diff --git a/expression.py b/expression.py index 59c58fb..214b344 100644 --- a/expression.py +++ b/expression.py @@ -1,190 +1,332 @@ #!/usr/bin/env python # encoding: utf-8 +from generic import Stack, flatten_list, expand_list +from fraction import Fraction + +def trace(f): + """ Permet de visualiser des appels récursifs à la fonction décorée """ + f.indent = 0 + def g(x): + print('| ' * f.indent + '|--', f.__name__, x) + f.indent += 1 + value = f(x) + print( '| ' * f.indent + '|--', 'return', repr(value)) + f.indent -= 1 + return value + return g class Expression(object): """A calculus expression. Today it can andle only expression with numbers later it will be able to manipulate unknown""" - priority = {"*" : 3, "/": 3, "+": 2, "-":2, "(": 1} + PRIORITY = {"*" : 3, "/": 3, "+": 2, "-":2, "(": 1} def __init__(self, exp): """ Initiate the expression :param exp: the expression. It can be a string or a list of tokens. It can be infix or postfix expression """ - self._exp = exp + if type(exp) == str: + self._exp = exp + self.tokens = self.str2tokens(exp) # les tokens seront alors stockés dans self.tokens temporairement + elif type(exp) == list: + self.tokens = exp - def simplify(self): - """ Generator which return steps for computing the expression """ - if self._exp == 0: - yield "Zero!" + self.find_fix() # Determine le fix et range la liste dans self.[fix]_tokens + + ## --------------------- + ## Mechanism functions + + def simplify(self, render = lambda x:str(x)): + """ Generator which return steps for computing the expression + @param render: function which render the list of token (postfix form now) + """ + if not self.can_go_further(): + yield render(self.postfix_tokens) # self.postfix_tokens devra être une propriété pour vérifier que quand on y accède, on a bien la forme voulu et si non, la calculer. else: - self.children = Expression(self._exp - 1) - yield "num: " + str(self._exp) - yield self.children.simplify().next() + self.compute_exp() # crée self.children et éventuellement les étapes intermédiaires pour y arriver. /!\ les étapes sont sous la forme de listes de tokens en postfix + for s in self.steps: + yield render(s) + for s in self.children.simplify(render = render): + yield render(s) + + def can_go_further(self): + """Check whether it's a last step or not. If not create self.children the next expression. + :returns: 1 if it's not the last step, 0 otherwise + """ + if len(self.tokens) == 1: + return 0 + else: + return 1 + + def compute_exp(self): + """ Create self.children with self.steps to go up to it """ + self.steps = [self.postfix_tokens] + + tokenList = self.postfix_tokens.copy() + tmpTokenList = [] + + while len(tokenList) > 2: + # on va chercher les motifs du genre A B + pour les calculer + if isNumber(tokenList[0]) and isNumber(tokenList[1]) and self.isOperator(tokenList[2]): + + # S'il y a une opération à faire + op1 = tokenList[0] + op2 = tokenList[1] + token = tokenList[2] + + res = doMath(token, op1, op2) + + tmpTokenList.append(res) + + # Comme on vient de faire le calcul, on peut détruire aussi les deux prochains termes + del tokenList[0:3] + else: + tmpTokenList.append(tokenList[0]) + + del tokenList[0] + tmpTokenList += tokenList + + steps += expand_list(tmpTokenList) + + self.steps += [steps[:-1]] + self.children = Expression(steps[-1]) + ## --------------------- ## String parsing ## @classmethod ???? - #def parseExp(self, exp): - # """ Parse the expression, ie tranform a string into a list of tokens + def str2tokens(self, exp): + """ Parse the expression, ie tranform a string into a list of tokens - # :param exp: The expression (a string) - # :returns: list of token + :param exp: The expression (a string) + :returns: list of token - # """ - # return exp.split(" ") + """ + return exp.split(" ") - ## --------------------- - ## "fix" recognition + # --------------------- + # "fix" recognition - #def get_fix(self, tokens): - # """ Give the "fix" of an expression - # infix -> A + B - # prefix -> + A B - # postfix -> A B + + @classmethod + def get_fix(self, tokens): + """ Give the "fix" of an expression + [A, +, B] -> infix + [+, A, B] -> prefix + [A, B, +] -> postfix + /!\ does not verify if the expression is correct/computable! - # :param exp: the expression (list of token) - # :returns: the "fix" (infix, postfix, prefix) + :param exp: the expression (list of token) + :returns: the "fix" (infix, postfix, prefix) - # """ - # if tokens[0] in "+-*/": - # return "prefix" - # elif token[0] not in "+-*/" ans token[1] not in "+-*/": - # return "postfix" - # else: - # return "infix" + """ + if self.isOperator(tokens[0]): + return "prefix" + elif not self.isOperator(tokens[0]) and not self.isOperator(tokens[1]): + return "postfix" + else: + return "infix" - ## ---------------------- - ## Expressions - tokens getters + def find_fix(self): + """ Recognize the fix of self.tokens and stock tokens in self.