# Grammalecte
# Grammar checker engine
import re
import sys
import os
import traceback
#import unicodedata
from itertools import chain
from ..graphspell.spellchecker import SpellChecker
from ..graphspell.echo import echo
from . import gc_options
from ..graphspell.tokenizer import Tokenizer
from .gc_rules_graph import dGraph, dRule
__all__ = [ "lang", "locales", "pkg", "name", "version", "author", \
"load", "parse", "getSpellChecker", \
"setOption", "setOptions", "getOptions", "getDefaultOptions", "getOptionsLabels", "resetOptions", "displayOptions", \
"ignoreRule", "resetIgnoreRules", "reactivateRule", "listRules", "displayRules" ]
__version__ = "${version}"
lang = "${lang}"
locales = ${loc}
pkg = "${implname}"
name = "${name}"
version = "${version}"
author = "${author}"
_rules = None # module gc_rules
# data
_sAppContext = "" # what software is running
_dOptions = None
_aIgnoredRules = set()
_oSpellChecker = None
_oTokenizer = None
#### Parsing
def parse (sText, sCountry="${country_default}", bDebug=False, dOptions=None, bContext=False):
"analyses the paragraph sText and returns list of errors"
#sText = unicodedata.normalize("NFC", sText)
aErrors = None
sRealText = sText
dDA = {} # Disambiguisator. Key = position; value = list of morphologies
dPriority = {} # Key = position; value = priority
dOpt = _dOptions if not dOptions else dOptions
bShowRuleId = option('idrule')
# parse paragraph
try:
sNew, aErrors = _proofread(sText, sRealText, 0, True, dDA, dPriority, sCountry, dOpt, bShowRuleId, bDebug, bContext)
if sNew:
sText = sNew
except:
raise
# cleanup
if " " in sText:
sText = sText.replace(" ", ' ') # nbsp
if " " in sText:
sText = sText.replace(" ", ' ') # nnbsp
if "'" in sText:
sText = sText.replace("'", "’")
if "‑" in sText:
sText = sText.replace("‑", "-") # nobreakdash
# parse sentences
for iStart, iEnd in _getSentenceBoundaries(sText):
if 4 < (iEnd - iStart) < 2000:
dDA.clear()
try:
# regex parser
_, errs = _proofread(sText[iStart:iEnd], sRealText[iStart:iEnd], iStart, False, dDA, dPriority, sCountry, dOpt, bShowRuleId, bDebug, bContext)
aErrors.update(errs)
# token parser
oSentence = TokenSentence(sText[iStart:iEnd], sRealText[iStart:iEnd], iStart)
_, errs = oSentence.parse(dPriority, sCountry, dOpt, bShowRuleId, bDebug, bContext)
aErrors.update(errs)
except:
raise
return aErrors.values() # this is a view (iterable)
def _getSentenceBoundaries (sText):
iStart = _zBeginOfParagraph.match(sText).end()
for m in _zEndOfSentence.finditer(sText):
yield (iStart, m.end())
iStart = m.end()
def _proofread (s, sx, nOffset, bParagraph, dDA, dPriority, sCountry, dOptions, bShowRuleId, bDebug, bContext):
dErrs = {}
bChange = False
for sOption, lRuleGroup in _getRules(bParagraph):
if not sOption or dOptions.get(sOption, False):
for zRegex, bUppercase, sLineId, sRuleId, nPriority, lActions in lRuleGroup:
if sRuleId not in _aIgnoredRules:
for m in zRegex.finditer(s):
bCondMemo = None
for sFuncCond, cActionType, sWhat, *eAct in lActions:
# action in lActions: [ condition, action type, replacement/suggestion/action[, iGroup[, message, URL]] ]
try:
bCondMemo = not sFuncCond or globals()[sFuncCond](s, sx, m, dDA, sCountry, bCondMemo)
if bCondMemo:
if cActionType == "-":
# grammar error
nErrorStart = nOffset + m.start(eAct[0])
