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.
ORPA-pyOpenRPA/Resources/WPy32-3720/python-3.7.2/Lib/site-packages/pygments/lexers/csound.py

457 lines
16 KiB

This file contains ambiguous Unicode characters!

This file contains ambiguous Unicode characters that may be confused with others in your current locale. If your use case is intentional and legitimate, you can safely ignore this warning. Use the Escape button to highlight these characters.

# -*- coding: utf-8 -*-
"""
pygments.lexers.csound
~~~~~~~~~~~~~~~~~~~~~~
Lexers for Csound languages.
:copyright: Copyright 2006-2017 by the Pygments team, see AUTHORS.
:license: BSD, see LICENSE for details.
"""
import re
from pygments.lexer import RegexLexer, bygroups, default, include, using, words
from pygments.token import Comment, Error, Keyword, Name, Number, Operator, Punctuation, \
String, Text, Whitespace
from pygments.lexers._csound_builtins import OPCODES, DEPRECATED_OPCODES
from pygments.lexers.html import HtmlLexer
from pygments.lexers.python import PythonLexer
from pygments.lexers.scripting import LuaLexer
__all__ = ['CsoundScoreLexer', 'CsoundOrchestraLexer', 'CsoundDocumentLexer']
newline = (r'((?:(?:;|//).*)*)(\n)', bygroups(Comment.Single, Text))
class CsoundLexer(RegexLexer):
tokens = {
'whitespace': [
(r'[ \t]+', Text),
(r'/[*](?:.|\n)*?[*]/', Comment.Multiline),
(r'(?:;|//).*$', Comment.Single),
(r'(\\)(\n)', bygroups(Whitespace, Text))
],
'preprocessor directives': [
(r'#(?:e(?:nd(?:if)?|lse)\b|##)|@@?[ \t]*\d+', Comment.Preproc),
(r'#include', Comment.Preproc, 'include directive'),
(r'#[ \t]*define', Comment.Preproc, 'define directive'),
(r'#(?:ifn?def|undef)\b', Comment.Preproc, 'macro directive')
],
'include directive': [
include('whitespace'),
(r'([^ \t]).*?\1', String, '#pop')
],
'define directive': [
(r'\n', Text),
include('whitespace'),
(r'([A-Z_a-z]\w*)(\()', bygroups(Comment.Preproc, Punctuation),
('#pop', 'macro parameter name list')),
(r'[A-Z_a-z]\w*', Comment.Preproc, ('#pop', 'before macro body'))
],
'macro parameter name list': [
include('whitespace'),
(r'[A-Z_a-z]\w*', Comment.Preproc),
(r"['#]", Punctuation),
(r'\)', Punctuation, ('#pop', 'before macro body'))
],
'before macro body': [
(r'\n', Text),
include('whitespace'),
(r'#', Punctuation, ('#pop', 'macro body'))
],
'macro body': [
(r'(?:\\(?!#)|[^#\\]|\n)+', Comment.Preproc),
(r'\\#', Comment.Preproc),
(r'(?<!\\)#', Punctuation, '#pop')
],
'macro directive': [
include('whitespace'),
(r'[A-Z_a-z]\w*', Comment.Preproc, '#pop')
],
'macro uses': [
(r'(\$[A-Z_a-z]\w*\.?)(\()', bygroups(Comment.Preproc, Punctuation),
'macro parameter value list'),
(r'\$[A-Z_a-z]\w*(?:\.|\b)', Comment.Preproc)
],
'macro parameter value list': [
(r'(?:[^\'#"{()]|\{(?!\{))+', Comment.Preproc),
(r"['#]", Punctuation),
(r'"', String, 'macro parameter value quoted string'),
(r'\{\{', String, 'macro parameter value braced string'),
(r'\(', Comment.Preproc, 'macro parameter value parenthetical'),
(r'\)', Punctuation, '#pop')
],
'macro parameter value quoted string': [
(r"\\[#'()]", Comment.Preproc),
(r"[#'()]", Error),
include('quoted string')
],
'macro parameter value braced string': [
(r"\\[#'()]", Comment.Preproc),
(r"[#'()]", Error),
include('braced string')
],
'macro parameter value parenthetical': [
(r'(?:[^\\()]|\\\))+', Comment.Preproc),
(r'\(', Comment.Preproc, '#push'),
(r'\)', Comment.Preproc, '#pop')
],
'whitespace and macro uses': [
include('whitespace'),
include('macro uses')
],
'numbers': [
(r'\d+[Ee][+-]?\d+|(\d+\.\d*|\d*\.\d+)([Ee][+-]?\d+)?', Number.Float),
(r'(0[Xx])([0-9A-Fa-f]+)', bygroups(Keyword.Type, Number.Hex)),
(r'\d+', Number.Integer)
],
'braced string': [
# Do nothing. This must be defined in subclasses.
]
}
class CsoundScoreLexer(CsoundLexer):
"""
For `Csound <https://csound.github.io>`_ scores.
