You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
AQuery/aquery_parser/utils.py

619 lines
16 KiB

# encoding: utf-8
#
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this file,
# You can obtain one at http://mozilla.org/MPL/2.0/.
#
# Contact: Kyle Lahnakoski (kyle@lahnakoski.com)
#
import ast
from mo_dots import is_data, is_null, Data, from_data
from mo_future import text, number_types, binary_type, flatten
from mo_imports import expect
from mo_parsing import *
from mo_parsing.utils import is_number, listwrap
unary_ops = expect("unary_ops")
class Call(object):
__slots__ = ["op", "args", "kwargs"]
def __init__(self, op, args, kwargs):
self.op = op
self.args = args
self.kwargs = kwargs
IDENT_CHAR = Regex("[@_$0-9A-Za-zÀ-ÖØ-öø-ƿ]").expr.parser_config.include
FIRST_IDENT_CHAR = "".join(set(IDENT_CHAR) - set("0123456789"))
SQL_NULL = Call("null", [], {})
null_locations = []
def keyword(keywords):
return And([
Keyword(k, caseless=True) for k in keywords.split(" ")
]).set_parser_name(keywords) / (lambda: keywords.replace(" ", "_"))
def flag(keywords):
"""
RETURN {keywords: True}
"""
return (keyword(keywords) / (lambda: True))(keywords.replace(" ", "_"))
def assign(key: str, value: ParserElement):
return keyword(key).suppress() + value(key.replace(" ", "_"))
def simple_op(op, args, kwargs):
if args is None:
kwargs[op] = {}
else:
kwargs[op] = args
return kwargs
def normal_op(op, args, kwargs):
output = Data(op=op)
args = listwrap(args)
if args and (not isinstance(args[0], dict) or args[0]):
output.args = args
if kwargs:
output.kwargs = kwargs
return from_data(output)
scrub_op = simple_op
def scrub(result):
if result is SQL_NULL:
return SQL_NULL
elif result == None:
return None
elif isinstance(result, text):
return result
elif isinstance(result, binary_type):
return result.decode("utf8")
elif isinstance(result, number_types):
return result
elif isinstance(result, Call):
kwargs = scrub(result.kwargs)
args = scrub(result.args)
if args is SQL_NULL:
null_locations.append((kwargs, result.op))
return scrub_op(result.op, args, kwargs)
elif isinstance(result, dict) and not result:
return result
elif isinstance(result, list):
output = [rr for r in result for rr in [scrub(r)]]
if not output:
return None
elif len(output) == 1:
return output[0]
else:
for i, v in enumerate(output):
if v is SQL_NULL:
null_locations.append((output, i))
return output
else:
# ATTEMPT A DICT INTERPRETATION
try:
kv_pairs = list(result.items())
except Exception as c:
print(c)
output = {k: vv for k, v in kv_pairs for vv in [scrub(v)] if not is_null(vv)}
if isinstance(result, dict) or output:
for k, v in output.items():
if v is SQL_NULL:
null_locations.append((output, k))
return output
return scrub(list(result))
def _chunk(values, size):
acc = []
for v in values:
acc.append(v)
if len(acc) == size:
yield acc
acc = []
if acc:
yield acc
def to_lambda(tokens):
params, op, expr = list(tokens)
return Call("lambda", [expr], {"params": list(params)})
def to_json_operator(tokens):
# ARRANGE INTO {op: params} FORMAT
length = len(tokens.tokens)
if length == 2:
if tokens.tokens[1].type.parser_name == "cast":
return Call("cast", list(tokens), {})
# UNARY OPERATOR
op = tokens.tokens[0].type.parser_name
if op == "neg" and is_number(tokens[1]):
return -tokens[1]
return Call(op, [tokens[1]], {})
elif length == 5:
# TRINARY OPERATOR
return Call(
tokens.tokens[1].type.parser_name, [tokens[0], tokens[2], tokens[4]], {}
)