if nErrorStart not in dErrs or nPriority > dPriority[nErrorStart]:
dErrs[nErrorStart] = _createRegexError(s, sx, sWhat, nOffset, m, eAct[0], sLineId, sRuleId, bUppercase, eAct[1], eAct[2], bShowRuleId, sOption, bContext)
dPriority[nErrorStart] = nPriority
elif cActionType == "~":
# text processor
s = _rewrite(s, sWhat, eAct[0], m, bUppercase)
bChange = True
if bDebug:
echo("~ " + s + " -- " + m.group(eAct[0]) + " # " + sLineId)
elif cActionType == "=":
# disambiguation
globals()[sWhat](s, m, dDA)
if bDebug:
echo("= " + m.group(0) + " # " + sLineId + "\nDA: " + str(dDA))
elif cActionType == ">":
# we do nothing, this test is just a condition to apply all following actions
pass
else:
echo("# error: unknown action at " + sLineId)
elif cActionType == ">":
break
except Exception as e:
raise Exception(str(e), "# " + sLineId + " # " + sRuleId)
if bChange:
return (s, dErrs)
return (False, dErrs)
def _createRegexWriterError (s, sx, sRepl, nOffset, m, iGroup, sLineId, sRuleId, bUppercase, sMsg, sURL, bShowRuleId, sOption, bContext):
"error for Writer (LO/OO)"
xErr = SingleProofreadingError()
#xErr = uno.createUnoStruct( "com.sun.star.linguistic2.SingleProofreadingError" )
xErr.nErrorStart = nOffset + m.start(iGroup)
xErr.nErrorLength = m.end(iGroup) - m.start(iGroup)
xErr.nErrorType = PROOFREADING
xErr.aRuleIdentifier = sRuleId
# suggestions
if sRepl[0:1] == "=":
sugg = globals()[sRepl[1:]](s, m)
if sugg:
if bUppercase and m.group(iGroup)[0:1].isupper():
xErr.aSuggestions = tuple(map(str.capitalize, sugg.split("|")))
else:
xErr.aSuggestions = tuple(sugg.split("|"))
else:
xErr.aSuggestions = ()
elif sRepl == "_":
xErr.aSuggestions = ()
else:
if bUppercase and m.group(iGroup)[0:1].isupper():
xErr.aSuggestions = tuple(map(str.capitalize, m.expand(sRepl).split("|")))
else:
xErr.aSuggestions = tuple(m.expand(sRepl).split("|"))
# Message
if sMsg[0:1] == "=":
sMessage = globals()[sMsg[1:]](s, m)
else:
sMessage = m.expand(sMsg)
xErr.aShortComment = sMessage # sMessage.split("|")[0] # in context menu
xErr.aFullComment = sMessage # sMessage.split("|")[-1] # in dialog
if bShowRuleId:
xErr.aShortComment += " # " + sLineId + " # " + sRuleId
# URL
if sURL:
p = PropertyValue()
p.Name = "FullCommentURL"
p.Value = sURL
xErr.aProperties = (p,)
else:
xErr.aProperties = ()
return xErr
def _createRegexDictError (s, sx, sRepl, nOffset, m, iGroup, sLineId, sRuleId, bUppercase, sMsg, sURL, bShowRuleId, sOption, bContext):
"error as a dictionary"
dErr = {}
dErr["nStart"] = nOffset + m.start(iGroup)
dErr["nEnd"] = nOffset + m.end(iGroup)
dErr["sLineId"] = sLineId
dErr["sRuleId"] = sRuleId
dErr["sType"] = sOption if sOption else "notype"
# suggestions
if sRepl[0:1] == "=":
sugg = globals()[sRepl[1:]](s, m)
if sugg:
if bUppercase and m.group(iGroup)[0:1].isupper():
dErr["aSuggestions"] = list(map(str.capitalize, sugg.split("|")))
else:
dErr["aSuggestions"] = sugg.split("|")
else:
dErr["aSuggestions"] = []
elif sRepl == "_":
dErr["aSuggestions"] = []
else:
if bUppercase and m.group(iGroup)[0:1].isupper():
dErr["aSuggestions"] = list(map(str.capitalize, m.expand(sRepl).split("|")))
else:
dErr["aSuggestions"] = m.expand(sRepl).split("|")
# Message
if sMsg[0:1] == "=":
sMessage = globals()[sMsg[1:]](s, m)
else:
sMessage = m.expand(sMsg)