[fix]_tokens """ + fix = self.get_fix(self.tokens) - #def tokens(self, fix = "infix"): - # """Get the list of tokens with the wanted fix + setattr(self, fix+"_tokens", self.tokens) - # :param fix: the fix wanted (infix default) - # :returns: list of tokens + # ---------------------- + # Expressions - tokens manipulation - # """ - # # Si ce fix n'a pas encore été enregistré - # if fix not in self._tokens: - # # Il peut venir de la version string - # if fix in self._string: - # self._tokens[fix] = self.parseExp(self._string[fix]) - # # Sinon il faut le calculer à partir d'une autre forme - # else: - # fix_transfo = "to" + fix.capitalize() - # getattr(self,fix_transfo)() + @property + def infix_tokens(self): + """ Return infix list of tokens. Verify if it has already been computed and compute it if not - # return self._tokens[fix] + :returns: infix list of tokens + """ + if hasattr(self, "_infix_tokens"): + return self._infix_tokens -###### On en est là,il faudra faire attention à bien vérifier ce que les "to..." enregistre (string ou tokens) + elif hasattr(self, "_postfix_tokens"): + self.infix_tokens = self.post2in_fix() + return self._infix_tokens - #def string(self, fix = "infix"): - # """Get the string with the wanted fix + else: + raise ValueError("Unkown fix") - # :param fix: the fix wanted (infix default) - # :returns: the string representing the expression + @infix_tokens.setter + def infix_tokens(self, val): + self._infix_tokens = val - # """ - # # Si ce fix n'a pas encore été enregistré - # if fix not in self._string: - # # Il peut venir de la version string - # if fix in self._tokens: - # self._string[fix] = self.parseExp(self._string[fix]) - # # Sinon il faut le calculer à partir d'une autre forme - # else: - # fix_transfo = "to" + fix.capitalize() - # getattr(self,fix_transfo)() + @property + def postfix_tokens(self): + """ Return postfix list of tokens. Verify if it has already been computed and compute it if not - # return self._string[fix] - # - ## ---------------------- - ## "fix" tranformations + :returns: postfix list of tokens + """ + if hasattr(self, "_postfix_tokens"): + return self._postfix_tokens - #def toPostfix(self): - # """ Transorm the expression into postfix form using the infix form""" - # pass + elif hasattr(self,"_infix_tokens"): + self.postfix_tokens = self.in2post_fix() + return self._postfix_tokens - #def toInfix(self): - # """ Tranform the expression into infix form using postfix form""" - # pass + else: + raise ValueError("Unkown fix") - ## --------------------- - ## Tools for placing parenthesis in infix notation + @postfix_tokens.setter + def postfix_tokens(self, val): + self._postfix_tokens = val - #def needPar(operande, operator, posi = "after"): - # """ Says whether or not the operande needs parenthesis + # ---------------------- + # "fix" tranformations - # :param operande: the operande - # :param operator: the operator - # :param posi: "after"(default) if the operande will be after the operator, "before" othewise - # :returns: bollean - # """ - # if isNumber(operande) and "-" in operande: - # return 1 - # elif not isNumber(operande): - # # Si c'est une grande expression ou un chiffre négatif - # stand_alone = get_main_op(operande) - # # Si la priorité de l'operande est plus faible que celle de l'opérateur - # minor_priority = self.priority[get_main_op(operande)] < self.priority[operator] - # # Si l'opérateur est -/ pour after ou juste / pour before - # special = (operator in "-/" and posi == "after") or (operator in "/" and posi == "before") + def in2post_fix(self): + """ From the self.infix_tokens list compute the corresponding self.postfix_tokens list """ + + opStack = Stack() + postfixList = [] - # return stand_alone and (minor_priority or special) - # else: - # return 0 - # -# J'#aime pas bien cette endroit faudrait que ce soit une méthode qui s'applique uniquement à l'expression en question (self) pas à n'importe quel string, ça serait plus propre. - #def get_main_op(exp): - # """ Gives the main operation of the expression + for token in self.infix_tokens: + if token == "(": + opStack.push(token) + elif token == ")": + topToken = opStack.pop() + while topToken != "(": + postfixList.append(topToken) + topToken = opStack.pop() + elif self.isOperator(token): + # On doit ajouter la condition == str sinon python ne veut pas tester l'appartenance à la chaine de caractère. + while (not opStack.isEmpty()) and (self.PRIORITY[opStack.peek()] >= self.PRIORITY[token]): + postfixList.append(opStack.pop()) + opStack.push(token) + else: + postfixList.append(token) - # :param exp: the expression - # :returns: the main operation (+, -, * or /) or 0 if the expression is only one element + while not opStack.isEmpty(): + postfixList.append(opStack.pop()) - # """ - # parStack = Stack() - # tokenList = exp.split(" ") + self.postfix_tokens = postfixList - # if len(tokenList) == 1: - # # Si l'expression n'est qu'un élément - # return 0 + def post2in_fix(self): + """ From the self.postfix_tokens list compute the corresponding self.infix_tokens list """ + operandeStack = Stack() - # main_op = [] + for token in postfixTokens: + if self.isOperator(token): + op2 = operandeStack.pop() + if self.needPar(op2, token, "after"): + op2 = ["( ", op2, " )"] + op1 = operandeStack.pop() + if self.needPar(op1, token, "before"): + op1 = ["( ", op1, " )"] + res = [op1, token, op2] - # for token in tokenList: - # if token == "(": - # parStack.push(token) - # elif token == ")": - # parStack.pop() - # elif token in "+-*/" and parStack.isEmpty(): - # main_op.append(token) + operandeStack.push(res) - # return min(main_op, key = lambda s: priority[s]) + else: + operandeStack.push(token) + + self.infix_tokens = flatten_list(operandeStack.pop()) + + # --------------------- + # Tools for placing parenthesis in infix notation + + @classmethod + def needPar(self, operande, operator, posi = "after"): + """Says whether or not the operande needs parenthesis + + :param operande: the operande + :param operator: the operator + :param posi: "after"(default) if the operande will be after the operator, "before" othewise + :returns: bollean + """ + + if isNumber(operande) and operande < 0: + return 1 + elif not isNumber(operande): + # Si c'est une grande expression ou un chiffre négatif + stand_alone = self.get_main_op(operande) + # Si la priorité de l'operande est plus faible que celle de l'opérateur + minor_priority = self.PRIORITY[self.get_main_op(operande)] < self.PRIORITY[operator] + # Si l'opérateur est -/ pour after ou juste / pour before + special = (operator in "-/" and posi == "after") or (operator in "/" and posi == "before") + + return stand_alone and (minor_priority or special) + else: + return 0 + + @classmethod + def get_main_op(self, tokens): + """Getting the main operation of the list of tokens + + :param exp: the list of tokens + :returns: the main operation (+, -, * or /) or 0 if the expression is only one element + + """ + parStack = Stack() + + if len(tokens) == 1: + # Si l'expression n'est qu'un élément + return 0 + + main_op = [] + + for token in tokens: + if token == "(": + parStack.push(token) + elif token == ")": + parStack.pop() + elif self.isOperator(token) and parStack.isEmpty(): + main_op.append(token) + + return min(main_op, key = lambda s: self.PRIORITY[s]) ## --------------------- ## Computing the expression - #def compute(self): - # """ Recursive method for computing as a student the expression - # :returns: list of steps needed to reach the result + @classmethod + def doMath(self, op, op1, op2): + """Compute "op1 op op2" or create a fraction - # """ - # pass - # - #def doMath(op, op1, op2): - # """Compute "op1 op op2" + :param op: operator + :param op1: first operande + :param op2: second operande + :returns: string representing the result - # :param op: operator - # :param op1: first operande - # :param op2: second operande - # :returns: string representing the result + """ + operations = {"+": "__add__", "-": "__sub__", "*": "__mul__"} + if op == "/": + ans = [Fraction(op1, op2)] + ans += ans[0].simplify() + return ans + else: + return getattr(op1,operations[op])(op2) - # """ - # return str(eval(op1 + op + op2)) + ## --------------------- + ## Recognize numbers and operators + + @classmethod + def isNumber(self, exp): + """Check if the expression can be a number + + :param exp: an expression + :returns: True if the expression can be a number and false otherwise + + """ + return type(exp) == int or type(exp) == Fraction + + @classmethod + def isOperator(self, exp): + """Check if the expression is an opération in "+-*/" + + :param exp: an expression + :returns: boolean + + """ + return (type(exp) == str and exp in "+-*/") if __name__ == '__main__': - a = Expression(10) + a = Expression("10 + 1 * 3") for i in a.simplify(): print(i) diff --git a/generic.py b/generic.py index 84f337b..8f03d85 100644 --- a/generic.py +++ b/generic.py @@ -57,6 +57,48 @@ class Stack(object): def __add__(self, addList): return self.items + addList + +def flatten_list(a, result=None): + """Flattens a nested list. + + >>> flatten_list([ [1, 2, [3, 4] ], [5, 6], 7]) + [1, 2, 3, 4, 5, 6, 7] + """ + if result is None: + result = [] + + for x in a: + if isinstance(x, list): + flatten_list(x, result) + else: + result.append(x) + + return result + +def expand_list(list_list): + """Expand list of list + + >>> expand_list([1,2,[3,4],5,[6,7,8]]) + [[1, 2, 3, 5, 6], [1, 2, 4, 5, 7], [1, 2, 4, 5, 8]] + >>> expand_list([1,2,4,5,6,7,8]) + [[1, 2, 4, 5, 6, 7, 8]] + + """ + list_in_list = [i for i in list_list if type(i) == list].copy() + + try: + nbr_ans_list = max([len(i) for i in list_in_list]) + + ans = [list_list.copy() for i in range(nbr_ans_list)] + for (i,l) in enumerate(ans): + for (j,e) in enumerate(l): + if type(e) == list: + ans[i][j] = e[min(i,len(e)-1)] + # S'il n'y a pas eut d'étapes intermédiaires (2e exemple) + except ValueError: + ans = [list_list] + + return ans # ----------------------------- # Reglages pour 'vim' # vim:set autoindent expandtab tabstop=4 shiftwidth=4: