Merge pull request #148 from mailgun/sergey/v1.4.2

bump version after adding support for Vietnamese format

.gitignore

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

setup.py

@@ -29,7 +29,7 @@ class InstallCommand(install):
 
 
 setup(name='talon',
-      version='1.3.4',
+      version='1.4.2',
       description=("Mailgun library "
                    "to extract message quotations and signatures."),
       long_description=open("README.rst").read(),
@@ -48,7 +48,7 @@ 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',

talon/html_quotations.py

@@ -94,6 +94,12 @@ def cut_microsoft_quote(html_message):
         #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']|"
         #windows mail
         "//div[@style='padding-top: 5px; "
         "border-top-color: rgb(229, 229, 229); "

talon/quotations.py

@@ -42,6 +42,8 @@ RE_ON_DATE_SMB_WROTE = re.compile(
             u'På',
             # Swedish, Danish
             'Den',
+            # Vietnamese
+            u'Vào',
         )),
         # Date and sender separator
         u'|'.join((
@@ -64,6 +66,8 @@ RE_ON_DATE_SMB_WROTE = re.compile(
             'schrieb',
             # Norwegian, Swedish
             'skrev',
+            # Vietnamese
+            u'đã viết',
         ))
     ))
 # Special case for languages where text is translated like this: 'on {date} wrote {somebody}:'
@@ -139,6 +143,21 @@ RE_FROM_COLON_OR_DATE_COLON = re.compile(u'(_+\r?\n)?[\s]*(:?[*]?{})[\s]?:[*]?.*
         'Date', 'Datum', u'Envoyé', 'Skickat', 'Sendt',
     ))), re.I)
 
+# ---- 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,
@@ -154,16 +173,17 @@ SPLITTER_PATTERNS = [
     re.compile('\S{3,10}, \d\d? \S{3,10} 20\d\d,? \d\d?:\d\d(:\d\d)?'
                '( \S+){3,6}@\S+:'),
     # Sent from Samsung MobileName <address@example.com> wrote:
-    re.compile('Sent from Samsung .*@.*> wrote')
+    re.compile('Sent from Samsung .*@.*> 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

talon/signature/bruteforce.py

@@ -1,15 +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'''
                (
@@ -28,7 +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'''
                (
@@ -45,7 +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)
@@ -112,7 +110,7 @@ def extract_signature(msg_body):
 
             return (stripped_body.strip(),
                     signature.strip())
-    except Exception as e:
+    except Exception:
         log.exception('ERROR extracting signature')
         return (msg_body, None)
 
@@ -163,7 +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))):
@@ -174,7 +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

@@ -1,16 +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__)
 
@@ -58,7 +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)
@@ -81,7 +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
@@ -96,7 +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):
@@ -111,3 +110,4 @@ def _process_marked_lines(lines, markers):
         return (lines[:-signature.end()], lines[-signature.end():])
 
     return (lines, None)
+

talon/signature/learning/classifier.py

@@ -6,9 +6,10 @@ 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():
@@ -29,4 +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

@@ -17,13 +17,14 @@ 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
-from six.moves import range
-
 
 SENDER_SUFFIX = '_sender'
 BODY_SUFFIX = '_body'
@@ -57,9 +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:

talon/utils.py

@@ -1,19 +1,18 @@
 # coding:utf-8
 
 from __future__ import absolute_import
-import logging
+
 from random import shuffle
-import chardet
+
 import cchardet
-import regex as re
+import chardet
-
-from lxml.html import html5parser
-from lxml.cssselect import CSSSelector
-
 import html5lib
+import regex as re
+import six
+from lxml.cssselect import CSSSelector
+from lxml.html import html5parser
 
 from talon.constants import RE_DELIMITER
-import six
 
 
 def safe_format(format_string, *args, **kwargs):
@@ -177,6 +176,8 @@ def html_to_text(string):
 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
@@ -189,6 +190,8 @@ def html_fromstring(s):
 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
@@ -203,7 +206,9 @@ def cssselect(expr, tree):
 
