|
| 1 | +# Copyright 2020 Google LLC |
| 2 | +# |
| 3 | +# Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +# you may not use this file except in compliance with the License. |
| 5 | +# You may obtain a copy of the License at |
| 6 | +# |
| 7 | +# http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +# |
| 9 | +# Unless required by applicable law or agreed to in writing, software |
| 10 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | +# See the License for the specific language governing permissions and |
| 13 | +# limitations under the License. |
| 14 | + |
| 15 | +from collections import namedtuple |
| 16 | +from collections import OrderedDict |
| 17 | +import itertools |
| 18 | +import re |
| 19 | + |
| 20 | +import enum |
| 21 | + |
| 22 | + |
| 23 | +Token = namedtuple("Token", ("type_", "lexeme", "pos")) |
| 24 | +StateTransition = namedtuple("StateTransition", ("new_state", "total_offset")) |
| 25 | + |
| 26 | +# Pattern matching is done with regexes, and the order in which the token patterns are |
| 27 | +# defined is important. |
| 28 | +# |
| 29 | +# Suppose we had the following token definitions: |
| 30 | +# * INT - a token matching integers, |
| 31 | +# * FLOAT - a token matching floating point numbers, |
| 32 | +# * DOT - a token matching a single literal dot character, i.e. "." |
| 33 | +# |
| 34 | +# The FLOAT token would have to be defined first, since we would want the input "1.23" |
| 35 | +# to be tokenized as a single FLOAT token, and *not* three tokens (INT, DOT, INT). |
| 36 | +# |
| 37 | +# Sometimes, however, different tokens match too similar patterns, and it is not |
| 38 | +# possible to define them in order that would avoid any ambiguity. One such case are |
| 39 | +# the OPT_VAL and PY_NUMBER tokens, as both can match an integer literal, say "42". |
| 40 | +# |
| 41 | +# In order to avoid the dilemmas, the lexer implements a concept of STATES. States are |
| 42 | +# used to split token definitions into subgroups, and in each lexer state only a single |
| 43 | +# subgroup is used for tokenizing the input. Lexer states can therefore be though of as |
| 44 | +# token namespaces. |
| 45 | +# |
| 46 | +# For example, while parsing the value of the "--params" option, we do not want to |
| 47 | +# "recognize" it as a single OPT_VAL token, but instead want to parse it as a Python |
| 48 | +# dictionary and verify its syntactial correctness. On the other hand, while parsing |
| 49 | +# the value of an option other than "--params", we do not really care about its |
| 50 | +# structure, and thus do not want to use any of the "Python tokens" for pattern matching. |
| 51 | +# |
| 52 | +# Since token definition order is important, an OrderedDict is needed with tightly |
| 53 | +# controlled member definitions (i.e. passed as a sequence, and *not* via kwargs). |
| 54 | +token_types = OrderedDict( |
| 55 | + [ |
| 56 | + ( |
| 57 | + "state_parse_pos_args", |
| 58 | + OrderedDict( |
| 59 | + [ |
| 60 | + ( |
| 61 | + "GOTO_PARSE_NON_PARAMS_OPTIONS", |
| 62 | + r"(?P<GOTO_PARSE_NON_PARAMS_OPTIONS>(?=--))", # double dash - starting the options list |
| 63 | + ), |
| 64 | + ( |
| 65 | + "DEST_VAR", |
| 66 | + r"(?P<DEST_VAR>[^\d\W]\w*)", # essentially a Python ID |
| 67 | + ), |
| 68 | + ] |
| 69 | + ), |
| 70 | + ), |
| 71 | + ( |
| 72 | + "state_parse_non_params_options", |
| 73 | + OrderedDict( |
| 74 | + [ |
| 75 | + ( |
| 76 | + "GOTO_PARSE_PARAMS_OPTION", |
| 77 | + r"(?P<GOTO_PARSE_PARAMS_OPTION>(?=--params(?:\s|=|--|$)))", # the --params option |
| 78 | + ), |
| 79 | + ("OPTION_SPEC", r"(?P<OPTION_SPEC>--\w+)"), |