dErr["sMessage"] = sMessage
if bShowRuleId:
dErr["sMessage"] += " # " + sLineId + " # " + sRuleId
# URL
dErr["URL"] = sURL if sURL else ""
# Context
if bContext:
dErr['sUnderlined'] = sx[m.start(iGroup):m.end(iGroup)]
dErr['sBefore'] = sx[max(0,m.start(iGroup)-80):m.start(iGroup)]
dErr['sAfter'] = sx[m.end(iGroup):m.end(iGroup)+80]
return dErr
def _createTokenWriterError (lToken, sSentence, sSentence0, sRepl, iFirstToken, nStart, nEnd, sLineId, sRuleId, bUppercase, sMsg, sURL, bShowRuleId, sOption, bContext):
"error for Writer (LO/OO)"
xErr = SingleProofreadingError()
#xErr = uno.createUnoStruct( "com.sun.star.linguistic2.SingleProofreadingError" )
xErr.nErrorStart = nStart
xErr.nErrorLength = nEnd - nStart
xErr.nErrorType = PROOFREADING
xErr.aRuleIdentifier = sRuleId
# suggestions
if sRepl[0:1] == "=":
sSugg = globals()[sRepl[1:]](lToken)
if sSugg:
if bUppercase and lToken[iFirstToken]["sValue"][0:1].isupper():
xErr.aSuggestions = tuple(map(str.capitalize, sSugg.split("|")))
else:
xErr.aSuggestions = tuple(sSugg.split("|"))
else:
xErr.aSuggestions = ()
elif sRepl == "_":
xErr.aSuggestions = ()
else:
if bUppercase and lToken[iFirstToken]["sValue"][0:1].isupper():
xErr.aSuggestions = tuple(map(str.capitalize, sRepl.split("|")))
else:
xErr.aSuggestions = tuple(sRepl.split("|"))
# Message
if sMsg[0:1] == "=":
sMessage = globals()[sMsg[1:]](lToken)
else:
sMessage = sMsg
xErr.aShortComment = sMessage # sMessage.split("|")[0] # in context menu
xErr.aFullComment = sMessage # sMessage.split("|")[-1] # in dialog
if bShowRuleId:
xErr.aShortComment += " " + sLineId + " # " + sRuleId
# URL
if sURL:
p = PropertyValue()
p.Name = "FullCommentURL"
p.Value = sURL
xErr.aProperties = (p,)
else:
xErr.aProperties = ()
return xErr
def _createTokenDictError (lToken, sSentence, sSentence0, sRepl, iFirstToken, nStart, nEnd, sLineId, sRuleId, bUppercase, sMsg, sURL, bShowRuleId, sOption, bContext):
"error as a dictionary"
dErr = {}
dErr["nStart"] = nStart
dErr["nEnd"] = nEnd
dErr["sLineId"] = sLineId
dErr["sRuleId"] = sRuleId
dErr["sType"] = sOption if sOption else "notype"
# suggestions
if sRepl[0:1] == "=":
sugg = globals()[sRepl[1:]](lToken)
if sugg:
if bUppercase and lToken[iFirstToken]["sValue"][0:1].isupper():
dErr["aSuggestions"] = list(map(str.capitalize, sugg.split("|")))
else:
dErr["aSuggestions"] = sugg.split("|")
else:
dErr["aSuggestions"] = []
elif sRepl == "_":
dErr["aSuggestions"] = []
else:
if bUppercase and lToken[iFirstToken]["sValue"][0:1].isupper():
dErr["aSuggestions"] = list(map(str.capitalize, sRepl.split("|")))
else:
dErr["aSuggestions"] = sRepl.split("|")
# Message
if sMsg[0:1] == "=":
sMessage = globals()[sMsg[1:]](lToken)
else:
sMessage = sMsg
dErr["sMessage"] = sMessage
if bShowRuleId:
dErr["sMessage"] += " " + sLineId + " # " + sRuleId
# URL
dErr["URL"] = sURL if sURL else ""
# Context
if bContext:
dErr['sUnderlined'] = sSentence0[dErr["nStart"]:dErr["nEnd"]]
dErr['sBefore'] = sSentence0[max(0,dErr["nStart"]-80):dErr["nStart"]]
dErr['sAfter'] = sSentence0[dErr["nEnd"]:dErr["nEnd"]+80]
return dErr
def _rewrite (sSentence, sRepl, iGroup, m, bUppercase):
"text processor: write <sRepl> in <sSentence> at <iGroup> position"
nLen = m.end(iGroup) - m.start(iGroup)
if sRepl == "*":
sNew = " " * nLen