.. versionadded:: 2.1
"""
name = 'Csound Score'
aliases = ['csound-score', 'csound-sco']
filenames = ['*.sco']
tokens = {
'root': [
(r'\n', Text),
include('whitespace and macro uses'),
include('preprocessor directives'),
(r'[abCdefiqstvxy]', Keyword),
# There is also a w statement that is generated internally and should not be
# used; see https://github.com/csound/csound/issues/750.
(r'z', Keyword.Constant),
# z is a constant equal to 800,000,000,000. 800 billion seconds is about
# 25,367.8 years. See also
# https://csound.github.io/docs/manual/ScoreTop.html and
# https://github.com/csound/csound/search?q=stof+path%3AEngine+filename%3Asread.c.
(r'([nNpP][pP])(\d+)', bygroups(Keyword, Number.Integer)),
(r'[mn]', Keyword, 'mark statement'),
include('numbers'),
(r'[!+\-*/^%&|<>#~.]', Operator),
(r'[()\[\]]', Punctuation),
(r'"', String, 'quoted string'),
(r'\{', Comment.Preproc, 'loop after left brace'),
],
'mark statement': [
include('whitespace and macro uses'),
(r'[A-Z_a-z]\w*', Name.Label),
(r'\n', Text, '#pop')
],
'quoted string': [
(r'"', String, '#pop'),
(r'[^"$]+', String),
include('macro uses'),
(r'[$]', String)
],
'loop after left brace': [
include('whitespace and macro uses'),
(r'\d+', Number.Integer, ('#pop', 'loop after repeat count')),
],
'loop after repeat count': [
include('whitespace and macro uses'),
(r'[A-Z_a-z]\w*', Comment.Preproc, ('#pop', 'loop'))
],
'loop': [
(r'\}', Comment.Preproc, '#pop'),
include('root')
],
# Braced strings are not allowed in Csound scores, but this is needed
# because the superclass includes it.
'braced string': [
(r'\}\}', String, '#pop'),
(r'[^}]|\}(?!\})', String)
]
}
class CsoundOrchestraLexer(CsoundLexer):
"""
For `Csound <https://csound.github.io>`_ orchestras.
.. versionadded:: 2.1
"""
name = 'Csound Orchestra'
aliases = ['csound', 'csound-orc']
filenames = ['*.orc', '*.udo']
user_defined_opcodes = set()
def opcode_name_callback(lexer, match):
opcode = match.group(0)
lexer.user_defined_opcodes.add(opcode)
yield match.start(), Name.Function, opcode
def name_callback(lexer, match):
name = match.group(1)
if name in OPCODES or name in DEPRECATED_OPCODES:
yield match.start(), Name.Builtin, name
if match.group(2):
yield match.start(2), Punctuation, match.group(2)
yield match.start(3), Keyword.Type, match.group(3)
elif name in lexer.user_defined_opcodes:
yield match.start(), Name.Function, name
else:
nameMatch = re.search(r'^(g?[afikSw])(\w+)', name)
if nameMatch:
yield nameMatch.start(1), Keyword.Type, nameMatch.group(1)
yield nameMatch.start(2), Name, nameMatch.group(2)
else:
yield match.start(), Name, name
if match.group(2):
yield match.start(2), Punctuation, match.group(2)
yield match.start(3), Name, match.group(3)
tokens = {
'root': [
(r'\n', Text),
(r'^([ \t]*)(\w+)(:)(?:[ \t]+|$)', bygroups(Text, Name.Label, Punctuation)),
