# Leo colorizer control file for omnimark mode.
# This file is in the public domain.
# Properties for omnimark mode.
properties = {
"indentNextLines": "\\s*((PROCESS|MARKUP|EXTERNAL|DOCUMENT|DTD|ELEMENT|FIND|TRANSLATE)((-|\\s).*|\\s*)|(DO|ELSE|REPEAT|MATCH|CASE|USING|GROUP|DEFINE|MACRO)(\\s+.*|\\s*))",
"lineComment": ";",
"noWordSep": ".-_",
}
# Attributes dict for omnimark_main ruleset.
omnimark_main_attributes_dict = {
"default": "null",
"digit_re": "",
"escape": "%",
"highlight_digits": "true",
"ignore_case": "true",
"no_word_sep": "",
}
# Dictionary of attributes dictionaries for omnimark mode.
attributesDictDict = {
"omnimark_main": omnimark_main_attributes_dict,
}
# Keywords dict for omnimark_main ruleset.
omnimark_main_keywords_dict = {
"#!": "keyword2",
"#additional-info": "keyword2",
"#appinfo": "keyword2",
"#args": "keyword2",
"#capacity": "keyword2",
"#charset": "keyword2",
"#class": "keyword2",
"#command-line-names": "keyword2",
"#console": "keyword2",
"#current-input": "keyword2",
"#current-output": "keyword2",
"#data": "keyword2",
"#doctype": "keyword2",
"#document": "keyword2",
"#dtd": "keyword2",
"#empty": "keyword2",
"#error": "keyword2",
"#error-code": "keyword2",
"#external-exception": "keyword2",
"#file-name": "keyword2",
"#first": "keyword2",
"#group": "keyword2",
"#implied": "keyword2",
"#item": "keyword2",
"#language-version": "keyword2",
"#last": "keyword2",
"#libpath": "keyword2",
"#library": "keyword2",
"#libvalue": "keyword2",
"#line-number": "keyword2",
"#main-input": "keyword2",
"#main-output": "keyword2",
"#markup-error-count": "keyword2",
"#markup-error-total": "keyword2",
"#markup-parser": "keyword2",
"#markup-warning-count": "keyword2",
"#markup-warning-total": "keyword2",
"#message": "keyword2",
"#none": "keyword2",
"#output": "keyword2",
"#platform-info": "keyword2",
"#process-input": "keyword2",
"#process-output": "keyword2",
"#program-error": "keyword2",
"#recovery-info": "keyword2",
"#sgml": "keyword2",
"#sgml-error-count": "keyword2",
"#sgml-error-total": "keyword2",
"#sgml-warning-count": "keyword2",
"#sgml-warning-total": "keyword2",
"#suppress": "keyword2",
"#syntax": "keyword2",
"abs": "operator",
"activate": "keyword1",
"active": "operator",
"after": "keyword2",
"again": "keyword1",
"always": "keyword1",
"ancestor": "keyword2",
"and": "operator",
"another": "keyword2",
"any": "keyword3",
"any-text": "keyword3",
"arg": "keyword2",
"as": "operator",
"assert": "keyword1",
"attached": "keyword2",
"attribute": "keyword2",
"attributes": "keyword2",
"base": "operator",
"bcd": "keyword2",
"before": "keyword2",
"binary": "operator",
"binary-input": "keyword2",
"binary-mode": "keyword2",
"binary-output": "keyword2",
"blank": "keyword3",
"break-width": "keyword2",
"buffer": "keyword2",
"buffered": "keyword2",
"by": "keyword2",
"case": "keyword1",
"catch": "keyword1",
"catchable": "keyword2",
"cdata": "keyword3",
"cdata-entity": "keyword2",
"ceiling": "operator",
"children": "keyword2",
"clear": "keyword1",
"close": "keyword1",
"closed": "keyword2",
"compiled-date": "operator",
"complement": "operator",
"conref": "keyword2",
"content": "keyword2",
"content-end": "keyword3",
"content-start": "keyword3",
"context-translate": "keyword2",
"copy": "keyword1",
"copy-clear": "keyword1",
"counter": "keyword2",
"created": "keyword2",
"creating": "operator",
"creator": "operator",
"cross-translate": "keyword2",
"current": "keyword2",
"data-attribute": "keyword2",
"data-attributes": "keyword2",
"data-content": "keyword2",
"data-letters": "keyword2",
"date": "operator",
"deactivate": "keyword1",
"declare": "keyword2",
"declared-conref": "keyword2",
"declared-current": "keyword2",
"declared-defaulted": "keyword2",
"declared-fixed": "keyword2",
"declared-implied": "keyword2",
"declared-required": "keyword2",
"decrement": "keyword1",
"default-entity": "keyword2",
"defaulted": "keyword2",
"defaulting": "keyword2",
"define": "keyword2",
"delimiter": "keyword2",
"difference": "operator",
"digit": "keyword3",
"directory": "keyword2",
"discard": "keyword1",
