Vehicle-Anti-Theft-Face-Rec.../venv/Lib/site-packages/pygments/lexers/basic.py

662 lines
27 KiB
Python

# -*- coding: utf-8 -*-
"""
pygments.lexers.basic
~~~~~~~~~~~~~~~~~~~~~
Lexers for BASIC like languages (other than VB.net).
:copyright: Copyright 2006-2020 by the Pygments team, see AUTHORS.
:license: BSD, see LICENSE for details.
"""
import re
from pygments.lexer import RegexLexer, bygroups, default, words, include
from pygments.token import Comment, Error, Keyword, Name, Number, \
Punctuation, Operator, String, Text, Whitespace
from pygments.lexers import _vbscript_builtins
__all__ = ['BlitzBasicLexer', 'BlitzMaxLexer', 'MonkeyLexer', 'CbmBasicV2Lexer',
'QBasicLexer', 'VBScriptLexer', 'BBCBasicLexer']
class BlitzMaxLexer(RegexLexer):
"""
For `BlitzMax <http://blitzbasic.com>`_ source code.
.. versionadded:: 1.4
"""
name = 'BlitzMax'
aliases = ['blitzmax', 'bmax']
filenames = ['*.bmx']
mimetypes = ['text/x-bmx']
bmax_vopwords = r'\b(Shl|Shr|Sar|Mod)\b'
bmax_sktypes = r'@{1,2}|[!#$%]'
bmax_lktypes = r'\b(Int|Byte|Short|Float|Double|Long)\b'
bmax_name = r'[a-z_]\w*'
bmax_var = (r'(%s)(?:(?:([ \t]*)(%s)|([ \t]*:[ \t]*\b(?:Shl|Shr|Sar|Mod)\b)'
r'|([ \t]*)(:)([ \t]*)(?:%s|(%s)))(?:([ \t]*)(Ptr))?)') % \
(bmax_name, bmax_sktypes, bmax_lktypes, bmax_name)
bmax_func = bmax_var + r'?((?:[ \t]|\.\.\n)*)([(])'
flags = re.MULTILINE | re.IGNORECASE
tokens = {
'root': [
# Text
(r'[ \t]+', Text),
(r'\.\.\n', Text), # Line continuation
# Comments
(r"'.*?\n", Comment.Single),
(r'([ \t]*)\bRem\n(\n|.)*?\s*\bEnd([ \t]*)Rem', Comment.Multiline),
# Data types
('"', String.Double, 'string'),
# Numbers
(r'[0-9]+\.[0-9]*(?!\.)', Number.Float),
(r'\.[0-9]*(?!\.)', Number.Float),
(r'[0-9]+', Number.Integer),
(r'\$[0-9a-f]+', Number.Hex),
(r'\%[10]+', Number.Bin),
# Other
(r'(?:(?:(:)?([ \t]*)(:?%s|([+\-*/&|~]))|Or|And|Not|[=<>^]))' %
(bmax_vopwords), Operator),
(r'[(),.:\[\]]', Punctuation),
(r'(?:#[\w \t]*)', Name.Label),
(r'(?:\?[\w \t]*)', Comment.Preproc),
# Identifiers
(r'\b(New)\b([ \t]?)([(]?)(%s)' % (bmax_name),
bygroups(Keyword.Reserved, Text, Punctuation, Name.Class)),
(r'\b(Import|Framework|Module)([ \t]+)(%s\.%s)' %
(bmax_name, bmax_name),
bygroups(Keyword.Reserved, Text, Keyword.Namespace)),
(bmax_func, bygroups(Name.Function, Text, Keyword.Type,
Operator, Text, Punctuation, Text,
Keyword.Type, Name.Class, Text,
Keyword.Type, Text, Punctuation)),
(bmax_var, bygroups(Name.Variable, Text, Keyword.Type, Operator,
Text, Punctuation, Text, Keyword.Type,
Name.Class, Text, Keyword.Type)),
(r'\b(Type|Extends)([ \t]+)(%s)' % (bmax_name),
bygroups(Keyword.Reserved, Text, Name.Class)),
# Keywords
(r'\b(Ptr)\b', Keyword.Type),
(r'\b(Pi|True|False|Null|Self|Super)\b', Keyword.Constant),
(r'\b(Local|Global|Const|Field)\b', Keyword.Declaration),
