You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
91 lines
3.4 KiB
91 lines
3.4 KiB
4 years ago
|
# -*- coding: utf-8 -*-
|
||
|
"""
|
||
|
pygments.lexers.usd
|
||
|
~~~~~~~~~~~~~~~~~~~
|
||
|
|
||
|
The module that parses Pixar's Universal Scene Description file format.
|
||
|
|
||
|
:copyright: Copyright 2006-2019 by the Pygments team, see AUTHORS.
|
||
|
:license: BSD, see LICENSE for details.
|
||
|
"""
|
||
|
|
||
|
from pygments.lexer import RegexLexer, bygroups
|
||
|
from pygments.lexer import words as words_
|
||
|
from pygments.lexers._usd_builtins import COMMON_ATTRIBUTES, KEYWORDS, \
|
||
|
OPERATORS, SPECIAL_NAMES, TYPES
|
||
|
from pygments.token import Comment, Keyword, Name, Number, Operator, \
|
||
|
Punctuation, String, Text, Whitespace
|
||
|
|
||
|
__all__ = ["UsdLexer"]
|
||
|
|
||
|
|
||
|
def _keywords(words, type_):
|
||
|
return [(words_(words, prefix=r"\b", suffix=r"\b"), type_)]
|
||
|
|
||
|
|
||
|
_TYPE = r"(\w+(?:\[\])?)"
|
||
|
_BASE_ATTRIBUTE = r"([\w_]+(?:\:[\w_]+)*)(?:(\.)(timeSamples))?"
|
||
|
_WHITESPACE = r"([ \t]+)"
|
||
|
|
||
|
|
||
|
class UsdLexer(RegexLexer):
|
||
|
"""
|
||
|
A lexer that parses Pixar's Universal Scene Description file format.
|
||
|
|
||
|
.. versionadded:: 2.6
|
||
|
"""
|
||
|
|
||
|
name = "USD"
|
||
|
aliases = ["usd", "usda"]
|
||
|
filenames = ["*.usd", "*.usda"]
|
||
|
|
||
|
tokens = {
|
||
|
"root": [
|
||
|
(r"(custom){_WHITESPACE}(uniform)(\s+){}(\s+){}(\s*)(=)".format(
|
||
|
_TYPE, _BASE_ATTRIBUTE, _WHITESPACE=_WHITESPACE),
|
||
|
bygroups(Keyword.Token, Whitespace, Keyword.Token, Whitespace,
|
||
|
Keyword.Type, Whitespace, Name.Attribute, Text,
|
||
|
Name.Keyword.Tokens, Whitespace, Operator)),
|
||
|
(r"(custom){_WHITESPACE}{}(\s+){}(\s*)(=)".format(
|
||
|
_TYPE, _BASE_ATTRIBUTE, _WHITESPACE=_WHITESPACE),
|
||
|
bygroups(Keyword.Token, Whitespace, Keyword.Type, Whitespace,
|
||
|
Name.Attribute, Text, Name.Keyword.Tokens, Whitespace,
|
||
|
Operator)),
|
||
|
(r"(uniform){_WHITESPACE}{}(\s+){}(\s*)(=)".format(
|
||
|
_TYPE, _BASE_ATTRIBUTE, _WHITESPACE=_WHITESPACE),
|
||
|
bygroups(Keyword.Token, Whitespace, Keyword.Type, Whitespace,
|
||
|
Name.Attribute, Text, Name.Keyword.Tokens, Whitespace,
|
||
|
Operator)),
|
||
|
(r"{}{_WHITESPACE}{}(\s*)(=)".format(
|
||
|
_TYPE, _BASE_ATTRIBUTE, _WHITESPACE=_WHITESPACE),
|
||
|
bygroups(Keyword.Type, Whitespace, Name.Attribute, Text,
|
||
|
Name.Keyword.Tokens, Whitespace, Operator)),
|
||
|
] +
|
||
|
_keywords(KEYWORDS, Keyword.Tokens) +
|
||
|
_keywords(SPECIAL_NAMES, Name.Builtins) +
|
||
|
_keywords(COMMON_ATTRIBUTES, Name.Attribute) +
|
||
|
[(r"\b\w+:[\w:]+\b", Name.Attribute)] +
|
||
|
_keywords(OPERATORS, Operator) + # more attributes
|
||
|
[(type_ + r"\[\]", Keyword.Type) for type_ in TYPES] +
|
||
|
_keywords(TYPES, Keyword.Type) +
|
||
|
[
|
||
|
(r"[\(\)\[\]{}]", Punctuation),
|
||
|
("#.*?$", Comment.Single),
|
||
|
(",", Punctuation),
|
||
|
(";", Punctuation), # ";"s are allowed to combine separate metadata lines
|
||
|
("=", Operator),
|
||
|
(r"[-]*([0-9]*[.])?[0-9]+(?:e[+-]*\d+)?", Number),
|
||
|
(r"'''(?:.|\n)*?'''", String),
|
||
|
(r'"""(?:.|\n)*?"""', String),
|
||
|
(r"'.*?'", String),
|
||
|
(r'".*?"', String),
|
||
|
(r"<(\.\./)*([\w/]+|[\w/]+\.\w+[\w:]*)>", Name.Namespace),
|
||
|
(r"@.*?@", String.Interpol),
|
||
|
(r'\(.*"[.\\n]*".*\)', String.Doc),
|
||
|
(r"\A#usda .+$", Comment.Hashbang),
|
||
|
(r"\s+", Whitespace),
|
||
|
(r"[\w_]+", Text),
|
||
|
(r"[_:\.]+", Punctuation),
|
||
|
],
|
||
|
}
|