from argparse import ArgumentParser import csv from parser import Parser import sqlite3 import string from scanner import Scanner import re import readline ################################################################################ # Predicates ################################################################################ def is_alpha(c): return c in string.ascii_letters def is_digit(c): return c in "0123456789" def is_alphanumeric(c): return is_alpha(c) or is_digit(c) def is_whitespace(c): return c in " \r\t\n" ################################################################################ # Tokenizer ################################################################################ AND = ("CONJUNCTION", "AND") OR = ("CONJUNCTION", "OR") NOT = ("PUNCTUATION", "NOT") COLON = ("PUNCTUATION", "COLON") LPAREN = ("PUNCTUATION", "LPAREN") RPAREN = ("PUNCTUATION", "RPAREN") def tokenize(x): s = Scanner(x) tokens = scan_tokens(s) return tokens def scan_tokens(s): result = [] while not s.exhausted(): if is_whitespace(s.peek()): s.advance() else: result.append(scan_token(s)) return result def scan_token(s): punctuation = { "-": NOT, ":": COLON, "(": LPAREN, ")": RPAREN, } c = s.peek() if c in punctuation: s.advance() return punctuation[c] if c == "\"": return tokenize_string(s) if c == "/": return tokenize_regex(s) if is_alpha(c): return tokenize_identifier(s) def tokenize_string(s): s.advance() # ignore opening 2x-quote current = "" while s.peek() != "\"" and not s.exhausted(): current += s.advance() if s.exhausted(): raise Exception("Unterminated string") s.advance() # ignore closing 2x-quote return ("STRING", current) def tokenize_regex(s): s.advance() # ignore opening forward-slash current = "" while s.peek() != "/" and not s.exhausted(): current += s.advance() if s.exhausted(): raise Exception("Unterminated regex") s.advance() # ignore closing forward-slash return ("REGEX", current) def tokenize_identifier(s): conjunctions = { "AND", "OR", } current = s.advance() while is_alphanumeric(s.peek()): current += s.advance() if current.upper() in conjunctions: return ("CONJUNCTION", current.upper()) else: return ("IDENTIFIER", current) ################################################################################ # Parser ################################################################################ # EBNF # Note: we order expression types by ascending levels of precedence. # # expression -> conjunction ; # conjunction -> selection ( ( "AND" | "OR" )? selection )* ; # selection -> "-"? IDENTIFIER ":" ( REGEX | STRING ) | grouping ; # grouping -> REGEX | STRING | "(" expression ")" ; def parse(x): tokens = tokenize(x) p = Parser(tokens) return expression(p) def expression(p): return conjunction(p) def conjunction(p): lhs = selection(p) # TODO(wpcarro): Support default AND conjuctions when they're undefined. while not p.exhausted() and p.match({AND, OR}): conj = p.peek(n=-1) rhs = selection(p) lhs = ("CONJUNCTION", conj[1], lhs, rhs) return lhs def selection(p): negate = False if p.peek() == NOT: negate = True p.advance() if p.peek()[0] != "IDENTIFIER": return grouping(p) ident = p.expect(lambda x: x[0] == "IDENTIFIER") colon = p.expect(lambda x: x[1] == "COLON") value = p.expect(lambda x: x[0] in {"REGEX", "STRING"}) return ("SELECTION", negate, ident[1], value) def grouping(p): if p.peek()[0] == "REGEX": return p.advance() if p.peek()[0] == "STRING": return p.advance() if p.peek() == LPAREN: p.advance() expr = expression(p) p.expect(lambda x: x == RPAREN) return ("GROUPING", expr) ################################################################################ # Compiler ################################################################################ def compile(source, table, columns): ast = parse(source) return "SELECT * FROM {} WHERE {};".format(table, do_compile(ast, columns)) def do_compile(ast, columns): if ast[0] == "REGEX": cols = "({})".format(" || ".join(columns)) return "{} REGEXP '.*{}.*'".format(cols, ast[1]) if ast[0] == "STRING": cols = "({})".format(" || ".join(columns)) return "{} LIKE '%{}%'".format(cols, ast[1]) if ast[0] == "SELECTION": return compile_selection(ast) if ast[0] == "CONJUNCTION": _, conj, lhs, rhs = ast lhs = do_compile(lhs, columns) rhs = do_compile(rhs, columns) return "{} {} {}".format(lhs, conj, rhs) if ast[0] == "GROUPING": return "({})".format(do_compile(ast[1], columns)) raise Exception("Unexpected AST: \"{}\"".format(ast)) def compile_selection(ast): _, negate, column, query = ast match = compile_query(negate, query) return "{} {}".format(column, match) def compile_query(negate, query): query_type, query_string = query if query_type == "REGEX": if negate: return "NOT REGEXP '.*{}.*'".format(query_string) return "REGEXP '.*{}.*'".format(query_string) if query_type == "STRING": if negate: return "NOT LIKE '%{}%'".format(query_string) return "LIKE '%{}%'".format(query_string) ################################################################################ # Helper Functions ################################################################################ def regexp(expr, x): reg = re.compile(expr) return reg.search(x) is not None ################################################################################ # Main ################################################################################ def main(csv_path=None, debug=False): # Import CSV to SQLite table = "main" con = sqlite3.connect(":memory:") con.create_function("REGEXP", 2, regexp) cur = con.cursor() with open(csv_path, "r") as f: r = csv.DictReader(f) columns = next(r).keys() # TODO(wpcarro): Use safer interpolation variant of "?" here and throughout. cur.execute("CREATE TABLE {} ({});".format(table, ",".join(columns))) rows = [tuple(row[col] for col in columns) for row in r] cur.executemany("INSERT INTO {} ({}) VALUES ({});".format(table, ",".join(columns), ",".join("?" for _ in columns)), rows) con.commit() while True: x = input("> ") if debug: print("tokens:\t{}".format(tokenize(x))) print("AST:\t{}".format(parse(x))) print("query:\t\"{}\"".format(compile(x, table, columns))) try: compile(x, table, columns) for row in cur.execute(compile(x, table, columns)): print("\t".join(str(cell) for cell in row)) except: print("Compilation error.") # TODO(wpcarro): Trap exits and ensure cleanup always runs. con.close() if __name__ == "__main__": parser = ArgumentParser() parser.add_argument("-f", "--file", dest="file", help="Path to the CSV from which to read", metavar="PATH") parser.add_argument("-d", "--debug", dest="debug", default=False, action="store_true", help="Enable debugging") args = parser.parse_args() main(csv_path=args.file, debug=args.debug)