(words((
'TNullMethodException', 'TNullFunctionException',
'TNullObjectException', 'TArrayBoundsException',
'TRuntimeException'), prefix=r'\b', suffix=r'\b'), Name.Exception),
(words((
'Strict', 'SuperStrict', 'Module', 'ModuleInfo',
'End', 'Return', 'Continue', 'Exit', 'Public', 'Private',
'Var', 'VarPtr', 'Chr', 'Len', 'Asc', 'SizeOf', 'Sgn', 'Abs', 'Min', 'Max',
'New', 'Release', 'Delete', 'Incbin', 'IncbinPtr', 'IncbinLen',
'Framework', 'Include', 'Import', 'Extern', 'EndExtern',
'Function', 'EndFunction', 'Type', 'EndType', 'Extends', 'Method', 'EndMethod',
'Abstract', 'Final', 'If', 'Then', 'Else', 'ElseIf', 'EndIf',
'For', 'To', 'Next', 'Step', 'EachIn', 'While', 'Wend', 'EndWhile',
'Repeat', 'Until', 'Forever', 'Select', 'Case', 'Default', 'EndSelect',
'Try', 'Catch', 'EndTry', 'Throw', 'Assert', 'Goto', 'DefData', 'ReadData',
'RestoreData'), prefix=r'\b', suffix=r'\b'),
Keyword.Reserved),
# Final resolve (for variable names and such)
(r'(%s)' % (bmax_name), Name.Variable),
],
'string': [
(r'""', String.Double),
(r'"C?', String.Double, '#pop'),
(r'[^"]+', String.Double),
],
}
class BlitzBasicLexer(RegexLexer):
"""
For `BlitzBasic <http://blitzbasic.com>`_ source code.
.. versionadded:: 2.0
"""
name = 'BlitzBasic'
aliases = ['blitzbasic', 'b3d', 'bplus']
filenames = ['*.bb', '*.decls']
mimetypes = ['text/x-bb']
bb_sktypes = r'@{1,2}|[#$%]'
bb_name = r'[a-z]\w*'
bb_var = (r'(%s)(?:([ \t]*)(%s)|([ \t]*)([.])([ \t]*)(?:(%s)))?') % \
(bb_name, bb_sktypes, bb_name)
flags = re.MULTILINE | re.IGNORECASE
tokens = {
'root': [
# Text
(r'[ \t]+', Text),
# Comments
(r";.*?\n", Comment.Single),
# Data types
('"', String.Double, 'string'),
# Numbers
(r'[0-9]+\.[0-9]*(?!\.)', Number.Float),
(r'\.[0-9]+(?!\.)', Number.Float),
(r'[0-9]+', Number.Integer),
(r'\$[0-9a-f]+', Number.Hex),
(r'\%[10]+', Number.Bin),
# Other
(words(('Shl', 'Shr', 'Sar', 'Mod', 'Or', 'And', 'Not',
'Abs', 'Sgn', 'Handle', 'Int', 'Float', 'Str',
'First', 'Last', 'Before', 'After'),
prefix=r'\b', suffix=r'\b'),
Operator),
(r'([+\-*/~=<>^])', Operator),
(r'[(),:\[\]\\]', Punctuation),
(r'\.([ \t]*)(%s)' % bb_name, Name.Label),
# Identifiers
(r'\b(New)\b([ \t]+)(%s)' % (bb_name),
bygroups(Keyword.Reserved, Text, Name.Class)),
(r'\b(Gosub|Goto)\b([ \t]+)(%s)' % (bb_name),
bygroups(Keyword.Reserved, Text, Name.Label)),
(r'\b(Object)\b([ \t]*)([.])([ \t]*)(%s)\b' % (bb_name),
bygroups(Operator, Text, Punctuation, Text, Name.Class)),
(r'\b%s\b([ \t]*)(\()' % bb_var,
bygroups(Name.Function, Text, Keyword.Type, Text, Punctuation,
Text, Name.Class, Text, Punctuation)),
(r'\b(Function)\b([ \t]+)%s' % bb_var,
bygroups(Keyword.Reserved, Text, Name.Function, Text, Keyword.Type,
Text, Punctuation, Text, Name.Class)),
(r'\b(Type)([ \t]+)(%s)' % (bb_name),
