# Create a Direct Acyclic Rule Graph (DARG)
import re
import traceback
import json
import darg
dDEF = {}
dACTIONS = {}
lFUNCTIONS = []
def prepareFunction (s):
s = s.replace("__also__", "bCondMemo")
s = s.replace("__else__", "not bCondMemo")
s = re.sub(r"isStart *\(\)", 'before(["<START>", ","])', s)
s = re.sub(r"isRealStart *\(\)", 'before(["<START>"])', s)
s = re.sub(r"isStart0 *\(\)", 'before0(["<START>", ","])', s)
s = re.sub(r"isRealStart0 *\(\)", 'before0(["<START>"])', s)
s = re.sub(r"isEnd *\(\)", 'after(["<END>", ","])', s)
s = re.sub(r"isRealEnd *\(\)", 'after(["<END>"])', s)
s = re.sub(r"isEnd0 *\(\)", 'after0(["<END>", ","])', s)
s = re.sub(r"isRealEnd0 *\(\)", 'after0(["<END>"])', s)
s = re.sub(r"(select|exclude)[(][\\](\d+)", '\\1(lToken[\\2]', s)
s = re.sub(r"define[(][\\](\d+)", 'define(lToken[\\1]', s)
s = re.sub(r"(morph|morphex|displayInfo)[(]\\(\d+)", '\\1(lToken[\\2]', s)
s = re.sub(r"token\(\s*(\d)", 'nextToken(\\1', s) # token(n)
s = re.sub(r"token\(\s*-(\d)", 'prevToken(\\1', s) # token(-n)
s = re.sub(r"before\(\s*", 'look(s[:m.start()], ', s) # before(s)
s = re.sub(r"after\(\s*", 'look(s[m.end():], ', s) # after(s)
s = re.sub(r"textarea\(\s*", 'look(s, ', s) # textarea(s)
s = re.sub(r"before_chk1\(\s*", 'look_chk1(dDA, s[:m.start()], 0, ', s) # before_chk1(s)
s = re.sub(r"after_chk1\(\s*", 'look_chk1(dDA, s[m.end():], m.end(), ', s) # after_chk1(s)
s = re.sub(r"textarea_chk1\(\s*", 'look_chk1(dDA, s, 0, ', s) # textarea_chk1(s)
s = re.sub(r"isEndOfNG\(\s*\)", 'isEndOfNG(dDA, s[m.end():], m.end())', s) # isEndOfNG(s)
s = re.sub(r"isNextNotCOD\(\s*\)", 'isNextNotCOD(dDA, s[m.end():], m.end())', s) # isNextNotCOD(s)
s = re.sub(r"isNextVerb\(\s*\)", 'isNextVerb(dDA, s[m.end():], m.end())', s) # isNextVerb(s)
s = re.sub(r"\bspell *[(]", '_oSpellChecker.isValid(', s)
s = re.sub(r"[\\](\d+)", 'lToken[\\1]', s)
return s
def changeReferenceToken (s, dPos):
for i in range(len(dPos), 0, -1):
s = s.replace("\\"+str(i), "\\"+dPos[i])
return s
def createRule (iLine, sRuleName, sTokenLine, sActions, nPriority):
# print(iLine, "//", sRuleName, "//", sTokenLine, "//", sActions, "//", nPriority)
lToken = sTokenLine.split()
# Calculate positions
dPos = {}
nGroup = 0
for i, sToken in enumerate(lToken):
if sToken.startswith("(") and sToken.endswith(")"):
lToken[i] = sToken[1:-1]
nGroup += 1
dPos[nGroup] = i
# Parse actions
for nAction, sAction in enumerate(sActions.split(" <<- ")):
if sAction.strip():
sActionId = sRuleName + "_a" + str(nAction)
aAction = createAction(sActionId, sAction, nGroup, nPriority, dPos)
if aAction:
dACTIONS[sActionId] = aAction
lResult = list(lToken)
lResult.extend(["##"+str(iLine), sActionId])
yield lResult
def createAction (sIdAction, sAction, nGroup, nPriority, dPos):
m = re.search("([-~=])(\\d+|)(:\\d+|)>> ", sAction)
if not m:
print(" # Error. No action found at: ", sIdAction)
print(" ==", sAction, "==")
return None
# Condition
sCondition = sAction[:m.start()].strip()
if sCondition:
sCondition = prepareFunction(sCondition)
sCondition = changeReferenceToken(sCondition, dPos)
lFUNCTIONS.append(("g_c_"+sIdAction, sCondition))
sCondition = "g_c_"+sIdAction
else:
sCondition = ""
# Action
cAction = m.group(1)
sAction = sAction[m.end():].strip()
sAction = changeReferenceToken(sAction, dPos)
iStartAction = int(m.group(2)) if m.group(2) else 0
iEndAction = int(m.group(3)[1:]) if m.group(3) else iStartAction
if nGroup:
iStartAction = dPos[iStartAction]
iEndAction = dPos[iEndAction]
if cAction == "-":
## error
iMsg = sAction.find(" # ")
if iMsg == -1:
sMsg = "# Error. Error message not found."