op = tokens[1]
if not isinstance(op, text):
op = op.type.parser_name
op = binary_ops.get(op, op)
if op == "eq":
if tokens[2] is SQL_NULL:
return Call("missing", tokens[0], {})
elif tokens[0] is SQL_NULL:
return Call("missing", tokens[2], {})
elif op == "neq":
if tokens[2] is SQL_NULL:
return Call("exists", tokens[0], {})
elif tokens[0] is SQL_NULL:
return Call("exists", tokens[2], {})
elif op == "eq!":
if tokens[2] is SQL_NULL:
return Call("missing", tokens[0], {})
elif tokens[0] is SQL_NULL:
return Call("missing", tokens[2], {})
elif op == "ne!":
if tokens[2] is SQL_NULL:
return Call("exists", tokens[0], {})
elif tokens[0] is SQL_NULL:
return Call("exists", tokens[2], {})
elif op == "is":
if tokens[2] is SQL_NULL:
return Call("missing", tokens[0], {})
else:
return Call("exists", tokens[0], {})
elif op == "is_not":
if tokens[2] is SQL_NULL:
return Call("exists", tokens[0], {})
else:
return Call("missing", tokens[0], {})
operands = [tokens[0], tokens[2]]
binary_op = Call(op, operands, {})
if op in {"add", "mul", "and", "or"}:
# ASSOCIATIVE OPERATORS
acc = []
for operand in operands:
while isinstance(operand, ParseResults) and isinstance(operand.type, Group):
# PARENTHESES CAUSE EXTRA GROUP LAYERS
operand = operand[0]
if isinstance(operand, ParseResults) and isinstance(
operand.type, Forward
):
operand = operand[0]
if isinstance(operand, Call) and operand.op == op:
acc.extend(operand.args)
elif isinstance(operand, list):
acc.append(operand)
elif isinstance(operand, dict) and operand.get(op):
acc.extend(operand.get(op))
else:
acc.append(operand)
binary_op = Call(op, acc, {})
return binary_op
def to_offset(tokens):
expr, offset = tokens.tokens
return Call("get", [expr, offset], {})
def to_window_mod(tokens):
expr, window = tokens.tokens
return Call("value", [expr], {**window})
def to_tuple_call(tokens):
# IS THIS ONE VALUE IN (), OR MANY?
tokens = list(tokens)
if len(tokens) == 1:
return [tokens[0]]
if all(isinstance(r, number_types) for r in tokens):
return [tokens]
if all(
isinstance(r, number_types) or (is_data(r) and "literal" in r.keys())
for r in tokens
):
candidate = {"literal": [r["literal"] if is_data(r) else r for r in tokens]}
return candidate
return [tokens]
binary_ops = {
"::": "cast",
"COLLATE": "collate",
"||": "concat",
"*": "mul",
"/": "div",
"%": "mod",
"+": "add",
"-": "sub",
"&": "binary_and",
"|": "binary_or",
"<": "lt",
"<=": "lte",
">": "gt",
">=": "gte",
"=": "eq",
"==": "eq",
"is distinct from": "eq!", # https://sparkbyexamples.com/apache-hive/hive-relational-arithmetic-logical-operators/
"is_distinct_from": "eq!",
"is not distinct from": "ne!",
"is_not_distinct_from": "ne!",
"<=>": "eq!", # https://sparkbyexamples.com/apache-hive/hive-relational-arithmetic-logical-operators/
"!=": "neq",
"<>": "neq",
"not in": "nin",
"in": "in",
"is_not": "neq",
"is": "eq",
"similar_to": "similar_to",
"like": "like",
"rlike": "rlike",
"not like": "not_like",
"not_like": "not_like",
"not rlike": "not_rlike",
"not_rlike": "not_rlike",
"not_simlilar_to": "not_similar_to",
"or": "or",
"and": "and",
"->": "lambda",
"union": "union",
"union_all": "union_all",
"union all": "union_all",
"except": "except",
"minus": "minus",
"intersect": "intersect",
}
is_set_op = ("union", "union_all", "except", "minus", "intersect")
def to_trim_call(tokens):
frum = tokens["from"]
if not frum:
return Call("trim", [tokens["chars"]], {"direction": tokens["direction"]})
return Call(
"trim",
[frum],
