >>> "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"")
+    tree = html_fromstring(s)
+
+    if tree is None:
+        return None
+
+    return html_tree_to_text(tree)
+
+
+def html_fromstring(s):
+    """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
+
+
+def html_document_fromstring(s):
+    """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
+
+
+def cssselect(expr, tree):
+    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):
     """Return True if the first 4KB contain charset spec
     """
-    return s.lower().find('html; charset=', 0, 4096) != -1
+    return s.lower().find(b'html; charset=', 0, 4096) != -1
 
 
 def _prepend_utf8_declaration(s):
@@ -173,15 +233,32 @@ def _rm_excessive_newlines(s):
 def _encode_utf8(s):
     """Encode in 'utf-8' if unicode
     """
-    return s.encode('utf-8') if isinstance(s, unicode) else s
+    return s.encode('utf-8') if isinstance(s, six.text_type) else s
 
 
-_UTF8_DECLARATION = ('<meta http-equiv="Content-Type" content="text/html;'
+def _html5lib_parser():
-                     'charset=utf-8">')
+    """
+    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(
+        # build lxml tree
+        html5lib.treebuilders.getTreeBuilder("lxml"),
+        # remove namespace value from inside lxml.html.html5paser element tag
+        # otherwise it yields something like "{http://www.w3.org/1999/xhtml}div"
+        # instead of "div", throwing the algo off
+        namespaceHTMLElements=False
+    )
 
 
+_UTF8_DECLARATION = (b'<meta http-equiv="Content-Type" content="text/html;'
+                     b'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

tests/__init__.py

@@ -1,3 +1,4 @@
+from __future__ import absolute_import
 from nose.tools import *
 from mock import *
 

tests/html_quotations_test.py

@@ -1,12 +1,13 @@
 # -*- coding: utf-8 -*-
 
-from . import *
+from __future__ import absolute_import
-from . fixtures import *
 
-import regex as re
+# noinspection PyUnresolvedReferences
+import re
 
 from talon import quotations, utils as u
-
+from . import *
+from .fixtures import *
 
 RE_WHITESPACE = re.compile("\s")
 RE_DOUBLE_WHITESPACE = re.compile("\s")
@@ -26,7 +27,7 @@ def test_quotation_splitter_inside_blockquote():
 
 </blockquote>"""
 
-    eq_("<html><body><p>Reply</p></body></html>",
+    eq_("<html><head></head><body>Reply</body></html>",
         RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
 
 
@@ -43,7 +44,7 @@ def test_quotation_splitter_outside_blockquote():
   </div>
 </blockquote>
 """
-    eq_("<html><body><p>Reply</p></body></html>",
+    eq_("<html><head></head><body>Reply</body></html>",
         RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
 
 
@@ -61,7 +62,7 @@ def test_regular_blockquote():
   </div>
 </blockquote>
 """
-    eq_("<html><body><p>Reply</p><blockquote>Regular</blockquote></body></html>",
+    eq_("<html><head></head><body>Reply<blockquote>Regular</blockquote></body></html>",
         RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
 
 
@@ -84,6 +85,7 @@ Reply
 
     reply = """
 <html>
+<head></head>
 <body>
 Reply
 
@@ -127,7 +129,18 @@ def test_gmail_quote():
     </div>
   </div>
 </div>"""
-    eq_("<html><body><p>Reply</p></body></html>",
+    eq_("<html><head></head><body>Reply</body></html>",
+        RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
+
+
+def test_gmail_quote_compact():
+    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_("<html><head></head><body>Reply</body></html>",
         RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
 
 
@@ -154,7 +167,7 @@ def test_unicode_in_reply():
   Quote
 </blockquote>""".encode("utf-8")
 
-    eq_("<html><body><p>Reply&#160;&#160;Text<br></p><div><br></div>"
+    eq_("<html><head></head><body>Reply&#160;&#160;Text<br><div><br></div>"
         "</body></html>",
         RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
 