include('whitespace and macro uses'),
include('preprocessor directives'),
(r'\binstr\b', Keyword.Declaration, 'instrument numbers and identifiers'),
(r'\bopcode\b', Keyword.Declaration, 'after opcode keyword'),
(r'\b(?:end(?:in|op))\b', Keyword.Declaration),
include('partial statements')
],
'partial statements': [
(r'\b(?:0dbfs|A4|k(?:r|smps)|nchnls(?:_i)?|sr)\b', Name.Variable.Global),
include('numbers'),
(r'\+=|-=|\*=|/=|<<|>>|<=|>=|==|!=|&&|\|\||[~¬]|[=!+\-*/^%&|<>#?:]', Operator),
(r'[(),\[\]]', Punctuation),
(r'"', String, 'quoted string'),
(r'\{\{', String, 'braced string'),
(words((
'do', 'else', 'elseif', 'endif', 'enduntil', 'fi', 'if', 'ithen', 'kthen',
'od', 'then', 'until', 'while',
), prefix=r'\b', suffix=r'\b'), Keyword),
(words(('return', 'rireturn'), prefix=r'\b', suffix=r'\b'), Keyword.Pseudo),
(r'\b[ik]?goto\b', Keyword, 'goto label'),
(r'\b(r(?:einit|igoto)|tigoto)(\(|\b)', bygroups(Keyword.Pseudo, Punctuation),
'goto label'),
(r'\b(c(?:g|in?|k|nk?)goto)(\(|\b)', bygroups(Keyword.Pseudo, Punctuation),
('goto label', 'goto argument')),
(r'\b(timout)(\(|\b)', bygroups(Keyword.Pseudo, Punctuation),
('goto label', 'goto argument', 'goto argument')),
(r'\b(loop_[gl][et])(\(|\b)', bygroups(Keyword.Pseudo, Punctuation),
('goto label', 'goto argument', 'goto argument', 'goto argument')),
(r'\bprintk?s\b', Name.Builtin, 'prints opcode'),
(r'\b(?:readscore|scoreline(?:_i)?)\b', Name.Builtin, 'Csound score opcode'),
(r'\bpyl?run[it]?\b', Name.Builtin, 'Python opcode'),
(r'\blua_(?:exec|opdef)\b', Name.Builtin, 'Lua opcode'),
(r'\bp\d+\b', Name.Variable.Instance),
(r'\b([A-Z_a-z]\w*)(?:(:)([A-Za-z]))?\b', name_callback)
],
'instrument numbers and identifiers': [
include('whitespace and macro uses'),
(r'\d+|[A-Z_a-z]\w*', Name.Function),
(r'[+,]', Punctuation),
(r'\n', Text, '#pop')
],
'after opcode keyword': [
include('whitespace and macro uses'),
(r'[A-Z_a-z]\w*', opcode_name_callback, ('#pop', 'opcode type signatures')),
(r'\n', Text, '#pop')
],
'opcode type signatures': [
include('whitespace and macro uses'),
# https://github.com/csound/csound/search?q=XIDENT+path%3AEngine+filename%3Acsound_orc.lex
(r'0|[afijkKoOpPStV\[\]]+', Keyword.Type),
(r',', Punctuation),
(r'\n', Text, '#pop')
],
'quoted string': [
(r'"', String, '#pop'),
(r'[^\\"$%)]+', String),
include('macro uses'),
include('escape sequences'),
include('format specifiers'),
(r'[\\$%)]', String)
],
'braced string': [
(r'\}\}', String, '#pop'),
(r'(?:[^\\%)}]|\}(?!\}))+', String),
include('escape sequences'),
include('format specifiers'),
(r'[\\%)]', String)
],
'escape sequences': [
# https://github.com/csound/csound/search?q=unquote_string+path%3AEngine+filename%3Acsound_orc_compile.c
(r'\\(?:[\\abnrt"]|[0-7]{1,3})', String.Escape)