{"characters": tokens["chars"], "direction": tokens["direction"]},
)
def to_json_call(tokens):
# ARRANGE INTO {op: params} FORMAT
op = tokens["op"].lower()
op = binary_ops.get(op, op)
params = tokens["params"]
if isinstance(params, (dict, str, int, Call)):
args = [params]
else:
args = list(params)
kwargs = {k: v for k, v in tokens.items() if k not in ("op", "params")}
return ParseResults(
tokens.type,
tokens.start,
tokens.end,
[Call(op, args, kwargs)],
tokens.failures,
)
def to_interval_call(tokens):
# ARRANGE INTO {interval: [amount, type]} FORMAT
params = tokens["params"]
if not params:
params = {}
if params.length() == 2:
return Call("interval", params, {})
return Call("add", [Call("interval", p, {}) for p in _chunk(params, size=2)], {})
def to_case_call(tokens):
cases = list(tokens["case"])
elze = tokens["else"]
if elze != None:
cases.append(elze)
return Call("case", cases, {})
def to_switch_call(tokens):
# CONVERT TO CLASSIC CASE STATEMENT
value = tokens["value"]
acc = []
for c in list(tokens["case"]):
acc.append(Call("when", [Call("eq", [value] + c.args, {})], c.kwargs))
elze = tokens["else"]
if elze != None:
acc.append(elze)
return Call("case", acc, {})
def to_when_call(tokens):
tok = tokens
return Call("when", [tok["when"]], {"then": tok["then"]})
def to_join_call(tokens):
op = " ".join(tokens["op"])
if tokens["join"]["name"]:
output = {op: {
"name": tokens["join"]["name"],
"value": tokens["join"]["value"],
}}
else:
output = {op: tokens["join"]}
output["on"] = tokens["on"]
output["using"] = tokens["using"]
return output
def to_expression_call(tokens):
if set(tokens.keys()) & {"over", "within", "filter"}:
return
return ParseResults(
tokens.type,
tokens.start,
tokens.end,
listwrap(tokens["value"]),
tokens.failures,
)
def to_over(tokens):
if not tokens:
return {}
def to_alias(tokens):
cols = tokens["col"]
name = tokens["name"]
if cols:
return {name: cols}
return name
def to_top_clause(tokens):
value = tokens["value"]
if not value:
return None
value = value.value()
if tokens["ties"]:
output = {}
output["ties"] = True
if tokens["percent"]:
output["percent"] = value
else:
output["value"] = value
return output
elif tokens["percent"]:
return {"percent": value}
else:
return [value]
def to_row(tokens):
columns = list(tokens)
if len(columns) > 1:
return {"select": [{"value": v[0]} for v in columns]}
else:
return {"select": {"value": columns[0]}}
def get_literal(value):
if isinstance(value, (int, float)):
return value
elif isinstance(value, Call):
return
elif value is SQL_NULL:
return value
elif "literal" in value:
return value["literal"]
def to_values(tokens):
rows = list(tokens)
if len(rows) > 1:
values = [
[get_literal(s["value"]) for s in listwrap(row["select"])] for row in rows
]
if all(flatten(values)):
return {"from": {"literal": values}}
return {"union_all": list(tokens)}
else:
return rows
def to_stack(tokens):
width = tokens["width"]
args = listwrap(tokens["args"])
return Call("stack", args, {"width": width})
def to_array(tokens):
types = list(tokens["type"])
args = list(tokens["args"])
output = Call("create_array", args, {})
if types:
output = Call("cast", [output, Call("array", types, {})], {})
return output
def to_map(tokens):
keys = tokens["keys"]
values = tokens["values"]
return Call("create_map", [keys, values], {})
def to_struct(tokens):
types = list(tokens["types"])
args = list(d for a in tokens["args"] for d in [a if a["name"] else a["value"]])