bygroups(Keyword.Reserved, Text, Name.Class)),
# Keywords
(r'\b(Pi|True|False|Null)\b', Keyword.Constant),
(r'\b(Local|Global|Const|Field|Dim)\b', Keyword.Declaration),
(words((
'End', 'Return', 'Exit', 'Chr', 'Len', 'Asc', 'New', 'Delete', 'Insert',
'Include', 'Function', 'Type', 'If', 'Then', 'Else', 'ElseIf', 'EndIf',
'For', 'To', 'Next', 'Step', 'Each', 'While', 'Wend',
'Repeat', 'Until', 'Forever', 'Select', 'Case', 'Default',
'Goto', 'Gosub', 'Data', 'Read', 'Restore'), prefix=r'\b', suffix=r'\b'),
Keyword.Reserved),
# Final resolve (for variable names and such)
# (r'(%s)' % (bb_name), Name.Variable),
(bb_var, bygroups(Name.Variable, Text, Keyword.Type,
Text, Punctuation, Text, Name.Class)),
],
'string': [
(r'""', String.Double),
(r'"C?', String.Double, '#pop'),
(r'[^"]+', String.Double),
],
}
class MonkeyLexer(RegexLexer):
"""
For
`Monkey <https://en.wikipedia.org/wiki/Monkey_(programming_language)>`_
source code.
.. versionadded:: 1.6
"""
name = 'Monkey'
aliases = ['monkey']
filenames = ['*.monkey']
mimetypes = ['text/x-monkey']
name_variable = r'[a-z_]\w*'
name_function = r'[A-Z]\w*'
name_constant = r'[A-Z_][A-Z0-9_]*'
name_class = r'[A-Z]\w*'
name_module = r'[a-z0-9_]*'
keyword_type = r'(?:Int|Float|String|Bool|Object|Array|Void)'
# ? == Bool // % == Int // # == Float // $ == String
keyword_type_special = r'[?%#$]'
flags = re.MULTILINE
tokens = {
'root': [
# Text
(r'\s+', Text),
# Comments
(r"'.*", Comment),
(r'(?i)^#rem\b', Comment.Multiline, 'comment'),
# preprocessor directives
(r'(?i)^(?:#If|#ElseIf|#Else|#EndIf|#End|#Print|#Error)\b', Comment.Preproc),
# preprocessor variable (any line starting with '#' that is not a directive)
(r'^#', Comment.Preproc, 'variables'),
# String
('"', String.Double, 'string'),
# Numbers
(r'[0-9]+\.[0-9]*(?!\.)', Number.Float),
(r'\.[0-9]+(?!\.)', Number.Float),
(r'[0-9]+', Number.Integer),
(r'\$[0-9a-fA-Z]+', Number.Hex),
(r'\%[10]+', Number.Bin),
# Native data types
(r'\b%s\b' % keyword_type, Keyword.Type),
# Exception handling
(r'(?i)\b(?:Try|Catch|Throw)\b', Keyword.Reserved),
(r'Throwable', Name.Exception),
# Builtins
(r'(?i)\b(?:Null|True|False)\b', Name.Builtin),
(r'(?i)\b(?:Self|Super)\b', Name.Builtin.Pseudo),
(r'\b(?:HOST|LANG|TARGET|CONFIG)\b', Name.Constant),
# Keywords
(r'(?i)^(Import)(\s+)(.*)(\n)',
bygroups(Keyword.Namespace, Text, Name.Namespace, Text)),
(r'(?i)^Strict\b.*\n', Keyword.Reserved),
(r'(?i)(Const|Local|Global|Field)(\s+)',
bygroups(Keyword.Declaration, Text), 'variables'),
(r'(?i)(New|Class|Interface|Extends|Implements)(\s+)',
bygroups(Keyword.Reserved, Text), 'classname'),
(r'(?i)(Function|Method)(\s+)',
bygroups(Keyword.Reserved, Text), 'funcname'),
(r'(?i)(?:End|Return|Public|Private|Extern|Property|'
r'Final|Abstract)\b', Keyword.Reserved),