@@ -180,6 +193,7 @@ def test_blockquote_disclaimer():
 
     stripped_html = """
 <html>
+  <head></head>
   <body>
   <div>
     <div>
@@ -211,7 +225,7 @@ def test_date_block():
   </div>
 </div>
 """
-    eq_('<html><body><div>message<br></div></body></html>',
+    eq_('<html><head></head><body><div>message<br></div></body></html>',
         RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
 
 
@@ -228,7 +242,7 @@ Subject: You Have New Mail From Mary!<br><br>
 text
 </div></div>
 """
-    eq_('<html><body><div>message<br></div></body></html>',
+    eq_('<html><head></head><body><div>message<br></div></body></html>',
         RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
 
 
@@ -246,7 +260,7 @@ def test_reply_shares_div_with_from_block():
 
   </div>
 </body>'''
-    eq_('<html><body><div>Blah<br><br></div></body></html>',
+    eq_('<html><head></head><body><div>Blah<br><br></div></body></html>',
         RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
 
 
@@ -257,23 +271,49 @@ def test_reply_quotations_share_block():
 
 
 def test_OLK_SRC_BODY_SECTION_stripped():
-    eq_('<html><body><div>Reply</div></body></html>',
+    eq_('<html><head></head><body><div>Reply</div></body></html>',
         RE_WHITESPACE.sub(
             '', quotations.extract_from_html(OLK_SRC_BODY_SECTION)))
 
 
 def test_reply_separated_by_hr():
-    eq_('<html><body><div>Hi<div>there</div></div></body></html>',
+    eq_('<html><head></head><body><div>Hi<div>there</div></div></body></html>',
         RE_WHITESPACE.sub(
             '', quotations.extract_from_html(REPLY_SEPARATED_BY_HR)))
 
 
+def test_from_block_and_quotations_in_separate_divs():
+    msg_body = '''
+Reply
+<div>
+  <hr/>
+  <div>
+    <font>
+      <b>From: bob@example.com</b>
+      <b>Date: Thu, 24 Mar 2016 08:07:12 -0700</b>
+    </font>
+  </div>
+  <div>
+    Quoted message
+  </div>
+</div>
+'''
+    eq_('<html><head></head><body>Reply<div><hr></div></body></html>',
+        RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
+
+
 def extract_reply_and_check(filename):
-    f = open(filename)
+    import sys
+    kwargs = {}
+    if sys.version_info > (3, 0):
+        kwargs["encoding"] = "utf8"
+
+    f = open(filename, **kwargs)
 
     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))
@@ -323,7 +363,8 @@ def test_CRLF():
     assert_false(symbol in extracted)
     eq_('<html></html>', RE_WHITESPACE.sub('', extracted))
 
-    msg_body = """Reply
+    msg_body = """My
+reply
 <blockquote>
 
   <div>
@@ -338,8 +379,8 @@ def test_CRLF():
     msg_body = msg_body.replace('\n', '\r\n')
     extracted = quotations.extract_from_html(msg_body)
     assert_false(symbol in extracted)
-    eq_("<html><body><p>Reply</p></body></html>",
+    # Keep new lines otherwise "My reply" becomes one word - "Myreply" 
-        RE_WHITESPACE.sub('', extracted))
+    eq_("<html><head></head><body>My\nreply\n</body></html>", extracted)
 
 
 def test_gmail_forwarded_msg():
@@ -347,3 +388,39 @@ def test_gmail_forwarded_msg():
 </div><br></div>"""
     extracted = quotations.extract_from_html(msg_body)
     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>
+  Reply
+  <div>
+    On 11-Apr-2011, at 6:54 PM, Bob &lt;bob@example.com&gt; wrote:
+  </div>
+
+  <div>
+    Test
+  </div>
+
+</blockquote>"""
+
+    eq_(RE_WHITESPACE.sub('', msg_body),
+        RE_WHITESPACE.sub('', quotations.extract_from_html(msg_body)))
+
+
+@patch.object(quotations, 'html_document_fromstring', Mock(return_value=None))
+def test_bad_html():
+    bad_html = "<html></html>"
+    eq_(bad_html, quotations.extract_from_html(bad_html))

tests/quotations_test.py

@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 
+from __future__ import absolute_import
 from . import *
 from . fixtures import *
 

tests/signature/bruteforce_test.py

@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 
+from __future__ import absolute_import
 from .. import *
 
 from talon.signature import bruteforce

tests/signature/extraction_test.py

@@ -1,13 +1,15 @@
 # -*- coding: utf-8 -*-
 
-from .. import *
+from __future__ import absolute_import
 
 import os
 
-from talon.signature.learning import dataset
+from six.moves import range
-from talon import signature
+
+from talon.signature import bruteforce, extraction, extract
 from talon.signature import extraction as e
-from talon.signature import bruteforce
+from talon.signature.learning import dataset
+from .. import *
 
 
 def test_message_shorter_SIGNATURE_MAX_LINES():
@@ -16,20 +18,25 @@ def test_message_shorter_SIGNATURE_MAX_LINES():
 
 Thanks in advance,
 Bob"""
-    text, extracted_signature = signature.extract(body, sender)
+    text, extracted_signature = extract(body, sender)
     eq_('\n'.join(body.splitlines()[:2]), text)
     eq_('\n'.join(body.splitlines()[-2:]), extracted_signature)
 
 
 def test_messages_longer_SIGNATURE_MAX_LINES():
+    import sys
+    kwargs = {}
+    if sys.version_info > (3, 0):
+        kwargs["encoding"] = "utf8"
+
     for filename in os.listdir(STRIPPED):
         filename = os.path.join(STRIPPED, filename)
         if not filename.endswith('_body'):
             continue
         sender, body = dataset.parse_msg_sender(filename)
-        text, extracted_signature = signature.extract(body, sender)
+        text, extracted_signature = extract(body, sender)
         extracted_signature = extracted_signature or ''
-        with open(filename[:-len('body')] + 'signature') as ms:
+        with open(filename[:-len('body')] + 'signature', **kwargs) as ms:
             msg_signature = ms.read()
             eq_(msg_signature.strip(), extracted_signature.strip())
             stripped_msg = body.strip()[:len(body.strip()) - len(msg_signature)]
@@ -45,7 +52,7 @@ Thanks in advance,
 some text which doesn't seem to be a signature at all
 Bob"""
 
-    text, extracted_signature = signature.extract(body, sender)
+    text, extracted_signature = extract(body, sender)
     eq_('\n'.join(body.splitlines()[:2]), text)
     eq_('\n'.join(body.splitlines()[-3:]), extracted_signature)
 
@@ -58,7 +65,7 @@ Thanks in advance,
 some long text here which doesn't seem to be a signature at all
 Bob"""
 
-    text, extracted_signature = signature.extract(body, sender)
+    text, extracted_signature = extract(body, sender)
     eq_('\n'.join(body.splitlines()[:-1]), text)
     eq_('Bob', extracted_signature)
 
@@ -66,13 +73,38 @@ Bob"""
 
     some *long* text here which doesn't seem to be a signature at all
     """
-    ((body, None), signature.extract(body, "david@example.com"))
+    ((body, None), extract(body, "david@example.com"))
 
 
 def test_basic():
     msg_body = 'Blah\r\n--\r\n\r\nSergey Obukhov'
     eq_(('Blah', '--\r\n\r\nSergey Obukhov'),
-        signature.extract(msg_body, 'Sergey'))
+        extract(msg_body, 'Sergey'))
+
+
+def test_capitalized():
+    msg_body = """Hi Mary,
+
+Do you still need a DJ for your wedding? I've included a video demo of one of our DJs available for your wedding date.
+
+DJ Doe 
+http://example.com
+Password: SUPERPASSWORD
+
+Would you like to check out more?
+
+
+At your service,
+
+John Smith
+Doe Inc
+555-531-7967"""
+
+    sig = """John Smith
+Doe Inc
+555-531-7967"""
+
+    eq_(sig, extract(msg_body, 'Doe')[1])
 
 
 def test_over_2_text_lines_after_signature():
@@ -83,25 +115,25 @@ def test_over_2_text_lines_after_signature():
     2 non signature lines in the end
     It's not signature
     """
-    text, extracted_signature = signature.extract(body, "Bob")
+    text, extracted_signature = extract(body, "Bob")
     eq_(extracted_signature, None)
 
 
 def test_no_signature():
     sender, body = "bob@foo.bar", "Hello"
-    eq_((body, None), signature.extract(body, sender))
+    eq_((body, None), extract(body, sender))
 
 
 def test_handles_unicode():
     sender, body = dataset.parse_msg_sender(UNICODE_MSG)
-    text, extracted_signature = signature.extract(body, sender)
+    text, extracted_signature = extract(body, sender)
 