output = Call("create_struct", args, {})
if types:
output = Call("cast", [output, Call("struct", types, {})], {})
return output
def to_select_call(tokens):
expr = tokens["value"]
if expr == "*":
return ["*"]
try:
call = expr[0][0]
if call.op == "value":
return {"name": tokens["name"], "value": call.args, **call.kwargs}
except:
pass
def to_union_call(tokens):
unions = tokens["union"]
if isinstance(unions, dict):
return unions
elif unions.type.parser_name == "unordered sql":
output = {k: v for k, v in unions.items()} # REMOVE THE Group()
else:
unions = list(unions)
sources = [unions[i] for i in range(0, len(unions), 2)]
operators = ["_".join(unions[i]) for i in range(1, len(unions), 2)]
acc = sources[0]
last_union = None
for op, so in list(zip(operators, sources[1:])):
if op == last_union and "union" in op:
acc[op] = acc[op] + [so]
else:
acc = {op: [acc, so]}
last_union = op
if not tokens["orderby"] and not tokens["offset"] and not tokens["limit"]:
return acc
else:
output = {"from": acc}
output["orderby"] = tokens["orderby"]
output["limit"] = tokens["limit"]
output["offset"] = tokens["offset"]
output["fetch"] = tokens["fetch"]
output["outfile"] = tokens["outfile"]
return output
def to_insert_call(tokens):
options = {
k: v for k, v in tokens.items() if k not in ["columns", "table", "query"]
}
query = tokens["query"]
columns = tokens["columns"]
try:
values = query["from"]["literal"]
if values:
if columns:
data = [dict(zip(columns, row)) for row in values]
return Call("insert", [tokens["table"]], {"values": data, **options})
else:
return Call("insert", [tokens["table"]], {"values": values, **options})
except Exception:
pass
return Call(
"insert", [tokens["table"]], {"columns": columns, "query": query, **options}
)
def to_query(tokens):
output = tokens["query"][0]
try:
output["with"] = tokens["with"]
output["with_recursive"] = tokens["with_recursive"]
return output
except Exception as cause:
return
def to_table(tokens):
output = dict(tokens)
if len(list(output.keys())) > 1:
return output
else:
return output["value"]
def unquote(tokens):
val = tokens[0]
if val.startswith("'") and val.endswith("'"):
val = "'" + val[1:-1].replace("''", "\\'") + "'"
elif val.startswith('"') and val.endswith('"'):
val = '"' + val[1:-1].replace('""', '\\"') + '"'
elif val.startswith("`") and val.endswith("`"):
val = '"' + val[1:-1].replace("``", "`").replace('"', '\\"') + '"'
elif val.startswith("[") and val.endswith("]"):
val = '"' + val[1:-1].replace("]]", "]").replace('"', '\\"') + '"'
elif val.startswith("+"):
val = val[1:]
un = ast.literal_eval(val).replace(".", "\\.")
return un
def to_string(tokens):
val = tokens[0]
val = "'" + val[1:-1].replace("''", "\\'") + "'"
return {"literal": ast.literal_eval(val)}
# NUMBERS
real_num = (
Regex(r"[+-]?(\d+\.\d*|\.\d+)([eE][+-]?\d+)?").set_parser_name("float")
/ (lambda t: float(t[0]))
)
def parse_int(tokens):
if "e" in tokens[0].lower():
return int(float(tokens[0]))
else:
return int(tokens[0])
int_num = Regex(r"[+-]?\d+([eE]\+?\d+)?").set_parser_name("int") / parse_int
hex_num = (
Regex(r"0x[0-9a-fA-F]+").set_parser_name("hex") / (lambda t: {"hex": t[0][2:]})
)
# STRINGS
ansi_string = Regex(r"\'(\'\'|[^'])*\'") / to_string
mysql_doublequote_string = Regex(r'\"(\"\"|[^"])*\"') / to_string
# BASIC IDENTIFIERS
ansi_ident = Regex(r'\"(\"\"|[^"])*\"') / unquote
mysql_backtick_ident = Regex(r"\`(\`\`|[^`])*\`") / unquote
sqlserver_ident = Regex(r"\[(\]\]|[^\]])*\]") / unquote