# Flow Control stuff
(r'(?i)(?:If|Then|Else|ElseIf|EndIf|'
r'Select|Case|Default|'
r'While|Wend|'
r'Repeat|Until|Forever|'
r'For|To|Until|Step|EachIn|Next|'
r'Exit|Continue)\s+', Keyword.Reserved),
# not used yet
(r'(?i)\b(?:Module|Inline)\b', Keyword.Reserved),
# Array
(r'[\[\]]', Punctuation),
# Other
(r'<=|>=|<>|\*=|/=|\+=|-=|&=|~=|\|=|[-&*/^+=<>|~]', Operator),
(r'(?i)(?:Not|Mod|Shl|Shr|And|Or)', Operator.Word),
(r'[(){}!#,.:]', Punctuation),
# catch the rest
(r'%s\b' % name_constant, Name.Constant),
(r'%s\b' % name_function, Name.Function),
(r'%s\b' % name_variable, Name.Variable),
],
'funcname': [
(r'(?i)%s\b' % name_function, Name.Function),
(r':', Punctuation, 'classname'),
(r'\s+', Text),
(r'\(', Punctuation, 'variables'),
(r'\)', Punctuation, '#pop')
],
'classname': [
(r'%s\.' % name_module, Name.Namespace),
(r'%s\b' % keyword_type, Keyword.Type),
(r'%s\b' % name_class, Name.Class),
# array (of given size)
(r'(\[)(\s*)(\d*)(\s*)(\])',
bygroups(Punctuation, Text, Number.Integer, Text, Punctuation)),
# generics
(r'\s+(?!<)', Text, '#pop'),
(r'<', Punctuation, '#push'),
(r'>', Punctuation, '#pop'),
(r'\n', Text, '#pop'),
default('#pop')
],
'variables': [
(r'%s\b' % name_constant, Name.Constant),
(r'%s\b' % name_variable, Name.Variable),
(r'%s' % keyword_type_special, Keyword.Type),
(r'\s+', Text),
(r':', Punctuation, 'classname'),
(r',', Punctuation, '#push'),
default('#pop')
],
'string': [
(r'[^"~]+', String.Double),
(r'~q|~n|~r|~t|~z|~~', String.Escape),
(r'"', String.Double, '#pop'),
],
'comment': [
(r'(?i)^#rem.*?', Comment.Multiline, "#push"),
(r'(?i)^#end.*?', Comment.Multiline, "#pop"),
(r'\n', Comment.Multiline),
(r'.+', Comment.Multiline),
],
}
class CbmBasicV2Lexer(RegexLexer):
"""
For CBM BASIC V2 sources.
.. versionadded:: 1.6
"""
name = 'CBM BASIC V2'
aliases = ['cbmbas']
filenames = ['*.bas']
flags = re.IGNORECASE
tokens = {
'root': [
(r'rem.*\n', Comment.Single),
(r'\s+', Text),
(r'new|run|end|for|to|next|step|go(to|sub)?|on|return|stop|cont'
r'|if|then|input#?|read|wait|load|save|verify|poke|sys|print#?'
r'|list|clr|cmd|open|close|get#?', Keyword.Reserved),
(r'data|restore|dim|let|def|fn', Keyword.Declaration),
(r'tab|spc|sgn|int|abs|usr|fre|pos|sqr|rnd|log|exp|cos|sin|tan|atn'
r'|peek|len|val|asc|(str|chr|left|right|mid)\$', Name.Builtin),
(r'[-+*/^<>=]', Operator),
(r'not|and|or', Operator.Word),
(r'"[^"\n]*.', String),
(r'\d+|[-+]?\d*\.\d*(e[-+]?\d+)?', Number.Float),
(r'[(),:;]', Punctuation),
(r'\w+[$%]?', Name),
]
}
def analyse_text(self, text):
# if it starts with a line number, it shouldn't be a "modern" Basic
# like VB.net
if re.match(r'\d+', text):
return 0.2
class QBasicLexer(RegexLexer):
"""
For
`QBasic <http://en.wikipedia.org/wiki/QBasic>`_
source code.
.. versionadded:: 2.0
"""
name = 'QBasic'
aliases = ['qbasic', 'basic']
filenames = ['*.BAS', '*.bas']
mimetypes = ['text/basic']
declarations = ('DATA', 'LET')
functions = (
'ABS', 'ASC', 'ATN', 'CDBL', 'CHR$', 'CINT', 'CLNG',
'COMMAND$', 'COS', 'CSNG', 'CSRLIN', 'CVD', 'CVDMBF', 'CVI',
'CVL', 'CVS', 'CVSMBF', 'DATE$', 'ENVIRON$', 'EOF', 'ERDEV',
'ERDEV$', 'ERL', 'ERR', 'EXP', 'FILEATTR', 'FIX', 'FRE',
'FREEFILE', 'HEX$', 'INKEY$', 'INP', 'INPUT$', 'INSTR', 'INT',
'IOCTL$', 'LBOUND', 'LCASE$', 'LEFT$', 'LEN', 'LOC', 'LOF',
'LOG', 'LPOS', 'LTRIM$', 'MID$', 'MKD$', 'MKDMBF$', 'MKI$',
'MKL$', 'MKS$', 'MKSMBF$', 'OCT$', 'PEEK', 'PEN', 'PLAY',
'PMAP', 'POINT', 'POS', 'RIGHT$', 'RND', 'RTRIM$', 'SADD',
'SCREEN', 'SEEK', 'SETMEM', 'SGN', 'SIN', 'SPACE$', 'SPC',
'SQR', 'STICK', 'STR$', 'STRIG', 'STRING$', 'TAB', 'TAN',
'TIME$', 'TIMER', 'UBOUND', 'UCASE$', 'VAL', 'VARPTR',
'VARPTR$', 'VARSEG'
)
metacommands = ('$DYNAMIC', '$INCLUDE', '$STATIC')
operators = ('AND', 'EQV', 'IMP', 'NOT', 'OR', 'XOR')
statements = (
'BEEP', 'BLOAD', 'BSAVE', 'CALL', 'CALL ABSOLUTE',
'CALL INTERRUPT', 'CALLS', 'CHAIN', 'CHDIR', 'CIRCLE', 'CLEAR',
'CLOSE', 'CLS', 'COLOR', 'COM', 'COMMON', 'CONST', 'DATA',
'DATE$', 'DECLARE', 'DEF FN', 'DEF SEG', 'DEFDBL', 'DEFINT',
'DEFLNG', 'DEFSNG', 'DEFSTR', 'DEF', 'DIM', 'DO', 'LOOP',
'DRAW', 'END', 'ENVIRON', 'ERASE', 'ERROR', 'EXIT', 'FIELD',
'FILES', 'FOR', 'NEXT', 'FUNCTION', 'GET', 'GOSUB', 'GOTO',
'IF', 'THEN', 'INPUT', 'INPUT #', 'IOCTL', 'KEY', 'KEY',
'KILL', 'LET', 'LINE', 'LINE INPUT', 'LINE INPUT #', 'LOCATE',
'LOCK', 'UNLOCK', 'LPRINT', 'LSET', 'MID$', 'MKDIR', 'NAME',
'ON COM', 'ON ERROR', 'ON KEY', 'ON PEN', 'ON PLAY',
'ON STRIG', 'ON TIMER', 'ON UEVENT', 'ON', 'OPEN', 'OPEN COM',
'OPTION BASE', 'OUT', 'PAINT', 'PALETTE', 'PCOPY', 'PEN',
'PLAY', 'POKE', 'PRESET', 'PRINT', 'PRINT #', 'PRINT USING',
'PSET', 'PUT', 'PUT', 'RANDOMIZE', 'READ', 'REDIM', 'REM',
'RESET', 'RESTORE', 'RESUME', 'RETURN', 'RMDIR', 'RSET', 'RUN',
'SCREEN', 'SEEK', 'SELECT CASE', 'SHARED', 'SHELL', 'SLEEP',
'SOUND', 'STATIC', 'STOP', 'STRIG', 'SUB', 'SWAP', 'SYSTEM',
'TIME$', 'TIMER', 'TROFF', 'TRON', 'TYPE', 'UEVENT', 'UNLOCK',
'VIEW', 'WAIT', 'WHILE', 'WEND', 'WIDTH', 'WINDOW', 'WRITE'
)
keywords = (
'ACCESS', 'ALIAS', 'ANY', 'APPEND', 'AS', 'BASE', 'BINARY',
'BYVAL', 'CASE', 'CDECL', 'DOUBLE', 'ELSE', 'ELSEIF', 'ENDIF',
'INTEGER', 'IS', 'LIST', 'LOCAL', 'LONG', 'LOOP', 'MOD',
'NEXT', 'OFF', 'ON', 'OUTPUT', 'RANDOM', 'SIGNAL', 'SINGLE',
