Unambiguous grammar for arithmetic expression with Unary + and -

I have just started self-studying the Dragon book of Compiler Design. I am working on a problem that says to design grammar for an expression containin开发者_JAVA技巧g binary +,-,*,/ and unary +,-

I came up with following

E -> E+T | E-T | T
T -> T*P | T/P | P
P -> +S | -S | S
S -> id | constant | (E)

However, there is an obvious flaw in it. According to this grammar, expressions like

1--3

are valid, which is an error in all programming languages I know. Though, expressions like

1+-+3
and
1- -3

must be valid. How can such a grammar be designed?

I believe your problem is with tokenization. You are identifying 1--3 as an error because you think it should be resolved as 1 --3 rather than 1 - -3, the latter being perfectly valid. So I think you problem comes because when you tokenize the string you are getting:

['1', '-', '-' , '3']

rather than:

['1', '--', '3']

I think you have one extra production rule

P -> +S | -S | S
S -> id | constant | (E)

can be shrinked to

P -> +P | -P | id | constant | (E)

With such grammar you will succesfully match exp "1+-+3" as valid.

You have tokenizer(scanner) problem! before passing token to parser You must differentiate between "-" and "--". you must define a token struct which contains a token type and value, then parse the token list. Also the rule P->--S must be added to the production rules!!