sURL = ""
print(sMsg + " Action id: " + sIdAction)
else:
sMsg = sAction[iMsg+3:].strip()
sAction = sAction[:iMsg].strip()
sURL = ""
mURL = re.search("[|] *(https?://.*)", sMsg)
if mURL:
sURL = mURL.group(1).strip()
sMsg = sMsg[:mURL.start(0)].strip()
if sMsg[0:1] == "=":
sMsg = prepareFunction(sMsg[1:])
lFUNCTIONS.append(("g_m_"+sIdAction, sMsg))
for x in re.finditer("group[(](\\d+)[)]", sMsg):
if int(x.group(1)) > nGroup:
print("# Error in groups in message at line " + sIdAction + " ("+str(nGroup)+" groups only)")
sMsg = "=g_m_"+sIdAction
else:
for x in re.finditer(r"\\(\d+)", sMsg):
if int(x.group(1)) > nGroup:
print("# Error in groups in message at line " + sIdAction + " ("+str(nGroup)+" groups only)")
if re.search("[.]\\w+[(]", sMsg):
print("# Error in message at line " + sIdAction + ": This message looks like code. Line should begin with =")
if sAction[0:1] == "=" or cAction == "=":
if "define" in sAction and not re.search(r"define\(\\\d+ *, *\[.*\] *\)", sAction):
print("# Error in action at line " + sIdAction + ": second argument for define must be a list of strings")
sAction = prepareFunction(sAction)
for x in re.finditer("group[(](\\d+)[)]", sAction):
if int(x.group(1)) > nGroup:
print("# Error in groups in replacement at line " + sIdAction + " ("+str(nGroup)+" groups only)")
else:
for x in re.finditer(r"\\(\d+)", sAction):
if int(x.group(1)) > nGroup:
print("# Error in groups in replacement at line " + sIdAction + " ("+str(nGroup)+" groups only)")
if re.search("[.]\\w+[(]|sugg\\w+[(]", sAction):
print("# Error in action at line " + sIdAction + ": This action looks like code. Line should begin with =")
if cAction == "-":
## error detected --> suggestion
if not sAction:
print("# Error in action at line " + sIdAction + ": This action is empty.")
if sAction[0:1] == "=":
lFUNCTIONS.append(("g_s_"+sIdAction, sAction[1:]))
sAction = "=g_s_"+sIdAction
elif sAction.startswith('"') and sAction.endswith('"'):
sAction = sAction[1:-1]
if not sMsg:
print("# Error in action at line " + sIdAction + ": The message is empty.")
return [sCondition, cAction, sAction, iStartAction, iEndAction, nPriority, sMsg, sURL]
elif cAction == "~":
## text processor
if not sAction:
print("# Error in action at line " + sIdAction + ": This action is empty.")
if sAction[0:1] == "=":
lFUNCTIONS.append(("g_p_"+sIdAction, sAction[1:]))
sAction = "=g_p_"+sIdAction
elif sAction.startswith('"') and sAction.endswith('"'):
sAction = sAction[1:-1]
return [sCondition, cAction, sAction, iStartAction, iEndAction]
elif cAction == "=":
## disambiguator
if sAction[0:1] == "=":
sAction = sAction[1:]
if not sAction:
print("# Error in action at line " + sIdAction + ": This action is empty.")
lFUNCTIONS.append(("g_d_"+sIdAction, sAction))
sAction = "g_d_"+sIdAction
return [sCondition, cAction, sAction]
elif cAction == ">":
## no action, break loop if condition is False
return [sCondition, cAction, ""]
else:
print("# Unknown action at line " + sIdAction)
return None
def make (spLang, sLang, bJavaScript):
"compile rules, returns a dictionary of values"
# for clarity purpose, don’t create any file here
print("> read graph rules file...")
try:
lRules = open(spLang + "/rules_graph.grx", 'r', encoding="utf-8").readlines()
except:
print("Error. Rules file in project [" + sLang + "] not found.")
exit()
# removing comments, zeroing empty lines, creating definitions, storing tests, merging rule lines
print(" parsing rules...")
global dDEF
lLine = []
lRuleLine = []
lTest = []
lOpt = []
lTokenLine = []
sActions = ""
nPriority = 4
for i, sLine in enumerate(lRules, 1):
sLine = sLine.rstrip()
if "\t" in sLine:
print("Error. Tabulation at line: ", i)
break
if sLine.startswith('#END'):
printBookmark(0, "BREAK BY #END", i)
break
elif sLine.startswith("#"):
pass
elif sLine.startswith("DEF:"):
m = re.match("DEF: +([a-zA-Z_][a-zA-Z_0-9]*) +(.+)$", sLine.strip())
if m:
dDEF["{"+m.group(1)+"}"] = m.group(2)
else:
print("Error in definition: ", end="")
print(sLine.strip())
elif sLine.startswith("TEST:"):
lTest.append("{:<8}".format(i) + " " + sLine[5:].strip())
elif sLine.startswith("TODO:"):
pass
elif sLine.startswith("!!"):
m = re.search("^!!+", sLine)
nExMk = len(m.group(0))
if sLine[nExMk:].strip():
printBookmark(nExMk-2, sLine[nExMk:].strip(), i)
elif sLine.startswith("__") and sLine.endswith("__"):
# new rule group
m = re.match("__(\\w+)(!\\d|)__", sLine)
if m:
sRuleName = m.group(1)
nPriority = int(m.group(2)[1:]) if m.group(2) else 4
else:
print("Error at rule group: ", sLine, " -- line:", i)
break
elif re.match("[ ]*$", sLine):
# empty line to end merging
for i, sTokenLine in lTokenLine:
lRuleLine.append((i, sRuleName, sTokenLine, sActions, nPriority))
lTokenLine = []
sActions = ""
sRuleName = ""
nPriority = 4
elif sLine.startswith((" ")):
# actions
sActions += " " + sLine.strip()
else:
lTokenLine.append([i, sLine.strip()])
# tests
print(" list tests...")
sGCTests = "\n".join(lTest)
sGCTestsJS = '{ "aData2": ' + json.dumps(lTest, ensure_ascii=False) + " }\n"
# processing rules
print(" preparing rules...")
lPreparedRule = []
for i, sRuleGroup, sTokenLine, sActions, nPriority in lRuleLine:
for lRule in createRule(i, sRuleGroup, sTokenLine, sActions, nPriority):
lPreparedRule.append(lRule)
# Graph creation
for e in lPreparedRule:
print(e)
oDARG = darg.DARG(lPreparedRule, sLang)
oRuleGraph = oDARG.createGraph()
# creating file with all functions callable by rules
print(" creating callables...")
sPyCallables = "# generated code, do not edit\n"
#sJSCallables = "// generated code, do not edit\nconst oEvalFunc = {\n"
for sFuncName, sReturn in lFUNCTIONS:
if sFuncName.startswith("g_c_"): # condition
sParams = "lToken, sCountry, bCondMemo"
elif sFuncName.startswith("g_m_"): # message
sParams = "lToken"
elif sFuncName.startswith("g_s_"): # suggestion
sParams = "lToken"
elif sFuncName.startswith("g_p_"): # preprocessor
sParams = "lToken"
elif sFuncName.startswith("g_d_"): # disambiguator
sParams = "lToken"
else:
print("# Unknown function type in [" + sFuncName + "]")
continue
sPyCallables += "def {} ({}):\n".format(sFuncName, sParams)
sPyCallables += " return " + sReturn + "\n"
#sJSCallables += " {}: function ({})".format(sFuncName, sParams) + " {\n"
#sJSCallables += " return " + jsconv.py2js(sReturn) + ";\n"
#sJSCallables += " },\n"
#sJSCallables += "}\n"
# Result
d = {
"graph_callables": sPyCallables,
"graph_gctests": None,
"rules_graph": oRuleGraph,
"rules_actions": dACTIONS
}
return d