'STEP', 'STRING', 'THEN', 'TO', 'UNTIL', 'USING', 'WEND'
)
tokens = {
'root': [
(r'\n+', Text),
(r'\s+', Text.Whitespace),
(r'^(\s*)(\d*)(\s*)(REM .*)$',
bygroups(Text.Whitespace, Name.Label, Text.Whitespace,
Comment.Single)),
(r'^(\s*)(\d+)(\s*)',
bygroups(Text.Whitespace, Name.Label, Text.Whitespace)),
(r'(?=[\s]*)(\w+)(?=[\s]*=)', Name.Variable.Global),
(r'(?=[^"]*)\'.*$', Comment.Single),
(r'"[^\n"]*"', String.Double),
(r'(END)(\s+)(FUNCTION|IF|SELECT|SUB)',
bygroups(Keyword.Reserved, Text.Whitespace, Keyword.Reserved)),
(r'(DECLARE)(\s+)([A-Z]+)(\s+)(\S+)',
bygroups(Keyword.Declaration, Text.Whitespace, Name.Variable,
Text.Whitespace, Name)),
(r'(DIM)(\s+)(SHARED)(\s+)([^\s(]+)',
bygroups(Keyword.Declaration, Text.Whitespace, Name.Variable,
Text.Whitespace, Name.Variable.Global)),
(r'(DIM)(\s+)([^\s(]+)',
bygroups(Keyword.Declaration, Text.Whitespace, Name.Variable.Global)),
(r'^(\s*)([a-zA-Z_]+)(\s*)(\=)',
bygroups(Text.Whitespace, Name.Variable.Global, Text.Whitespace,
Operator)),
(r'(GOTO|GOSUB)(\s+)(\w+\:?)',
bygroups(Keyword.Reserved, Text.Whitespace, Name.Label)),
(r'(SUB)(\s+)(\w+\:?)',
bygroups(Keyword.Reserved, Text.Whitespace, Name.Label)),
include('declarations'),
include('functions'),
include('metacommands'),
include('operators'),
include('statements'),
include('keywords'),
(r'[a-zA-Z_]\w*[$@#&!]', Name.Variable.Global),
(r'[a-zA-Z_]\w*\:', Name.Label),
(r'\-?\d*\.\d+[@|#]?', Number.Float),
(r'\-?\d+[@|#]', Number.Float),
(r'\-?\d+#?', Number.Integer.Long),
(r'\-?\d+#?', Number.Integer),
(r'!=|==|:=|\.=|<<|>>|[-~+/\\*%=<>&^|?:!.]', Operator),
(r'[\[\]{}(),;]', Punctuation),
(r'[\w]+', Name.Variable.Global),
],
# can't use regular \b because of X$()
# XXX: use words() here
'declarations': [
(r'\b(%s)(?=\(|\b)' % '|'.join(map(re.escape, declarations)),
Keyword.Declaration),
],
'functions': [
(r'\b(%s)(?=\(|\b)' % '|'.join(map(re.escape, functions)),
Keyword.Reserved),
],
'metacommands': [
(r'\b(%s)(?=\(|\b)' % '|'.join(map(re.escape, metacommands)),
Keyword.Constant),
],
'operators': [
(r'\b(%s)(?=\(|\b)' % '|'.join(map(re.escape, operators)), Operator.Word),
],
'statements': [
(r'\b(%s)\b' % '|'.join(map(re.escape, statements)),
Keyword.Reserved),
],
'keywords': [
(r'\b(%s)\b' % '|'.join(keywords), Keyword),
],
}
def analyse_text(text):
if '$DYNAMIC' in text or '$STATIC' in text:
return 0.9
class VBScriptLexer(RegexLexer):
"""
VBScript is scripting language that is modeled on Visual Basic.
