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/engine/ast.py

104 lines
3.0 KiB

from typing import List
import uuid
class TableInfo:
def __init__(self, table_name, cols, cxt:'Context'):
# statics
self.table_name = table_name
self.columns_byname = dict() # column_name, type
self.columns = []
for c in cols:
k9name = self.table_name + c['name']
if k9name in cxt.k9cols_byname: # duplicate names?
root = cxt.k9cols_byname[k9name]
k9name = k9name + root[3]
root[3] += 1
# column: (k9name, type, original col_object, dup_count)
col_object = (k9name, (list(c['type'].keys()))[0], c, 1)
cxt.k9cols_byname[k9name] = col_object
self.columns_byname[c['name']] = col_object
self.columns.append(col_object)
# runtime
self.n_rows = 0 # number of cols
self.order = [] # assumptions
cxt.tables_byname[self.table_name] = self # construct reverse map
@property
def n_cols(self):
return len(self.columns)
def get_k9colname(self, col_name):
return self.columns_byname[col_name][0]
def parse_tablenames(self, str):
# TODO: deal with alias
return self.get_k9colname(str)
class Context:
def __init__(self):
self.tables:List[TableInfo] = []
self.tables_byname = dict()
self.k9cols_byname = dict()
self.udf_map = dict()
self.k9code = ''
def add_table(self, table_name, cols):
tbl = TableInfo(table_name, cols, self)
self.tables.append(tbl)
return tbl
def gen_tmptable(self):
from engine.utils import base62uuid
return f'tmp{base62uuid()}'
def emit(self, codelet):
self.k9code += codelet + '\n'
def emit_nonewline(self, codelet):
self.k9code += codelet
def __str__(self):
return self.k9code
class ast_node:
types = dict()
def __init__(self, parent:"ast_node", node, context:Context = None):
self.context = parent.context if context is None else context
self.init(node)
self.produce(node)
self.spawn(node)
self.consume(node)
def emit(self, code):
self.context.emit(code)
def emit_no_ln(self, code):
self.context.emit_nonewline(code)
name = 'null'
# each ast node has 3 stages.
# `produce' generates info for child nodes
# `spawn' populates child nodes
# `consume' consumes info from child nodes and finalizes codegen
# For simple operators, there may not be need for some of these stages
def init(self, _):
pass
def produce(self, _):
pass
def spawn(self, _):
pass
def consume(self, _):
pass
# include classes in module as first order operators
def include(objs):
import inspect
for _, cls in inspect.getmembers(objs):
if inspect.isclass(cls) and issubclass(cls, ast_node):
ast_node.types[cls.name] = cls