I’m making early steps with the Lark library in Python and am really looking forward to replacing a lot of awful if statements with an EBNF parser..!
The task here is interpreting the times written in railway timetables. 12:34 means the train will stop, 12/34 means the train is not expected to stop, 12d34 means the train may stop to set down passengers only… and there’s at least four more variations like this. Times are to the half-minute: 12:34h means 12:34:30 and it’s that h I’m having trouble with.
I have a dataclass and enum to store the parsed values:
from dataclasses import dataclass
from enum import Enum, auto
@dataclass(frozen=True)
class TTime:
class StopMode(Enum):
STOPPING = auto()
PASSING = auto()
SET_DOWN = auto() # etc...
hour: int
minute: int
second: int = 0
stopmode: StopMode = StopMode.PASSING
from lark import Lark, Transformer
class TreeToTTime(Transformer):
def hour(self, n):
return int("".join(n))
def minute(self, n):
return int("".join(n))
def stopmode(self, n):
(n,) = n
return n
def stopping(self, _):
return TTime.StopMode.STOPPING
def passing(self, _):
return TTime.StopMode.PASSING
def setdown(self, _):
return TTime.StopMode.SET_DOWN
def halfminute(self, _):
return True
def ttime(self, args):
hour, stopmode, minute, halfminute = args
second = 30 if halfminute else 0
return TTime(hour, minute, second, stopmode)
ttime_parser = Lark(
r"""
ttime : hour stopmode minute halfminute?
hour : ("0".."2")? DIGIT
minute : ("0".."5") DIGIT
stopmode : stopping | passing | setdown
stopping : ":"
passing : "/"
setdown : "d"
halfminute : "h" | "H"
%import common.DIGIT
%import common.WS
%ignore WS
""",
start="ttime",
)
def parse_ttime(text):
tree = ttime_parser.parse(text)
return TreeToTTime().transform(tree)
import pytest
@pytest.mark.parametrize(
"text,expected",
[
(" 0:00", TTime(0, 0, 0, TTime.StopMode.STOPPING)),
("13:20", TTime(13, 20, 0, TTime.StopMode.STOPPING)),
("00/00", TTime(0, 0, 0, TTime.StopMode.PASSING)),
("12d34", TTime(12, 34, 0, TTime.StopMode.SET_DOWN)),
# ("01:23h", TTime(1, 23, 30, TTime.StopMode.STOPPING)),
],
)
def test_parse_times(text, expected):
assert expected == parse_time(text)
The problem is the args passed to ttime. If halfminute is absent i.e. 12:34, args has three values: (12, StopMode.STOPPING, 34). If halfminute is present i.e. 12:34h it’s four values: (12, StopMode.STOPPING, 34, True).
How do I configure the transformer here to include the absence of a token? For 12:34 I’d like args to be [12, StopMode.STOPPING, 34, False] for example.
I could write:
hour, stopmode, minute, halfminute, *_ = args + (False,)
but I feel this is not the right approach – it only works because the halfminute token is the final token for example.
Any other pointers gratefully received – I feel I’ve made rapid progress this evening but there’s still a lot I’ll be missing!
The answer is to mark halfminute as optional with square brackets not a question mark:
ttime : hour stopmode minute [halfminute]
With the maybe_placeholder parser argument set to True (which is the default), [x] will parse as a token for x if matched or None if not.
A more minimal example:
from lark import Lark
grammar = """
start : [A] B?
A : "a"
B : "b"
"""
parser = Lark(grammar)
for x in ("a", "b", "ab", ""):
print(parser.parse(x))
This outputs:
Tree(Token('RULE', 'start'), [Token('A', 'a')])
Tree(Token('RULE', 'start'), [None, Token('B', 'b')])
Tree(Token('RULE', 'start'), [Token('A', 'a'), Token('B', 'b')])
Tree(Token('RULE', 'start'), [None])
i.e. [A] is either present with a value or is None. B? is either present with a value or absent.
There’s a bunch of other problems with my code here which I’m still working out – however it does technically work 🙂
I’ve iterated on the grammar and the transformer to something I’m much happier with. (Previous iterations in the post history.)
time : HOUR stopmode MINUTE [HALFMINUTE]
HOUR : ("0".."2")? DIGIT
MINUTE : ("0".."5")? DIGIT
HALFMINUTE : "h" | "H" | "½"
stopmode: STOPPING | PASSING | SET_DOWN | IF_REQUIRED
| REQUEST_STOP | DWELL_TIME | THROUGH_LINE
STOPPING : ":"
PASSING : "/"
SET_DOWN : "d"
IF_REQUIRED : "*"
REQUEST_STOP : "r"
DWELL_TIME : "w"
THROUGH_LINE : "t"
%import common.DIGIT
%import common.WS
%ignore WS
class TreeToTTime(Transformer):
def time(self, args):
hour, stopmode, minute, second = args
return TTime(hour, minute, second or 0, stopmode)
def HOUR(self, n):
return int("".join(n))
def MINUTE(self, n):
return int("".join(n))
def HALFMINUTE(self, _):
return 30
def stopmode(self, args):
(mode,) = args
return getattr(TTime.StopMode, mode.type, TTime.StopMode.PASSING)
There is now only a coupling between the terminal names in the grammar and the enum value names in the domain object, which I think is quite acceptable!