elif sRepl == "_":
sNew = sRepl + " " * (nLen-1)
elif sRepl[0:1] == "=":
sNew = globals()[sRepl[1:]](sSentence, m)
sNew = sNew + " " * (nLen-len(sNew))
if bUppercase and m.group(iGroup)[0:1].isupper():
sNew = sNew.capitalize()
else:
sNew = m.expand(sRepl)
sNew = sNew + " " * (nLen-len(sNew))
return sSentence[0:m.start(iGroup)] + sNew + sSentence[m.end(iGroup):]
def ignoreRule (sRuleId):
_aIgnoredRules.add(sRuleId)
def resetIgnoreRules ():
_aIgnoredRules.clear()
def reactivateRule (sRuleId):
_aIgnoredRules.discard(sRuleId)
def listRules (sFilter=None):
"generator: returns typle (sOption, sLineId, sRuleId)"
if sFilter:
try:
zFilter = re.compile(sFilter)
except:
echo("# Error. List rules: wrong regex.")
sFilter = None
for sOption, lRuleGroup in chain(_getRules(True), _getRules(False)):
for _, _, sLineId, sRuleId, _, _ in lRuleGroup:
if not sFilter or zFilter.search(sRuleId):
yield (sOption, sLineId, sRuleId)
def displayRules (sFilter=None):
echo("List of rules. Filter: << " + str(sFilter) + " >>")
for sOption, sLineId, sRuleId in listRules(sFilter):
echo("{:<10} {:<10} {}".format(sOption, sLineId, sRuleId))
#### init
try:
# LibreOffice / OpenOffice
from com.sun.star.linguistic2 import SingleProofreadingError
from com.sun.star.text.TextMarkupType import PROOFREADING
from com.sun.star.beans import PropertyValue
#import lightproof_handler_${implname} as opt
_createRegexError = _createRegexWriterError
_createTokenError = _createTokenWriterError
except ImportError:
_createRegexError = _createRegexDictError
_createTokenError = _createTokenDictError
def load (sContext="Python"):
global _oSpellChecker
global _sAppContext
global _dOptions
global _oTokenizer
try:
_oSpellChecker = SpellChecker("${lang}", "${dic_main_filename_py}", "${dic_extended_filename_py}", "${dic_community_filename_py}", "${dic_personal_filename_py}")
_sAppContext = sContext
_dOptions = dict(gc_options.getOptions(sContext)) # duplication necessary, to be able to reset to default
_oTokenizer = _oSpellChecker.getTokenizer()
_oSpellChecker.activateStorage()
except:
traceback.print_exc()
def setOption (sOpt, bVal):
if sOpt in _dOptions:
_dOptions[sOpt] = bVal
def setOptions (dOpt):
for sKey, bVal in dOpt.items():
if sKey in _dOptions:
_dOptions[sKey] = bVal
def getOptions ():
return _dOptions
def getDefaultOptions ():
return dict(gc_options.getOptions(_sAppContext))
def getOptionsLabels (sLang):
return gc_options.getUI(sLang)
def displayOptions (sLang):
echo("List of options")
echo("\n".join( [ k+":\t"+str(v)+"\t"+gc_options.getUI(sLang).get(k, ("?", ""))[0] for k, v in sorted(_dOptions.items()) ] ))
echo("")
def resetOptions ():
global _dOptions
_dOptions = dict(gc_options.getOptions(_sAppContext))
def getSpellChecker ():
return _oSpellChecker
def _getRules (bParagraph):
try:
if not bParagraph:
return _rules.lSentenceRules
return _rules.lParagraphRules
except:
_loadRules()
if not bParagraph:
return _rules.lSentenceRules
return _rules.lParagraphRules
def _loadRules ():
from . import gc_rules
global _rules
_rules = gc_rules
# compile rules regex
for lRuleGroup in chain(_rules.lParagraphRules, _rules.lSentenceRules):
for rule in lRuleGroup[1]:
try:
rule[0] = re.compile(rule[0])
except:
echo("Bad regular expression in # " + str(rule[2]))
rule[0] = "(?i)<Grammalecte>"