"divide": "operator",
"do": "keyword1",
"doctype": "keyword2",
"document": "keyword2",
"document-element": "keyword2",
"document-end": "keyword2",
"document-start": "keyword2",
"domain-free": "keyword2",
"done": "keyword1",
"down-translate": "keyword2",
"drop": "operator",
"dtd": "keyword2",
"dtd-end": "keyword2",
"dtd-start": "keyword2",
"dtds": "keyword2",
"element": "keyword2",
"elements": "keyword2",
"else": "keyword1",
"elsewhere": "keyword2",
"empty": "keyword2",
"entities": "keyword2",
"entity": "keyword2",
"epilog-start": "keyword2",
"equal": "operator",
"equals": "operator",
"escape": "keyword2",
"except": "keyword1",
"exists": "operator",
"exit": "keyword1",
"external": "keyword2",
"external-data-entity": "keyword2",
"external-entity": "keyword2",
"external-function": "keyword2",
"external-output-function": "keyword2",
"external-text-entity": "keyword2",
"false": "keyword2",
"file": "operator",
"find": "keyword2",
"find-end": "keyword2",
"find-start": "keyword2",
"floor": "operator",
"flush": "keyword1",
"for": "keyword1",
"format": "keyword1",
"function": "keyword2",
"function-library": "keyword2",
"general": "keyword2",
"global": "keyword2",
"greater-equal": "operator",
"greater-than": "operator",
"group": "keyword2",
"groups": "keyword2",
"halt": "keyword1",
"halt-everything": "keyword1",
"has": "operator",
"hasnt": "operator",
"heralded-names": "keyword2",
"id": "keyword2",
"id-checking": "keyword2",
"idref": "keyword2",
"idrefs": "keyword2",
"ignore": "keyword2",
"implied": "keyword2",
"in": "keyword2",
"in-library": "keyword2",
"include": "keyword2",
"include-end": "keyword2",
"include-guard": "keyword2",
"include-start": "keyword2",
"inclusion": "keyword2",
"increment": "keyword1",
"initial": "keyword2",
"initial-size": "keyword2",
"input": "keyword1",
"insertion-break": "keyword2",
"instance": "keyword2",
"integer": "keyword2",
"internal": "keyword2",
"invalid-data": "keyword2",
"is": "operator",
"isnt": "operator",
"item": "operator",
"join": "keyword1",
"key": "operator",
"keyed": "keyword2",
"last": "operator",
"lastmost": "operator",
"lc": "keyword3",
"length": "operator",
"less-equal": "operator",
"less-than": "operator",
"letter": "keyword3",
"letters": "keyword2",
"library": "keyword2",
"line-end": "keyword3",
"line-start": "keyword3",
"literal": "operator",
"ln": "operator",
"local": "keyword2",
"log": "keyword1",
"log10": "operator",
"lookahead": "operator",
"macro": "keyword2",
"macro-end": "keyword2",
"marked-section": "keyword2",
"markup-comment": "keyword2",
"markup-error": "keyword2",
"markup-parser": "keyword2",
"markup-wrapper": "keyword2",
"mask": "operator",
"match": "keyword1",
"matches": "operator",
"minus": "operator",
"mixed": "keyword2",
"modifiable": "keyword2",
"modulo": "operator",
"name": "operator",
"name-letters": "keyword2",
"namecase": "keyword2",
"named": "keyword2",
"names": "keyword2",
"ndata-entity": "keyword2",
"negate": "operator",
"nested-referents": "keyword2",
"new": "keyword1",
"newline": "keyword2",
"next": "keyword1",
"nmtoken": "keyword2",
"nmtokens": "keyword2",
"no": "keyword2",
"no-default-io": "keyword2",
"non-cdata": "keyword3",
"non-implied": "keyword2",
"non-sdata": "keyword3",
"not": "operator",
"not-reached": "keyword1",
"notation": "keyword2",
"null": "keyword3",
"number": "keyword2",
"number-of": "operator",
"numbers": "keyword2",
"nutoken": "keyword2",
"nutokens": "keyword2",
"occurrence": "operator",
"of": "operator",
"opaque": "keyword2",
"open": "keyword1",
"optional": "keyword2",
"or": "operator",
"output": "keyword1",
"output-to": "keyword1",
"over": "keyword1",
"parameter": "keyword2",
"parent": "keyword2",
"past": "keyword2",
"pattern": "keyword2",
"pcdata": "keyword3",
"plus": "keyword2",
"preparent": "keyword2",
"previous": "keyword2",
"process": "keyword2",
"process-end": "keyword2",
"process-start": "keyword2",
"processing-instruction": "keyword2",
"prolog-end": "keyword2",
"prolog-in-error": "keyword2",
"proper": "keyword2",
"public": "keyword2",
"put": "keyword1",
"rcdata": "keyword3",