 
 def html_too_big(s):
-    return s.count('<') > _MAX_TAGS_COUNT
+    if isinstance(s, six.text_type):
+        s = s.encode('utf8')
+    return s.count(b'<') > _MAX_TAGS_COUNT
 
 
 def _contains_charset_spec(s):
@@ -248,7 +253,6 @@ def _html5lib_parser():
 _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']
 

tests/html_quotations_test.py

@@ -1,13 +1,13 @@
 # -*- coding: utf-8 -*-
 
 from __future__ import absolute_import
-from . 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")
@@ -303,7 +303,12 @@ Reply
 
 
 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)

tests/signature/extraction_test.py

@@ -1,16 +1,16 @@
 # -*- coding: utf-8 -*-
 
 from __future__ import absolute_import
-from .. import *
 
 import os
 
-from talon.signature.learning import dataset
-from talon import signature
-from talon.signature import extraction as e
-from talon.signature import bruteforce
 from six.moves import range
 
+from talon.signature import bruteforce, extraction, extract
+from talon.signature import extraction as e
+from talon.signature.learning import dataset
+from .. import *
+
 
 def test_message_shorter_SIGNATURE_MAX_LINES():
     sender = "bob@foo.bar"
@@ -18,20 +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)]
@@ -47,7 +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)
 
@@ -60,7 +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)
 
@@ -68,13 +73,13 @@ 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():
@@ -99,7 +104,7 @@ Doe Inc
 Doe Inc
 555-531-7967"""
 
-    eq_(sig, signature.extract(msg_body, 'Doe')[1])
+    eq_(sig, extract(msg_body, 'Doe')[1])
 
 
 def test_over_2_text_lines_after_signature():
@@ -110,25 +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():

tests/text_quotations_test.py

@@ -35,6 +35,19 @@ On 11-Apr-2011, at 6:54 PM, Roman Tkachenko <romant@example.com> wrote:
 
     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
@@ -142,7 +155,8 @@ def _check_pattern_original_message(original_message_indicator):
 -----{}-----
 
 Test"""
-    eq_('Test reply', quotations.extract_from_plain(msg_body.format(six.text_type(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')
@@ -165,6 +179,17 @@ 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
 
@@ -376,6 +401,14 @@ 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...

tests/utils_test.py

@@ -1,12 +1,12 @@
 # coding:utf-8
 
 from __future__ import absolute_import
-from . import *
 
-from talon import utils as u
 import cchardet
 import six
-from lxml import html
+
+from talon import utils as u
+from . import *
 
 
 def test_get_delimiter():
@@ -29,7 +29,9 @@ def test_unicode():
 
 def test_detect_encoding():
     eq_('ascii', u.detect_encoding(b'qwe').lower())
-    eq_ ('iso-8859-2', u.detect_encoding(u'Versi\xf3n'.encode('iso-8859-2')).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:
@@ -39,7 +41,9 @@ def test_detect_encoding():
 
 def test_quick_detect_encoding():
     eq_('ascii', u.quick_detect_encoding(b'qwe').lower())
-    eq_ ('windows-1252', u.quick_detect_encoding(u'Versi\xf3n'.encode('windows-1252')).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())
 
 
@@ -111,15 +115,16 @@ font: 13px 'Lucida Grande', Arial, sans-serif;
 
 
 def test_comment_no_parent():
-    s = "<!-- COMMENT 1 --> no comment"
+    s = b'<!-- COMMENT 1 --> no comment'
     d = u.html_document_fromstring(s)
-    eq_("no comment", u.html_tree_to_text(d))
+    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):
@@ -154,5 +159,5 @@ def test_html_too_big():
 
 @patch.object(u, '_MAX_TAGS_COUNT', 3)
 def test_html_to_text():
-    eq_("Hello", u.html_to_text("<div>Hello</div>"))
+    eq_(b"Hello", u.html_to_text("<div>Hello</div>"))
     eq_(None, u.html_to_text("<div><span>Hi</span></div>"))