def _getPath ():
return os.path.join(os.path.dirname(sys.modules[__name__].__file__), __name__ + ".py")
#### common functions
# common regexes
_zEndOfSentence = re.compile(r'([.?!:;…][ .?!… »”")]*|.$)')
_zBeginOfParagraph = re.compile(r"^\W*")
_zEndOfParagraph = re.compile(r"\W*$")
_zNextWord = re.compile(r" +(\w[\w-]*)")
_zPrevWord = re.compile(r"(\w[\w-]*) +$")
def option (sOpt):
"return True if option sOpt is active"
return _dOptions.get(sOpt, False)
def displayInfo (dDA, tWord):
"for debugging: retrieve info of word"
if not tWord:
echo("> nothing to find")
return True
lMorph = _oSpellChecker.getMorph(tWord[1])
if not lMorph:
echo("> not in dictionary")
return True
if tWord[0] in dDA:
echo("DA: " + str(dDA[tWord[0]]))
echo("FSA: " + str(lMorph))
return True
def morph (dDA, tWord, sPattern, bStrict=True, bNoWord=False):
"analyse a tuple (position, word), return True if sPattern in morphologies (disambiguation on)"
if not tWord:
return bNoWord
lMorph = dDA[tWord[0]] if tWord[0] in dDA else _oSpellChecker.getMorph(tWord[1])
if not lMorph:
return False
p = re.compile(sPattern)
if bStrict:
return all(p.search(s) for s in lMorph)
return any(p.search(s) for s in lMorph)
def morphex (dDA, tWord, sPattern, sNegPattern, bNoWord=False):
"analyse a tuple (position, word), returns True if not sNegPattern in word morphologies and sPattern in word morphologies (disambiguation on)"
if not tWord:
return bNoWord
lMorph = dDA[tWord[0]] if tWord[0] in dDA else _oSpellChecker.getMorph(tWord[1])
if not lMorph:
return False
# check negative condition
np = re.compile(sNegPattern)
if any(np.search(s) for s in lMorph):
return False
# search sPattern
p = re.compile(sPattern)
return any(p.search(s) for s in lMorph)
def analyse (sWord, sPattern, bStrict=True):
"analyse a word, return True if sPattern in morphologies (disambiguation off)"
lMorph = _oSpellChecker.getMorph(sWord)
if not lMorph:
return False
p = re.compile(sPattern)
if bStrict:
return all(p.search(s) for s in lMorph)
return any(p.search(s) for s in lMorph)
def analysex (sWord, sPattern, sNegPattern):
"analyse a word, returns True if not sNegPattern in word morphologies and sPattern in word morphologies (disambiguation off)"
lMorph = _oSpellChecker.getMorph(sWord)
if not lMorph:
return False
# check negative condition
np = re.compile(sNegPattern)
if any(np.search(s) for s in lMorph):
return False
# search sPattern
p = re.compile(sPattern)
return any(p.search(s) for s in lMorph)
## functions to get text outside pattern scope
# warning: check compile_rules.py to understand how it works
def nextword (s, iStart, n):
"get the nth word of the input string or empty string"
m = re.match("(?: +[\\w%-]+){" + str(n-1) + "} +([\\w%-]+)", s[iStart:])
if not m:
return None
return (iStart+m.start(1), m.group(1))
def prevword (s, iEnd, n):
"get the (-)nth word of the input string or empty string"
m = re.search("([\\w%-]+) +(?:[\\w%-]+ +){" + str(n-1) + "}$", s[:iEnd])
if not m:
return None
return (m.start(1), m.group(1))
def nextword1 (s, iStart):
"get next word (optimization)"
m = _zNextWord.match(s[iStart:])
if not m:
return None
return (iStart+m.start(1), m.group(1))
def prevword1 (s, iEnd):
"get previous word (optimization)"
m = _zPrevWord.search(s[:iEnd])
if not m:
return None
return (m.start(1), m.group(1))
def look (s, sPattern, sNegPattern=None):
"seek sPattern in s (before/after/fulltext), if sNegPattern not in s"
if sNegPattern and re.search(sNegPattern, s):
return False
if re.search(sPattern, s):
return True
return False
def look_chk1 (dDA, s, nOffset, sPattern, sPatternGroup1, sNegPatternGroup1=None):
"returns True if s has pattern sPattern and m.group(1) has pattern sPatternGroup1"
m = re.search(sPattern, s)
if not m:
return False
try:
sWord = m.group(1)
nPos = m.start(1) + nOffset
except:
return False
if sNegPatternGroup1:
return morphex(dDA, (nPos, sWord), sPatternGroup1, sNegPatternGroup1)
return morph(dDA, (nPos, sWord), sPatternGroup1, False)
#### Disambiguator
def select (dDA, nPos, sWord, sPattern, lDefault=None):
if not sWord:
return True
if nPos in dDA:
return True
lMorph = _oSpellChecker.getMorph(sWord)
if not lMorph or len(lMorph) == 1:
return True
lSelect = [ sMorph for sMorph in lMorph if re.search(sPattern, sMorph) ]
if lSelect:
if len(lSelect) != len(lMorph):
dDA[nPos] = lSelect
elif lDefault:
dDA[nPos] = lDefault
return True
def exclude (dDA, nPos, sWord, sPattern, lDefault=None):
if not sWord:
return True
if nPos in dDA:
return True
lMorph = _oSpellChecker.getMorph(sWord)
if not lMorph or len(lMorph) == 1:
return True
lSelect = [ sMorph for sMorph in lMorph if not re.search(sPattern, sMorph) ]
if lSelect:
if len(lSelect) != len(lMorph):
dDA[nPos] = lSelect
elif lDefault:
dDA[nPos] = lDefault
return True
def define (dDA, nPos, lMorph):
dDA[nPos] = lMorph
return True
#### GRAMMAR CHECKER PLUGINS
${plugins}
#### CALLABLES (generated code)
${callables}
#### TOKEN SENTENCE CHECKER
class TokenSentence:
def __init__ (self, sSentence, sSentence0, iStart):
self.sSentence = sSentence
self.sSentence0 = sSentence0
self.iStart = iStart
self.lToken = list(_oTokenizer.genTokens(sSentence, True))
def _getNextMatchingNodes (self, dToken, dNode):
"generator: return nodes where <dToken> “values” match <dNode> arcs"
# token value
if dToken["sValue"] in dNode:
#print("value found: ", dToken["sValue"])
yield dGraph[dNode[dToken["sValue"]]]
# token lemmas
if "<lemmas>" in dNode:
for sLemma in _oSpellChecker.getLemma(dToken["sValue"]):
if sLemma in dNode["<lemmas>"]:
#print("lemma found: ", sLemma)
yield dGraph[dNode["<lemmas>"][sLemma]]
# universal arc
if "*" in dNode:
#print("generic arc")
yield dGraph[dNode["*"]]
# regex value arcs
if "<re_value>" in dNode:
for sRegex in dNode["<re_value>"]:
if re.search(sRegex, dToken["sValue"]):
#print("value regex matching: ", sRegex)
yield dGraph[dNode["<re_value>"][sRegex]]
# regex morph arcs
if "<re_morph>" in dNode:
for sRegex in dNode["<re_morph>"]:
if "¬" not in sRegex:
# no anti-pattern
if any(re.search(sRegex, sMorph) for sMorph in _oSpellChecker.getMorph(dToken["sValue"])):
yield dGraph[dNode["<re_morph>"][sRegex]]
else:
# there is an anti-pattern
sPattern, sNegPattern = sRegex.split("¬", 1)
if sNegPattern == "*":
# all morphologies must match with <sPattern>
if all(re.search(sPattern, sMorph) for sMorph in _oSpellChecker.getMorph(dToken["sValue"])):
yield dGraph[dNode["<re_morph>"][sRegex]]
else:
if sNegPattern and any(re.search(sNegPattern, sMorph) for sMorph in _oSpellChecker.getMorph(dToken["sValue"])):
