pyparsing v2.3.0 Release Notes

Release Date: 2018-10-01 // about 3 years ago
    • ๐Ÿ†• NEW SUPPORT FOR UNICODE CHARACTER RANGES This release introduces the pyparsing_unicode namespace class, defining a series of language character sets to simplify the definition of alphas, nums, alphanums, and printables in the following language sets: . Arabic . Chinese . Cyrillic . Devanagari . Greek . Hebrew . Japanese (including Kanji, Katakana, and Hirigana subsets) . Korean . Latin1 (includes 7 and 8-bit Latin characters) . Thai . CJK (combination of Chinese, Japanese, and Korean sets)

    For example, your code can define words using:

    korean_word = Word(pyparsing_unicode.Korean.alphas)
    

    See their use in the updated examples greetingInGreek.py and greetingInKorean.py.

    This namespace class also offers access to these sets using their unicode identifiers.

    • ๐Ÿ“œ POSSIBLE API CHANGE: Fixed bug where a parse action that explicitly returned the input ParseResults could add another nesting level in the results if the current expression had a results name.

      vals = pp.OneOrMore(pp.pyparsing_common.integer)("int_values")
      
      def add_total(tokens):
          tokens['total'] = sum(tokens)
          return tokens  # this line can be removed
      
      vals.addParseAction(add_total)
      print(vals.parseString("244 23 13 2343").dump())
      

    Before the fix, this code would print (note the extra nesting level):

    [244, 23, 13, 2343]
    - int_values: [244, 23, 13, 2343]
      - int_values: [244, 23, 13, 2343]
      - total: 2623
    - total: 2623
    

    With the fix, this code now prints:

    [244, 23, 13, 2343]
    - int_values: [244, 23, 13, 2343]
    - total: 2623
    

    This fix will change the structure of ParseResults returned if a program defines a parse action that returns the tokens that were sent in. This is not necessary, and statements like "return tokens" in the example above can be safely deleted prior to upgrading to this release, in order to avoid the bug and get the new behavior.

    Reported by seron in Issue #22, nice catch!

    • ๐Ÿ›  POSSIBLE API CHANGE: Fixed a related bug where a results name erroneously created a second level of hierarchy in the returned ParseResults. The intent for accumulating results names into ParseResults is that, in the absence of Group'ing, all names get merged into a common namespace. This allows us to write:

      key_value_expr = (Word(alphas)("key") + '=' + Word(nums)("value")) result = key_value_expr.parseString("a = 100")

    and have result structured as {"key": "a", "value": "100"} instead of [{"key": "a"}, {"value": "100"}].

    However, if a named expression is used in a higher-level non-Group expression that also has a name, a false sub-level would be created in the namespace:

        num = pp.Word(pp.nums)
        num_pair = ("[" + (num("A") + num("B"))("values") + "]")
        U = num_pair.parseString("[ 10 20 ]")
        print(U.dump())
    

    Since there is no grouping, "A", "B", and "values" should all appear at the same level in the results, as:

        ['[', '10', '20', ']']
        - A: '10'
        - B: '20'
        - values: ['10', '20']
    

    Instead, an extra level of "A" and "B" show up under "values":

        ['[', '10', '20', ']']
        - A: '10'
        - B: '20'
        - values: ['10', '20']
          - A: '10'
          - B: '20'
    

    This bug has been fixed. Now, if this hierarchy is desired, then a Group should be added:

        num_pair = ("[" + pp.Group(num("A") + num("B"))("values") + "]")
    

    Giving:

        ['[', ['10', '20'], ']']
        - values: ['10', '20']
          - A: '10'
          - B: '20'
    

    But in no case should "A" and "B" appear in multiple levels. This bug-fix fixes that.

    If you have current code which relies on this behavior, then add or remove Groups as necessary to get your intended results structure.

    Reported by Athanasios Anastasiou.

    • ๐Ÿ“œ IndexError's raised in parse actions will get explicitly reraised as ParseExceptions that wrap the original IndexError. Since IndexError sometimes occurs as part of pyparsing's normal parsing logic, IndexErrors that are raised during a parse action may have gotten silently reinterpreted as parsing errors. To retain the information from the IndexError, these exceptions will now be raised as ParseExceptions that reference the original IndexError. This wrapping will only be visible when run under Python3, since it emulates "raise ... from ..." syntax.

    Addresses Issue #4, reported by guswns0528.

    • โž• Added Char class to simplify defining expressions of a single character. (Char("abc") is equivalent to Word("abc", exact=1))

    • โž• Added class PrecededBy to perform lookbehind tests. PrecededBy is used in the same way as FollowedBy, passing in an expression that must occur just prior to the current parse location.

    For fixed-length expressions like a Literal, Keyword, Char, or a Word with an exact or maxLen length given, PrecededBy(expr) is sufficient. For varying length expressions like a Word with no given maximum length, PrecededBy must be constructed with an integer retreat argument, as in PrecededBy(Word(alphas, nums), retreat=10), to specify the maximum number of characters pyparsing must look backward to make a match. pyparsing will check all the values from 1 up to retreat characters back from the current parse location.

    When stepping backwards through the input string, PrecededBy does not skip over whitespace.

    PrecededBy can be created with a results name so that, even though it always returns an empty parse result, the result can include named results.

    Idea first suggested in Issue #30 by Freakwill.

    • โšก๏ธ Updated FollowedBy to accept expressions that contain named results, so that results names defined in the lookahead expression will be returned, even though FollowedBy always returns an empty list. Inspired by the same feature implemented in PrecededBy.