.. versionadded:: 2.4
"""
name = 'VBScript'
aliases = ['vbscript']
filenames = ['*.vbs', '*.VBS']
flags = re.IGNORECASE
tokens = {
'root': [
(r"'[^\n]*", Comment.Single),
(r'\s+', Whitespace),
('"', String.Double, 'string'),
('&h[0-9a-f]+', Number.Hex),
# Float variant 1, for example: 1., 1.e2, 1.2e3
(r'[0-9]+\.[0-9]*(e[+-]?[0-9]+)?', Number.Float),
(r'\.[0-9]+(e[+-]?[0-9]+)?', Number.Float), # Float variant 2, for example: .1, .1e2
(r'[0-9]+e[+-]?[0-9]+', Number.Float), # Float variant 3, for example: 123e45
(r'[0-9]+', Number.Integer),
('#.+#', String), # date or time value
(r'(dim)(\s+)([a-z_][a-z0-9_]*)',
bygroups(Keyword.Declaration, Whitespace, Name.Variable), 'dim_more'),
(r'(function|sub)(\s+)([a-z_][a-z0-9_]*)',
bygroups(Keyword.Declaration, Whitespace, Name.Function)),
(r'(class)(\s+)([a-z_][a-z0-9_]*)',
bygroups(Keyword.Declaration, Whitespace, Name.Class)),
(r'(const)(\s+)([a-z_][a-z0-9_]*)',
bygroups(Keyword.Declaration, Whitespace, Name.Constant)),
(r'(end)(\s+)(class|function|if|property|sub|with)',
bygroups(Keyword, Whitespace, Keyword)),
(r'(on)(\s+)(error)(\s+)(goto)(\s+)(0)',
bygroups(Keyword, Whitespace, Keyword, Whitespace, Keyword, Whitespace, Number.Integer)),
(r'(on)(\s+)(error)(\s+)(resume)(\s+)(next)',
bygroups(Keyword, Whitespace, Keyword, Whitespace, Keyword, Whitespace, Keyword)),
(r'(option)(\s+)(explicit)', bygroups(Keyword, Whitespace, Keyword)),
(r'(property)(\s+)(get|let|set)(\s+)([a-z_][a-z0-9_]*)',
bygroups(Keyword.Declaration, Whitespace, Keyword.Declaration, Whitespace, Name.Property)),
(r'rem\s.*[^\n]*', Comment.Single),
(words(_vbscript_builtins.KEYWORDS, suffix=r'\b'), Keyword),
(words(_vbscript_builtins.OPERATORS), Operator),
(words(_vbscript_builtins.OPERATOR_WORDS, suffix=r'\b'), Operator.Word),
(words(_vbscript_builtins.BUILTIN_CONSTANTS, suffix=r'\b'), Name.Constant),
(words(_vbscript_builtins.BUILTIN_FUNCTIONS, suffix=r'\b'), Name.Builtin),
(words(_vbscript_builtins.BUILTIN_VARIABLES, suffix=r'\b'), Name.Builtin),
(r'[a-z_][a-z0-9_]*', Name),
(r'\b_\n', Operator),
(words(r'(),.:'), Punctuation),
(r'.+(\n)?', Error)
],
'dim_more': [
(r'(\s*)(,)(\s*)([a-z_][a-z0-9]*)',
bygroups(Whitespace, Punctuation, Whitespace, Name.Variable)),
default('#pop'),
],
'string': [
(r'[^"\n]+', String.Double),
(r'\"\"', String.Double),
(r'"', String.Double, '#pop'),
(r'\n', Error, '#pop'), # Unterminated string
],
}
class BBCBasicLexer(RegexLexer):
"""
BBC Basic was supplied on the BBC Micro, and later Acorn RISC OS.
It is also used by BBC Basic For Windows.