"read-only": "keyword2",
"readable": "keyword2",
"referent": "keyword2",
"referents": "keyword2",
"referents-allowed": "keyword2",
"referents-displayed": "keyword2",
"referents-not-allowed": "keyword2",
"remainder": "keyword2",
"remove": "keyword1",
"reopen": "keyword1",
"repeat": "keyword1",
"repeated": "keyword2",
"replacement-break": "keyword2",
"reset": "keyword1",
"rethrow": "keyword1",
"return": "keyword1",
"reversed": "keyword2",
"round": "operator",
"save": "keyword1",
"save-clear": "keyword1",
"scan": "keyword1",
"sdata": "keyword3",
"sdata-entity": "keyword2",
"select": "keyword1",
"set": "keyword1",
"sgml": "keyword1",
"sgml-comment": "keyword2",
"sgml-declaration-end": "keyword2",
"sgml-dtd": "keyword2",
"sgml-dtds": "keyword2",
"sgml-error": "keyword2",
"sgml-in": "keyword1",
"sgml-out": "keyword1",
"sgml-parse": "keyword1",
"sgml-parser": "keyword1",
"shift": "operator",
"silent-referent": "keyword2",
"size": "keyword2",
"skip": "keyword1",
"source": "keyword2",
"space": "keyword3",
"specified": "keyword2",
"sqrt": "operator",
"status": "operator",
"stream": "keyword2",
"subdoc-entity": "keyword2",
"subdocument": "keyword2",
"subdocuments": "keyword2",
"subelement": "keyword2",
"submit": "keyword1",
"succeed": "keyword1",
"suppress": "keyword1",
"switch": "keyword2",
"symbol": "keyword2",
"system": "keyword2",
"system-call": "keyword1",
"take": "operator",
"test-system": "keyword1",
"text": "keyword3",
"text-mode": "keyword2",
"this": "operator",
"throw": "keyword1",
"thrown": "keyword2",
"times": "keyword2",
"to": "keyword1",
"token": "keyword2",
"translate": "keyword2",
"true": "keyword2",
"truncate": "operator",
"uc": "keyword3",
"ul": "operator",
"unanchored": "operator",
"unattached": "keyword2",
"unbuffered": "keyword2",
"union": "operator",
"unless": "keyword1",
"up-translate": "keyword2",
"usemap": "operator",
"using": "keyword1",
"value": "keyword2",
"value-end": "keyword3",
"value-start": "keyword3",
"valued": "keyword2",
"variable": "keyword2",
"when": "keyword1",
"white-space": "keyword3",
"with": "operator",
"word-end": "keyword3",
"word-start": "keyword3",
"writable": "keyword2",
"xml": "keyword2",
"xml-dtd": "keyword2",
"xml-dtds": "keyword2",
"xml-parse": "keyword1",
"yes": "keyword2",
}
# Dictionary of keywords dictionaries for omnimark mode.
keywordsDictDict = {
"omnimark_main": omnimark_main_keywords_dict,
}
# Rules for omnimark_main ruleset.
def omnimark_rule0(colorer, s, i):
return colorer.match_eol_span(s, i, kind="comment1", seq="#!",
at_line_start=False, at_whitespace_end=False, at_word_start=False,
delegate="", exclude_match=False)
def omnimark_rule1(colorer, s, i):
return colorer.match_eol_span(s, i, kind="comment1", seq=";",
at_line_start=False, at_whitespace_end=False, at_word_start=False,
delegate="", exclude_match=False)
def omnimark_rule2(colorer, s, i):
return colorer.match_span_regexp(s, i, kind="invalid", begin="\"((?!$)[^\"])*$", end="$",
at_line_start=False, at_whitespace_end=False, at_word_start=False,
delegate="",exclude_match=False,
no_escape=False, no_line_break=True, no_word_break=False)
def omnimark_rule3(colorer, s, i):
return colorer.match_span(s, i, kind="literal1", begin="\"", end="\"",
at_line_start=False, at_whitespace_end=False, at_word_start=False,
delegate="",exclude_match=False,
no_escape=False, no_line_break=True, no_word_break=False)
def omnimark_rule4(colorer, s, i):
return colorer.match_span_regexp(s, i, kind="invalid", begin="'((?!$)[^'])*$", end="$",
at_line_start=False, at_whitespace_end=False, at_word_start=False,
delegate="",exclude_match=False,
no_escape=False, no_line_break=True, no_word_break=False)
def omnimark_rule5(colorer, s, i):
return colorer.match_span(s, i, kind="literal1", begin="'", end="'",
at_line_start=False, at_whitespace_end=False, at_word_start=False,
delegate="",exclude_match=False,
no_escape=False, no_line_break=True, no_word_break=False)