| 80 | + ("OPTION_EQ", r"(?P<OPTION_EQ>=)"), |
| 81 | + ("OPT_VAL", r"(?P<OPT_VAL>\S+?(?=\s|--|$))"), |
| 82 | + ] |
| 83 | + ), |
| 84 | + ), |
| 85 | + ( |
| 86 | + "state_parse_params_option", |
| 87 | + OrderedDict( |
| 88 | + [ |
| 89 | + ( |
| 90 | + "PY_STRING", |
| 91 | + r"(?P<PY_STRING>(?:{})|(?:{}))".format( |
| 92 | + r"'(?:[^'\\]|\.)*'", |
| 93 | + r'"(?:[^"\\]|\.)*"', # single and double quoted strings |
| 94 | + ), |
| 95 | + ), |
| 96 | + ("PARAMS_OPT_SPEC", r"(?P<PARAMS_OPT_SPEC>--params(?=\s|=|--|$))"), |
| 97 | + ("PARAMS_OPT_EQ", r"(?P<PARAMS_OPT_EQ>=)"), |
| 98 | + ( |
| 99 | + "GOTO_PARSE_NON_PARAMS_OPTIONS", |
| 100 | + r"(?P<GOTO_PARSE_NON_PARAMS_OPTIONS>(?=--\w+))", # found another option spec |
| 101 | + ), |
| 102 | + ("PY_BOOL", r"(?P<PY_BOOL>True|False)"), |
| 103 | + ("DOLLAR_PY_ID", r"(?P<DOLLAR_PY_ID>\$[^\d\W]\w*)"), |
| 104 | + ( |
| 105 | + "PY_NUMBER", |
| 106 | + r"(?P<PY_NUMBER>-?[1-9]\d*(?:\.\d+)?(:?[e|E][+-]?\d+)?)", |
| 107 | + ), |
| 108 | + ("SQUOTE", r"(?P<SQUOTE>')"), |
| 109 | + ("DQUOTE", r'(?P<DQUOTE>")'), |
| 110 | + ("COLON", r"(?P<COLON>:)"), |
| 111 | + ("COMMA", r"(?P<COMMA>,)"), |
| 112 | + ("LCURL", r"(?P<LCURL>\{)"), |
| 113 | + ("RCURL", r"(?P<RCURL>})"), |
| 114 | + ("LSQUARE", r"(?P<LSQUARE>\[)"), |
| 115 | + ("RSQUARE", r"(?P<RSQUARE>])"), |
| 116 | + ("LPAREN", r"(?P<LPAREN>\()"), |
| 117 | + ("RPAREN", r"(?P<RPAREN>\))"), |
| 118 | + ] |
| 119 | + ), |
| 120 | + ), |
| 121 | + ( |
| 122 | + "common", |
| 123 | + OrderedDict( |
| 124 | + [ |
| 125 | + ("WS", r"(?P<WS>\s+)"), |
| 126 | + ("EOL", r"(?P<EOL>$)"), |
| 127 | + ( |
| 128 | + # anything not a whitespace or matched by something else |
| 129 | + "UNKNOWN", |
| 130 | + r"(?P<UNKNOWN>\S+)", |
| 131 | + ), |
| 132 | + ] |
| 133 | + ), |
| 134 | + ), |
| 135 | + ] |
| 136 | +) |
| 137 | + |
| 138 | + |
| 139 | +# The _generate_next_value_() enum hook is only available in Python 3.6+, thus we |
| 140 | +# need to do some acrobatics to implement an "auto str enum" base class. Implementation |
| 141 | +# based on the recipe provided by the very author of the Enum library: |
| 142 | +# https://stackoverflow.com/a/32313954/5040035 |
| 143 | +class StrEnumMeta(enum.EnumMeta): |
| 144 | + @classmethod |
| 145 | + def __prepare__(metacls, name, bases, **kwargs): |
| 146 | + # Having deterministic enum members definition order is nice. |
| 147 | + return OrderedDict() |
| 148 | + |
| 149 | + def __new__(metacls, name, bases, oldclassdict): |
| 150 | + # Scan through the declared enum members and convert any value that is a plain |
| 151 | + # empty tuple into a `str` of the name instead. |
| 152 | + newclassdict = enum._EnumDict() |
| 153 | + for key, val in oldclassdict.items(): |
| 154 | + if val == (): |
| 155 | + val = key |
| 156 | + newclassdict[key] = val |
| 157 | + return super(StrEnumMeta, metacls).__new__(metacls, name, bases, newclassdict) |
| 158 | + |
| 159 | + |
| 160 | +# The @six.add_metaclass decorator does not work, Enum complains about _sunder_ names, |
| 161 | +# and we cannot use class syntax directly, because the Python 3 version would cause |
| 162 | +# a syntax error under Python 2. |
| 163 | +AutoStrEnum = StrEnumMeta( |
| 164 | + "AutoStrEnum", |
| 165 | + (str, enum.Enum), |
| 166 | + {"__doc__": "Base enum class for for name=value str enums."}, |
| 167 | +) |
| 168 | + |
| 169 | +TokenType = AutoStrEnum( |
| 170 | + "TokenType", |
| 171 | + [ |
| 172 | + (name, name) |
| 173 | + for name in itertools.chain.from_iterable(token_types.values()) |
| 174 | + if not name.startswith("GOTO_") |