continue
if any(re.search(sPattern, sMorph) for sMorph in _oSpellChecker.getMorph(dToken["sValue"])):
yield dGraph[dNode["<re_morph>"][sRegex]]
def parse (self, dPriority, sCountry="${country_default}", dOptions=None, bShowRuleId=False, bDebug=False, bContext=False):
dErr = {}
dPriority = {} # Key = position; value = priority
dOpt = _dOptions if not dOptions else dOptions
lPointer = []
bChange = False
for dToken in self.lToken:
# check arcs for each existing pointer
lNewPointer = []
for i, dPointer in enumerate(lPointer):
bValid = False
bFirst = True
for dNode in self._getNextMatchingNodes(dToken, dPointer["dNode"]):
if bFirst:
dPointer["dNode"] = dNode
else:
lNewPointer.append({"nOffset": dPointer["nOffset"], "dNode": dNode})
bFirst = False
bValid = True
if not bValid:
del lPointer[i]
lPointer.extend(lNewPointer)
# check arcs of first nodes
for dNode in self._getNextMatchingNodes(dToken, dGraph[0]):
lPointer.append({"nOffset": dToken["i"], "dNode": dNode})
# check if there is rules to check for each pointer
for dPointer in lPointer:
if "<rules>" in dPointer["dNode"]:
bHasChanged, errs = self._executeActions(dPointer["dNode"]["<rules>"], dPointer["nOffset"]-1, dPriority, dOpt, sCountry, bShowRuleId, bContext)
dErr.update(errs)
if bHasChanged:
bChange = True
if dErr:
print(dErr)
return (bChange, dErr)
def _executeActions (self, dNode, nTokenOffset, dPriority, dOpt, sCountry, bShowRuleId, bContext):
#print(locals())
dErrs = {}
bChange = False
for sLineId, nextNodeKey in dNode.items():
for sRuleId in dGraph[nextNodeKey]:
print(sRuleId)
bCondMemo = None
sFuncCond, cActionType, sWhat, *eAct = dRule[sRuleId]
# action in lActions: [ condition, action type, replacement/suggestion/action[, iTokenStart, iTokenEnd[, nPriority, message, URL]] ]
try:
bCondMemo = not sFuncCond or globals()[sFuncCond](self.lToken, nTokenOffset, sCountry, bCondMemo)
if bCondMemo:
if cActionType == "-":
# grammar error
print("-")
nTokenErrorStart = nTokenOffset + eAct[0]
nTokenErrorEnd = nTokenOffset + eAct[1]
nErrorStart = self.iStart + self.lToken[nTokenErrorStart]["nStart"]
nErrorEnd = self.iStart + self.lToken[nTokenErrorEnd]["nEnd"]
if nErrorStart not in dErrs or eAct[2] > dPriority[nErrorStart]:
dErrs[nErrorStart] = _createTokenError(self.lToken, self.sSentence, self.sSentence0, sWhat, nTokenErrorStart, nErrorStart, nErrorEnd, sLineId, sRuleId, True, eAct[3], eAct[4], bShowRuleId, "notype", bContext)
dPriority[nErrorStart] = eAct[2]
elif cActionType == "~":
# text processor
print("~")
self._rewrite(sWhat, nErrorStart, nErrorEnd)
elif cActionType == "@":
# jump
print("@")
self._jump(sWhat)
elif cActionType == "=":
# disambiguation
print("=")
globals()[sWhat](self.lToken)
elif cActionType == ">":
# we do nothing, this test is just a condition to apply all following actions
print(">")
pass
else:
print("# error: unknown action at " + sLineId)
elif cActionType == ">":
break
except Exception as e:
raise Exception(str(e), sLineId)
return bChange, dErrs
def _rewrite (self, sWhat, nErrorStart, nErrorEnd):
"text processor: rewrite tokens between <nErrorStart> and <nErrorEnd> position"
lTokenValue = sWhat.split("|")
if len(lTokenValue) != (nErrorEnd - nErrorStart + 1):
print("Error. Text processor: number of replacements != number of tokens.")