 
-@patch.object(signature.extraction, 'has_signature')
+@patch.object(extraction, 'has_signature')
 def test_signature_extract_crash(has_signature):
     has_signature.side_effect = Exception('Bam!')
     msg_body = u'Blah\r\n--\r\n\r\nСергей'
-    eq_((msg_body, None), signature.extract(msg_body, 'Сергей'))
+    eq_((msg_body, None), extract(msg_body, 'Сергей'))
 
 
 def test_mark_lines():
@@ -127,20 +159,20 @@ def test_mark_lines():
 
 def test_process_marked_lines():
     # no signature found
-    eq_((range(5), None), e._process_marked_lines(range(5), 'telt'))
+    eq_((list(range(5)), None), e._process_marked_lines(list(range(5)), 'telt'))
 
     # signature in the middle of the text
-    eq_((range(9), None), e._process_marked_lines(range(9), 'tesestelt'))
+    eq_((list(range(9)), None), e._process_marked_lines(list(range(9)), 'tesestelt'))
 
     # long line splits signature
-    eq_((range(7), [7, 8]),
+    eq_((list(range(7)), [7, 8]),
-        e._process_marked_lines(range(9), 'tsslsless'))
+        e._process_marked_lines(list(range(9)), 'tsslsless'))
 
-    eq_((range(20), [20]),
+    eq_((list(range(20)), [20]),
-        e._process_marked_lines(range(21), 'ttttttstttesllelelets'))
+        e._process_marked_lines(list(range(21)), 'ttttttstttesllelelets'))
 
     # some signature lines could be identified as text
-    eq_(([0], range(1, 9)), e._process_marked_lines(range(9), 'tsetetest'))
+    eq_(([0], list(range(1, 9))), e._process_marked_lines(list(range(9)), 'tsetetest'))
 
-    eq_(([], range(5)),
+    eq_(([], list(range(5))),
-        e._process_marked_lines(range(5), "ststt"))
+        e._process_marked_lines(list(range(5)), "ststt"))

tests/signature/learning/dataset_test.py

@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 
+from __future__ import absolute_import
 from ... import *
 import os
 

tests/signature/learning/featurespace_test.py

@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 
+from __future__ import absolute_import
 from ... import *
 
 from talon.signature.learning import featurespace as fs

tests/signature/learning/helpers_test.py

@@ -1,11 +1,13 @@
 # -*- coding: utf-8 -*-
 
+from __future__ import absolute_import
 from ... import *
 
 import regex as re
 
 from talon.signature.learning import helpers as h
 from talon.signature.learning.helpers import *
+from six.moves import range
 
 # First testing regex constants.
 VALID = '''
@@ -154,7 +156,7 @@ def test_extract_names():
         # check that extracted names could be compiled
         try:
             re.compile("|".join(extracted_names))
-        except Exception, e:
+        except Exception as e:
             ok_(False, ("Failed to compile extracted names {}"
                         "\n\nReason: {}").format(extracted_names, e))
         if expected_names:
@@ -190,10 +192,11 @@ def test_punctuation_percent(categories_percent):
 def test_capitalized_words_percent():
     eq_(0.0, h.capitalized_words_percent(''))
     eq_(100.0, h.capitalized_words_percent('Example Corp'))
-    eq_(50.0, h.capitalized_words_percent('Qqq qqq QQQ 123 sss'))
+    eq_(50.0, h.capitalized_words_percent('Qqq qqq Aqs 123 sss'))
     eq_(100.0, h.capitalized_words_percent('Cell 713-444-7368'))
     eq_(100.0, h.capitalized_words_percent('8th Floor'))
     eq_(0.0, h.capitalized_words_percent('(212) 230-9276'))
+    eq_(50.0, h.capitalized_words_percent('Password: REMARKABLE'))
 