| 175 | + ], |
| 176 | +) |
| 177 | + |
| 178 | + |
| 179 | +class LexerState(AutoStrEnum): |
| 180 | + PARSE_POS_ARGS = () # parsing positional arguments |
| 181 | + PARSE_NON_PARAMS_OPTIONS = () # parsing options other than "--params" |
| 182 | + PARSE_PARAMS_OPTION = () # parsing the "--params" option |
| 183 | + STATE_END = () |
| 184 | + |
| 185 | + |
| 186 | +class Lexer(object): |
| 187 | + """Lexical analyzer for tokenizing the cell magic input line.""" |
| 188 | + |
| 189 | + _GRAND_PATTERNS = { |
| 190 | + LexerState.PARSE_POS_ARGS: re.compile( |
| 191 | + "|".join( |
| 192 | + itertools.chain( |
| 193 | + token_types["state_parse_pos_args"].values(), |
| 194 | + token_types["common"].values(), |
| 195 | + ) |
| 196 | + ) |
| 197 | + ), |
| 198 | + LexerState.PARSE_NON_PARAMS_OPTIONS: re.compile( |
| 199 | + "|".join( |
| 200 | + itertools.chain( |
| 201 | + token_types["state_parse_non_params_options"].values(), |
| 202 | + token_types["common"].values(), |
| 203 | + ) |
| 204 | + ) |
| 205 | + ), |
| 206 | + LexerState.PARSE_PARAMS_OPTION: re.compile( |
| 207 | + "|".join( |
| 208 | + itertools.chain( |
| 209 | + token_types["state_parse_params_option"].values(), |
| 210 | + token_types["common"].values(), |
| 211 | + ) |
| 212 | + ) |
| 213 | + ), |
| 214 | + } |
| 215 | + |
| 216 | + def __init__(self, input_text): |
| 217 | + self._text = input_text |
| 218 | + |
| 219 | + def __iter__(self): |
| 220 | + # Since re.scanner does not seem to support manipulating inner scanner states, |
| 221 | + # we need to implement lexer state transitions manually using special |
| 222 | + # non-capturing lookahead token patterns to signal when a state transition |
| 223 | + # should be made. |
| 224 | + # Since we don't have "nested" states, we don't really need a stack and |
| 225 | + # this simple mechanism is sufficient. |
| 226 | + state = LexerState.PARSE_POS_ARGS |
| 227 | + offset = 0 # the number of characters processed so far |
| 228 | + |
| 229 | + while state != LexerState.STATE_END: |
| 230 | + token_stream = self._find_state_tokens(state, offset) |
| 231 | + |
| 232 | + for maybe_token in token_stream: # pragma: NO COVER |
| 233 | + if isinstance(maybe_token, StateTransition): |
| 234 | + state = maybe_token.new_state |
| 235 | + offset = maybe_token.total_offset |
| 236 | + break |
| 237 | + |
| 238 | + if maybe_token.type_ != TokenType.WS: |
| 239 | + yield maybe_token |
| 240 | + |
| 241 | + if maybe_token.type_ == TokenType.EOL: |
| 242 | + state = LexerState.STATE_END |
| 243 | + break |
| 244 | + |
| 245 | + def _find_state_tokens(self, state, current_offset): |
| 246 | + """Scan the input for current state's tokens starting at ``current_offset``. |
| 247 | +
|
| 248 | + Args: |
| 249 | + state (LexerState): The current lexer state. |
| 250 | + current_offset (int): The offset in the input text, i.e. the number |
| 251 | + of characters already scanned so far. |
| 252 | +
|
| 253 | + Yields: |
| 254 | + The next ``Token`` or ``StateTransition`` instance. |
| 255 | + """ |
| 256 | + pattern = self._GRAND_PATTERNS[state] |
| 257 | + scanner = pattern.finditer(self._text, current_offset) |
| 258 | + |
| 259 | + for match in scanner: # pragma: NO COVER |
| 260 | + token_type = match.lastgroup |
| 261 | + |
| 262 | + if token_type.startswith("GOTO_"): |
| 263 | + yield StateTransition( |
| 264 | + new_state=getattr(LexerState, token_type[5:]), # w/o "GOTO_" prefix |
| 265 | + total_offset=match.start(), |
| 266 | + ) |
| 267 | + |
| 268 | + yield Token(token_type, match.group(), match.start()) |
0 commit comments