def omnimark_rule6(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="&",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule7(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="|",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule8(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="+",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule9(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="=",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule10(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="/",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule11(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="<",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule12(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq=">",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule13(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="~",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule14(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="@",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule15(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="$",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule16(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="%",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule17(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="^",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule18(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="*",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule19(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="?",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule20(colorer, s, i):
return colorer.match_seq(s, i, kind="operator", seq="!",
at_line_start=False, at_whitespace_end=False, at_word_start=False, delegate="")
def omnimark_rule21(colorer, s, i):
return colorer.match_keywords(s, i)
# Rules dict for omnimark_main ruleset.
rulesDict1 = {
"!": [omnimark_rule20,omnimark_rule21,],
"\"": [omnimark_rule2,omnimark_rule3,],
"#": [omnimark_rule0,omnimark_rule21,],
"$": [omnimark_rule15,],
"%": [omnimark_rule16,],
"&": [omnimark_rule6,],
"'": [omnimark_rule4,omnimark_rule5,],
"*": [omnimark_rule18,],
"+": [omnimark_rule8,],
"-": [omnimark_rule21,],
"/": [omnimark_rule10,],
"0": [omnimark_rule21,],
"1": [omnimark_rule21,],
"2": [omnimark_rule21,],
"3": [omnimark_rule21,],
"4": [omnimark_rule21,],
"5": [omnimark_rule21,],
"6": [omnimark_rule21,],
"7": [omnimark_rule21,],
"8": [omnimark_rule21,],
"9": [omnimark_rule21,],
";": [omnimark_rule1,],
"<": [omnimark_rule11,],
"=": [omnimark_rule9,],
">": [omnimark_rule12,],
"?": [omnimark_rule19,],
"@": [omnimark_rule14,omnimark_rule21,],
"A": [omnimark_rule21,],
"B": [omnimark_rule21,],
"C": [omnimark_rule21,],
"D": [omnimark_rule21,],
"E": [omnimark_rule21,],
"F": [omnimark_rule21,],
"G": [omnimark_rule21,],
"H": [omnimark_rule21,],
"I": [omnimark_rule21,],
"J": [omnimark_rule21,],
"K": [omnimark_rule21,],
"L": [omnimark_rule21,],
"M": [omnimark_rule21,],
"N": [omnimark_rule21,],
"O": [omnimark_rule21,],
"P": [omnimark_rule21,],
"Q": [omnimark_rule21,],
"R": [omnimark_rule21,],
"S": [omnimark_rule21,],
"T": [omnimark_rule21,],
"U": [omnimark_rule21,],
"V": [omnimark_rule21,],
"W": [omnimark_rule21,],
"X": [omnimark_rule21,],
"Y": [omnimark_rule21,],
"Z": [omnimark_rule21,],
"^": [omnimark_rule17,],
"a": [omnimark_rule21,],
"b": [omnimark_rule21,],
"c": [omnimark_rule21,],
"d": [omnimark_rule21,],
"e": [omnimark_rule21,],
"f": [omnimark_rule21,],
"g": [omnimark_rule21,],
"h": [omnimark_rule21,],
"i": [omnimark_rule21,],
"j": [omnimark_rule21,],
"k": [omnimark_rule21,],
"l": [omnimark_rule21,],
"m": [omnimark_rule21,],
"n": [omnimark_rule21,],
"o": [omnimark_rule21,],
"p": [omnimark_rule21,],
"q": [omnimark_rule21,],
"r": [omnimark_rule21,],
"s": [omnimark_rule21,],
"t": [omnimark_rule21,],
"u": [omnimark_rule21,],
"v": [omnimark_rule21,],
"w": [omnimark_rule21,],
"x": [omnimark_rule21,],
"y": [omnimark_rule21,],
"z": [omnimark_rule21,],
"|": [omnimark_rule7,],
"~": [omnimark_rule13,],
}
# x.rulesDictDict for omnimark mode.
rulesDictDict = {
"omnimark_main": rulesDict1,
}
# Import dict for omnimark mode.
importDict = {}
|