],
# Format specifiers are highlighted in all strings, even though only
# fprintks https://csound.github.io/docs/manual/fprintks.html
# fprints https://csound.github.io/docs/manual/fprints.html
# printf/printf_i https://csound.github.io/docs/manual/printf.html
# printks https://csound.github.io/docs/manual/printks.html
# prints https://csound.github.io/docs/manual/prints.html
# sprintf https://csound.github.io/docs/manual/sprintf.html
# sprintfk https://csound.github.io/docs/manual/sprintfk.html
# work with strings that contain format specifiers. In addition, these
# opcodes handling of format specifiers is inconsistent:
# - fprintks, fprints, printks, and prints do accept %a and %A
# specifiers, but cant accept %s specifiers.
# - printf, printf_i, sprintf, and sprintfk dont accept %a and %A
# specifiers, but can accept %s specifiers.
# See https://github.com/csound/csound/issues/747 for more information.
'format specifiers': [
(r'%[#0\- +]*\d*(?:\.\d+)?[diuoxXfFeEgGaAcs]', String.Interpol),
(r'%%', String.Escape)
],
'goto argument': [
include('whitespace and macro uses'),
(r',', Punctuation, '#pop'),
include('partial statements')
],
'goto label': [
include('whitespace and macro uses'),
(r'\w+', Name.Label, '#pop'),
default('#pop')
],
'prints opcode': [
include('whitespace and macro uses'),
(r'"', String, 'prints quoted string'),
default('#pop')
],
'prints quoted string': [
(r'\\\\[aAbBnNrRtT]', String.Escape),
(r'%[!nNrRtT]|[~^]{1,2}', String.Escape),
include('quoted string')
],
'Csound score opcode': [
include('whitespace and macro uses'),
(r'\{\{', String, 'Csound score'),
(r'\n', Text, '#pop')
],
'Csound score': [
(r'\}\}', String, '#pop'),
(r'([^}]+)|\}(?!\})', using(CsoundScoreLexer))
],
'Python opcode': [
include('whitespace and macro uses'),
(r'\{\{', String, 'Python'),
(r'\n', Text, '#pop')
],
'Python': [
(r'\}\}', String, '#pop'),
(r'([^}]+)|\}(?!\})', using(PythonLexer))
],
'Lua opcode': [
include('whitespace and macro uses'),
(r'\{\{', String, 'Lua'),
(r'\n', Text, '#pop')
],
'Lua': [
(r'\}\}', String, '#pop'),
(r'([^}]+)|\}(?!\})', using(LuaLexer))
]
}
class CsoundDocumentLexer(RegexLexer):
"""
For `Csound <https://csound.github.io>`_ documents.
.. versionadded:: 2.1
"""
name = 'Csound Document'
aliases = ['csound-document', 'csound-csd']
filenames = ['*.csd']
# These tokens are based on those in XmlLexer in pygments/lexers/html.py. Making
# CsoundDocumentLexer a subclass of XmlLexer rather than RegexLexer may seem like a
# better idea, since Csound Document files look like XML files. However, Csound
# Documents can contain Csound comments (preceded by //, for example) before and
# after the root element, unescaped bitwise AND & and less than < operators, etc. In
# other words, while Csound Document files look like XML files, they may not actually
# be XML files.
tokens = {
'root': [
(r'/[*](.|\n)*?[*]/', Comment.Multiline),
(r'(?:;|//).*$', Comment.Single),
(r'[^/;<]+|/(?!/)', Text),
(r'<\s*CsInstruments', Name.Tag, ('orchestra', 'tag')),
(r'<\s*CsScore', Name.Tag, ('score', 'tag')),
(r'<\s*[Hh][Tt][Mm][Ll]', Name.Tag, ('HTML', 'tag')),
(r'<\s*[\w:.-]+', Name.Tag, 'tag'),
(r'<\s*/\s*[\w:.-]+\s*>', Name.Tag)
],
'orchestra': [
(r'<\s*/\s*CsInstruments\s*>', Name.Tag, '#pop'),
(r'(.|\n)+?(?=<\s*/\s*CsInstruments\s*>)', using(CsoundOrchestraLexer))
],
'score': [
(r'<\s*/\s*CsScore\s*>', Name.Tag, '#pop'),
(r'(.|\n)+?(?=<\s*/\s*CsScore\s*>)', using(CsoundScoreLexer))
],
'HTML': [
(r'<\s*/\s*[Hh][Tt][Mm][Ll]\s*>', Name.Tag, '#pop'),
(r'(.|\n)+?(?=<\s*/\s*[Hh][Tt][Mm][Ll]\s*>)', using(HtmlLexer))
],
'tag': [
(r'\s+', Text),
(r'[\w.:-]+\s*=', Name.Attribute, 'attr'),
(r'/?\s*>', Name.Tag, '#pop')
],
'attr': [
(r'\s+', Text),
(r'".*?"', String, '#pop'),
(r"'.*?'", String, '#pop'),
(r'[^\s>]+', String, '#pop')
]
}