ISRI Arabic Stemmer broken on Python >= 3.6

[b2m]

The ISRI ARabic Stemmer (src.whoosh.lang.isry.py) does not work on Python >= 3.6.

Exception: re.error: bad escape \u at position 0.
Reason: changed behavior of re.sub.

Changed in version 3.6: Unknown escapes in pattern consisting of '\' and an ASCII letter now are errors.

(quoted from https://docs.python.org/3/library/re.html)

Code snippet to reproduce:

from whoosh.analysis import LanguageAnalyzer

analyzer = LanguageAnalyzer(lang='ar')
[(token.text, token.stopped) for token in analyzer("This is a test")]

Codesamples with bad escape sequences:

• whoosh/src/whoosh/lang/isri.py

  Line 87 in 26153e2

  self.re_short_vowels = re.compile(r'[\u064B-\u0652]')

• whoosh/src/whoosh/lang/isri.py

  Line 88 in 26153e2

  self.re_hamza = re.compile(r'[\u0621\u0624\u0626]')

• whoosh/src/whoosh/lang/isri.py

  Line 89 in 26153e2

  self.re_intial_hamza = re.compile(r'^[\u0622\u0623\u0625]')

• whoosh/src/whoosh/lang/isri.py

  Line 190 in 26153e2

  self.stm = self.re_intial_hamza.sub(r'\u0627', self.stm)

• whoosh/src/whoosh/lang/isri.py

  Line 194 in 26153e2

  self.stm = self.re_intial_hamza.sub(r'\u0627', self.stm)

[stevennic]

Thanks Benjamin, good catch. I have submitted a PR to fix this.

[nijel]

Should be fixed by #557