 
 def test_has_signature():
@@ -204,7 +207,7 @@ def test_has_signature():
                         'sender@example.com'))
     assert_false(h.has_signature('http://www.example.com/555-555-5555',
                                  'sender@example.com'))
-    long_line = ''.join(['q' for e in xrange(28)])
+    long_line = ''.join(['q' for e in range(28)])
     assert_false(h.has_signature(long_line + ' sender', 'sender@example.com'))
     # wont crash on an empty string
     assert_false(h.has_signature('', ''))

tests/text_quotations_test.py

@@ -1,5 +1,6 @@
 # -*- coding: utf-8 -*-
 
+from __future__ import absolute_import
 from . import *
 from . fixtures import *
 
@@ -7,6 +8,9 @@ import os
 
 import email.iterators
 from talon import quotations
+import six
+from six.moves import range
+from six import StringIO
 
 
 @patch.object(quotations, 'MAX_LINES_COUNT', 1)
@@ -24,6 +28,32 @@ def test_pattern_on_date_somebody_wrote():
 
 On 11-Apr-2011, at 6:54 PM, Roman Tkachenko <romant@example.com> wrote:
 
+>
+> Test
+>
+> Roman"""
+
+    eq_("Test reply", quotations.extract_from_plain(msg_body))
+
+def test_pattern_on_date_polymail():
+    msg_body = """Test reply
+
+On Tue, Apr 11, 2017 at 10:07 PM John Smith
+
+<
+mailto:John Smith <johnsmith@gmail.com>
+> wrote:
+Test quoted data
+"""
+
+    eq_("Test reply", quotations.extract_from_plain(msg_body))
+
+
+def test_pattern_sent_from_samsung_smb_wrote():
+    msg_body = """Test reply
+
+Sent from Samsung MobileName <address@example.com> wrote:
+
 >
 > Test
 >
@@ -89,6 +119,38 @@ On 11-Apr-2011, at 6:54 PM, Roman Tkachenko <romant@example.com> sent:
     eq_("Test reply", quotations.extract_from_plain(msg_body))
 
 
+def test_appointment():
+    msg_body = """Response
+
+10/19/2017 @ 9:30 am for physical therapy
+Bla
+1517 4th Avenue Ste 300
+London CA 19129, 555-421-6780
+
+John Doe, FCLS
+Mailgun Inc
+555-941-0697
+
+From: from@example.com [mailto:from@example.com]
+Sent: Wednesday, October 18, 2017 2:05 PM
+To: John Doer - SIU <jd@example.com>
+Subject: RE: Claim # 5551188-1
+
+Text"""
+
+    expected = """Response
+
+10/19/2017 @ 9:30 am for physical therapy
+Bla
+1517 4th Avenue Ste 300
+London CA 19129, 555-421-6780
+
+John Doe, FCLS
+Mailgun Inc
+555-941-0697"""
+    eq_(expected, quotations.extract_from_plain(msg_body))
+
+
 def test_line_starts_with_on():
     msg_body = """Blah-blah-blah
 On blah-blah-blah"""
@@ -125,7 +187,8 @@ def _check_pattern_original_message(original_message_indicator):
 -----{}-----
 
 Test"""
-    eq_('Test reply', quotations.extract_from_plain(msg_body.format(unicode(original_message_indicator))))
+    eq_('Test reply', quotations.extract_from_plain(
+        msg_body.format(six.text_type(original_message_indicator))))
 
 def test_english_original_message():
     _check_pattern_original_message('Original Message')
@@ -148,6 +211,17 @@ Test reply"""
     eq_("Test reply", quotations.extract_from_plain(msg_body))
 
 
+def test_android_wrote():
+    msg_body = """Test reply
+
+---- John Smith wrote ----
+
+> quoted
+> text
+"""
+    eq_("Test reply", quotations.extract_from_plain(msg_body))
+
+
 def test_reply_wraps_quotations():
     msg_body = """Test reply
 