return
for i, sValue in zip(range(nErrorStart, nErrorEnd+1), lTokenValue):
self.lToken[i]["sValue"] = sValue
def _jump (self, sWhat):
try:
nFrom, nTo = sWhat.split(">")
self.lToken[int(nFrom)]["iJump"] = int(nTo)
except:
print("# Error. Jump failed: ", sWhat)
traceback.print_exc()
return
#### Analyse tokens
def g_morph (dToken, sPattern, sNegPattern=""):
"analyse a token, return True if <sNegPattern> not in morphologies and <sPattern> in morphologies"
if "lMorph" in dToken:
lMorph = dToken["lMorph"]
else:
lMorph = _oSpellChecker.getMorph(dToken["sValue"])
if not lMorph:
return False
# check negative condition
if sNegPattern:
if sNegPattern == "*":
# all morph must match sPattern
zPattern = re.compile(sPattern)
return all(zPattern.search(sMorph) for sMorph in lMorph)
else:
zNegPattern = re.compile(sNegPattern)
if any(zNegPattern.search(sMorph) for sMorph in lMorph):
return False
# search sPattern
zPattern = re.compile(sPattern)
return any(zPattern.search(sMorph) for sMorph in lMorph)
def g_analyse (dToken, sPattern, sNegPattern=""):
"analyse a token, return True if <sNegPattern> not in morphologies and <sPattern> in morphologies (disambiguation off)"
lMorph = _oSpellChecker.getMorph(dToken["sValue"])
if not lMorph:
return False
# check negative condition
if sNegPattern:
if sNegPattern == "*":
zPattern = re.compile(sPattern)
return all(zPattern.search(sMorph) for sMorph in lMorph)
else:
zNegPattern = re.compile(sNegPattern)
if any(zNegPattern.search(sMorph) for sMorph in lMorph):
return False
# search sPattern
zPattern = re.compile(sPattern)
return any(zPattern.search(sMorph) for sMorph in lMorph)
#### Disambiguator
def g_select (dToken, sPattern, lDefault=None):
"select morphologies for <dToken> according to <sPattern>, always return True"
lMorph = dToken["lMorph"] if "lMorph" in dToken else _oSpellChecker.getMorph(dToken["sValue"])
if not lMorph or len(lMorph) == 1:
return True
lSelect = [ sMorph for sMorph in lMorph if re.search(sPattern, sMorph) ]
if lSelect:
if len(lSelect) != len(lMorph):
dToken["lMorph"] = lSelect
elif lDefault:
dToken["lMorph"] = lDefault
return True
def g_exclude (dToken, sPattern, lDefault=None):
"select morphologies for <dToken> according to <sPattern>, always return True"
lMorph = dToken["lMorph"] if "lMorph" in dToken else _oSpellChecker.getMorph(dToken["sValue"])
if not lMorph or len(lMorph) == 1:
return True
lSelect = [ sMorph for sMorph in lMorph if not re.search(sPattern, sMorph) ]
if lSelect:
if len(lSelect) != len(lMorph):
dToken["lMorph"] = lSelect
elif lDefault:
dToken["lMorph"] = lDefault
return True
def g_define (dToken, lMorph):
"set morphologies of <dToken>, always return True"
dToken["lMorph"] = lMorph
return True
#### CALLABLES (generated code)
${graph_callables}