.. versionadded:: 2.4
"""
base_keywords = ['OTHERWISE', 'AND', 'DIV', 'EOR', 'MOD', 'OR', 'ERROR',
'LINE', 'OFF', 'STEP', 'SPC', 'TAB', 'ELSE', 'THEN',
'OPENIN', 'PTR', 'PAGE', 'TIME', 'LOMEM', 'HIMEM', 'ABS',
'ACS', 'ADVAL', 'ASC', 'ASN', 'ATN', 'BGET', 'COS', 'COUNT',
'DEG', 'ERL', 'ERR', 'EVAL', 'EXP', 'EXT', 'FALSE', 'FN',
'GET', 'INKEY', 'INSTR', 'INT', 'LEN', 'LN', 'LOG', 'NOT',
'OPENUP', 'OPENOUT', 'PI', 'POINT', 'POS', 'RAD', 'RND',
'SGN', 'SIN', 'SQR', 'TAN', 'TO', 'TRUE', 'USR', 'VAL',
'VPOS', 'CHR$', 'GET$', 'INKEY$', 'LEFT$', 'MID$',
'RIGHT$', 'STR$', 'STRING$', 'EOF', 'PTR', 'PAGE', 'TIME',
'LOMEM', 'HIMEM', 'SOUND', 'BPUT', 'CALL', 'CHAIN', 'CLEAR',
'CLOSE', 'CLG', 'CLS', 'DATA', 'DEF', 'DIM', 'DRAW', 'END',
'ENDPROC', 'ENVELOPE', 'FOR', 'GOSUB', 'GOTO', 'GCOL', 'IF',
'INPUT', 'LET', 'LOCAL', 'MODE', 'MOVE', 'NEXT', 'ON',
'VDU', 'PLOT', 'PRINT', 'PROC', 'READ', 'REM', 'REPEAT',
'REPORT', 'RESTORE', 'RETURN', 'RUN', 'STOP', 'COLOUR',
'TRACE', 'UNTIL', 'WIDTH', 'OSCLI']
basic5_keywords = ['WHEN', 'OF', 'ENDCASE', 'ENDIF', 'ENDWHILE', 'CASE',
'CIRCLE', 'FILL', 'ORIGIN', 'POINT', 'RECTANGLE', 'SWAP',
'WHILE', 'WAIT', 'MOUSE', 'QUIT', 'SYS', 'INSTALL',
'LIBRARY', 'TINT', 'ELLIPSE', 'BEATS', 'TEMPO', 'VOICES',
'VOICE', 'STEREO', 'OVERLAY', 'APPEND', 'AUTO', 'CRUNCH',
'DELETE', 'EDIT', 'HELP', 'LIST', 'LOAD', 'LVAR', 'NEW',
'OLD', 'RENUMBER', 'SAVE', 'TEXTLOAD', 'TEXTSAVE',
'TWIN', 'TWINO', 'INSTALL', 'SUM', 'BEAT']
name = 'BBC Basic'
aliases = ['bbcbasic']
filenames = ['*.bbc']
tokens = {
'root': [
(r"[0-9]+", Name.Label),
(r"(\*)([^\n]*)",
bygroups(Keyword.Pseudo, Comment.Special)),
default('code'),
],
'code': [
(r"(REM)([^\n]*)",
bygroups(Keyword.Declaration, Comment.Single)),
(r'\n', Whitespace, 'root'),
(r'\s+', Whitespace),
(r':', Comment.Preproc),
# Some special cases to make functions come out nicer
(r'(DEF)(\s*)(FN|PROC)([A-Za-z_@][\w@]*)',
bygroups(Keyword.Declaration, Whitespace, Keyword.Declaration, Name.Function)),
(r'(FN|PROC)([A-Za-z_@][\w@]*)',
bygroups(Keyword, Name.Function)),
(r'(GOTO|GOSUB|THEN|RESTORE)(\s*)(\d+)',
bygroups(Keyword, Whitespace, Name.Label)),
(r'(TRUE|FALSE)', Keyword.Constant),
(r'(PAGE|LOMEM|HIMEM|TIME|WIDTH|ERL|ERR|REPORT\$|POS|VPOS|VOICES)', Keyword.Pseudo),
(words(base_keywords), Keyword),
(words(basic5_keywords), Keyword),
('"', String.Double, 'string'),
('%[01]{1,32}', Number.Bin),
('&[0-9a-f]{1,8}', Number.Hex),
(r'[+-]?[0-9]+\.[0-9]*(E[+-]?[0-9]+)?', Number.Float),
(r'[+-]?\.[0-9]+(E[+-]?[0-9]+)?', Number.Float),
(r'[+-]?[0-9]+E[+-]?[0-9]+', Number.Float),
(r'[+-]?\d+', Number.Integer),
(r'([A-Za-z_@][\w@]*[%$]?)', Name.Variable),
(r'([+\-]=|[$!|?+\-*/%^=><();]|>=|<=|<>|<<|>>|>>>|,)', Operator),
],
'string': [
(r'[^"\n]+', String.Double),
(r'"', String.Double, '#pop'),
(r'\n', Error, 'root'), # Unterminated string
],
}
def analyse_text(text):
if text.startswith('10REM >') or text.startswith('REM >'):
return 0.9