@@ -359,6 +433,14 @@ Op 17-feb.-2015, om 13:18 heeft Julius Caesar <pantheon@rome.com> het volgende g
 Small batch beard laboris tempor, non listicle hella Tumblr heirloom.
 """))
 
+def test_vietnamese_from_block():
+    eq_('Hello', quotations.extract_from_plain(
+    u"""Hello
+
+Vào 14:24 8 tháng 6, 2017, Hùng Nguyễn <hungnguyen@xxx.com> đã viết:
+
+> Xin chào
+"""))
 
 def test_quotation_marker_false_positive():
     msg_body = """Visit us now for assistance...
@@ -371,6 +453,7 @@ 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> >
@@ -412,7 +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))
 
 
@@ -482,11 +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',
              '',
@@ -649,6 +735,15 @@ def test_preprocess_postprocess_2_links():
     eq_(msg_body, quotations.extract_from_plain(msg_body))
 
 
+def body_iterator(msg, decode=False):
+    for subpart in msg.walk():
+        payload = subpart.get_payload(decode=decode)
+        if isinstance(payload, six.text_type):
+            yield payload
+        else:
+            yield payload.decode('utf8')
+
+
 def test_standard_replies():
     for filename in os.listdir(STANDARD_REPLIES):
         filename = os.path.join(STANDARD_REPLIES, filename)
@@ -656,8 +751,8 @@ def test_standard_replies():
             continue
         with open(filename) as f:
             message = email.message_from_file(f)
-            body = email.iterators.typed_subpart_iterator(message, subtype='plain').next()
+            body = next(email.iterators.typed_subpart_iterator(message, subtype='plain'))
-            text = ''.join(email.iterators.body_line_iterator(body, True))
+            text = ''.join(body_iterator(body, True))
 
             stripped_text = quotations.extract_from_plain(text)
             reply_text_fn = filename[:-4] + '_reply_text'
@@ -670,3 +765,77 @@ def test_standard_replies():
                 "'%(reply)s' != %(stripped)s for %(fn)s" % \
                 {'reply': reply_text, 'stripped': stripped_text,
                  'fn': filename}
+
+
+def test_split_email():
+    msg = """From: Mr. X
+    Date: 24 February 2016
+    To: Mr. Y
+    Subject: Hi
+    Attachments: none
+    Goodbye.
+    From: Mr. Y
+    To: Mr. X
+    Date: 24 February 2016
+    Subject: Hi
+    Attachments: none
+
+    Hello.
+
+        On 24th February 2016 at 09.32am, Conal wrote:
+
+        Hey!
+
+        On Mon, 2016-10-03 at 09:45 -0600, Stangel, Dan wrote:
+        > Mohan,
+        >
+        > We have not yet migrated the systems.
+        >
+        > Dan
+        >
+        > > -----Original Message-----
+        > > Date: Mon, 2 Apr 2012 17:44:22 +0400
+        > > Subject: Test
+        > > From: bob@xxx.mailgun.org
+        > > To: xxx@gmail.com; xxx@hotmail.com; xxx@yahoo.com; xxx@aol.com; xxx@comcast.net; xxx@nyc.rr.com
+        > >
+        > > Hi
+        > >
+        > > > From: bob@xxx.mailgun.org
+        > > > To: xxx@gmail.com; xxx@hotmail.com; xxx@yahoo.com; xxx@aol.com; xxx@comcast.net; xxx@nyc.rr.com
+        > > > Date: Mon, 2 Apr 2012 17:44:22 +0400
+        > > > Subject: Test
+        > > > Hi
+        > > >
+        > >
+        >
+        >
+"""
+    expected_markers = "stttttsttttetesetesmmmmmmsmmmmmmmmmmmmmmmm"
+    markers = quotations.split_emails(msg)
+    eq_(markers, expected_markers)
+
+
+
+def test_feedback_below_left_unparsed():
+    msg_body = """Please enter your feedback below. Thank you.
+
+------------------------------------- Enter Feedback Below -------------------------------------
+
+The user experience was unparallelled. Please continue production. I'm sending payment to ensure
+that this line is intact."""
+
+    parsed = quotations.extract_from_plain(msg_body)
+    eq_(msg_body, parsed.decode('utf8'))
+
+
+def test_appointment():
+    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.decode('utf8'))

tests/utils_test.py

@@ -1,9 +1,12 @@
 # coding:utf-8
 
-from . import *
+from __future__ import absolute_import
+
+import cchardet
+import six
 
 from talon import utils as u
-import cchardet
+from . import *
 
 
 def test_get_delimiter():
@@ -14,49 +17,53 @@ def test_get_delimiter():
 
 def test_unicode():
     eq_(u'hi', u.to_unicode('hi'))
-    eq_ (type(u.to_unicode('hi')), unicode )
+    eq_(type(u.to_unicode('hi')), six.text_type)
-    eq_ (type(u.to_unicode(u'hi')), unicode )
+    eq_(type(u.to_unicode(u'hi')), six.text_type)
-    eq_ (type(u.to_unicode('привет')), unicode )
+    eq_(type(u.to_unicode('привет')), six.text_type)
-    eq_ (type(u.to_unicode(u'привет')), unicode )
+    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('Versi\xf3n', precise=True))
+    eq_(u"Versión", u.to_unicode(u'Versi\xf3n'.encode('iso-8859-2'), precise=True))
 
 
 def test_detect_encoding():
-    eq_ ('ascii', u.detect_encoding('qwe').lower())
+    eq_('ascii', u.detect_encoding(b'qwe').lower())
-    eq_ ('iso-8859-2', u.detect_encoding('Versi\xf3n').lower())
+    ok_(u.detect_encoding(
-    eq_ ('utf-8', u.detect_encoding('привет').lower())
+        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').lower())
+        eq_('utf-8', u.detect_encoding('qwe'.encode('utf8')).lower())
 
 
 def test_quick_detect_encoding():
-    eq_ ('ascii', u.quick_detect_encoding('qwe').lower())
+    eq_('ascii', u.quick_detect_encoding(b'qwe').lower())
-    eq_ ('windows-1252', u.quick_detect_encoding('Versi\xf3n').lower())
+    ok_(u.quick_detect_encoding(
-    eq_ ('utf-8', u.quick_detect_encoding('привет').lower())
+        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("qwe"))
+    eq_('ascii', u.quick_detect_encoding(b"qwe"))
-    cchardet_detect.assert_called_once_with("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("qwe"))
+    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("qwe"))
+    eq_('utf-8', u.quick_detect_encoding(b"qwe"))
     ok_(detect_encoding.called)
 
 
@@ -73,11 +80,11 @@ Haha
 </p>
 </body>"""
     text = u.html_to_text(html)
-    eq_("Hello world! \n\n  * One! \n  * Two \nHaha", text)
+    eq_(b"Hello world! \n\n  * One! \n  * Two \nHaha", text)
-    eq_("привет!", u.html_to_text("<b>привет!</b>"))
+    eq_(u"привет!", u.html_to_text("<b>привет!</b>").decode('utf8'))
 
     html = '<body><br/><br/>Hi</body>'
-    eq_ ('Hi', u.html_to_text(html))
+    eq_(b'Hi', u.html_to_text(html))
 
     html = """Hi
 <style type="text/css">
@@ -97,11 +104,60 @@ font: 13px 'Lucida Grande', Arial, sans-serif;
 
 }
 </style>"""
-    eq_ ('Hi', u.html_to_text(html))
+    eq_(b'Hi', u.html_to_text(html))
 
     html = """<div>
 <!-- COMMENT 1 -->
 <span>TEXT 1</span>
 <p>TEXT 2 <!-- COMMENT 2 --></p>
 </div>"""
-    eq_('TEXT 1 \nTEXT 2', u.html_to_text(html))
+    eq_(b'TEXT 1 \nTEXT 2', u.html_to_text(html))
+
+
+def test_comment_no_parent():
+    s = b'<!-- 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)
+
+
+@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>"))

train.py

@@ -1,3 +1,4 @@
+from __future__ import absolute_import
 from talon.signature import EXTRACTOR_FILENAME, EXTRACTOR_DATA
 from talon.signature.learning.classifier import train, init