"""Utility functions and classes for supporting query conditions. Classes: `CompileCondition` Container for a compiled condition. Functions: `compile_condition` Compile a condition and extract usable index conditions. `call_on_recarr` Evaluate a function over a structured array. """ from __future__ import annotations import re from typing import Any, TYPE_CHECKING from collections.abc import Callable, Iterable import numpy as np import numexpr as ne from .utils import lazyattr from .utilsextension import get_nested_field if TYPE_CHECKING: from .table import Column _no_matching_opcode = re.compile(r"[^a-z]([a-z]+)_([a-z]+)[^a-z]") # E.g. "gt" and "bfc" from "couldn't find matching opcode for 'gt_bfc'". def _unsupported_operation_error(exception: Exception) -> Exception: r"""Make the \"no matching opcode\" Numexpr `exception` more clear. A new exception of the same kind is returned. """ message = exception.args[0] op, types = _no_matching_opcode.search(message).groups() newmessage = "unsupported operand types for *%s*: " % op newmessage += ", ".join( ne.necompiler.typecode_to_kind[t] for t in types[1:] ) return exception.__class__(newmessage) def _check_indexable_cmp( getidxcmp: Callable[ [ne.expressions.ExpressionNode, frozenset[str]], tuple[Any, str, Any] ], ) -> Callable[ [ne.expressions.ExpressionNode, frozenset[str]], tuple[Any, str, Any] ]: """Decorate `getidxcmp` to check the returned indexable comparison. This does some extra checking that Numexpr would perform later on the comparison if it was compiled within a complete condition. """ def newfunc( exprnode: ne.expressions.ExpressionNode, indexedcols: frozenset[str], ) -> tuple[Any, str, Any]: result = getidxcmp(exprnode, indexedcols) if result[0] is not None: try: ne.necompiler.typeCompileAst( ne.necompiler.expressionToAST(exprnode) ) except NotImplementedError as nie: # Try to make this Numexpr error less cryptic. raise _unsupported_operation_error(nie) return result newfunc.__name__ = getidxcmp.__name__ newfunc.__doc__ = getidxcmp.__doc__ return newfunc @_check_indexable_cmp def _get_indexable_cmp( exprnode: ne.expressions.ExpressionNode, indexedcols: frozenset[str], ) -> tuple[Any, str, Any] | tuple[None, None, None]: """Get the indexable variable-constant comparison in `exprnode`. A tuple of (variable, operation, constant) is returned if `exprnode` is a variable-constant (or constant-variable) comparison, and the variable is in `indexedcols`. A normal variable can also be used instead of a constant: a tuple with its name will appear instead of its value. Otherwise, the values in the tuple are ``None``. """ not_indexable = (None, None, None) turncmp = { "lt": "gt", "le": "ge", "eq": "eq", "ge": "le", "gt": "lt", } def get_cmp( var: ne.expressions.ExpressionNode, const: ne.expressions.ExpressionNode, op: str, ) -> tuple[Any, str, Any] | None: var_value, const_value = var.value, const.value if ( var.astType == "variable" and var_value in indexedcols and const.astType in ["constant", "variable"] ): if const.astType == "variable": const_value = (const_value,) return (var_value, op, const_value) return None def is_indexed_boolean(node: ne.expressions.ExpressionNode) -> bool: return ( node.astType == "variable" and node.astKind == "bool" and node.value in indexedcols ) # Boolean variables are indexable by themselves. if is_indexed_boolean(exprnode): return (exprnode.value, "eq", True) # And so are negations of boolean variables. if exprnode.astType == "op" and exprnode.value == "invert": child = exprnode.children[0] if is_indexed_boolean(child): return (child.value, "eq", False) # A negation of an expression will be returned as ``~child``. # The indexability of the negated expression will be decided later on. if child.astKind == "bool": return (child, "invert", None) # Check node type. Only comparisons are indexable from now on. if exprnode.astType != "op": return not_indexable cmpop = exprnode.value if cmpop not in turncmp: return not_indexable # Look for a variable-constant comparison in both directions. left, right = exprnode.children cmp_ = get_cmp(left, right, cmpop) if cmp_: return cmp_ cmp_ = get_cmp(right, left, turncmp[cmpop]) if cmp_: return cmp_ return not_indexable def _equiv_expr_node( x: Any | ne.expressions.ExpressionNode, y: Any | ne.expressions.ExpressionNode, ) -> bool: """Return Truen whether two ExpressionNodes are equivalent. This is needed because '==' is overridden on ExpressionNode to return a new ExpressionNode. """ if not isinstance(x, ne.expressions.ExpressionNode) and not isinstance( y, ne.expressions.ExpressionNode ): return x == y elif ( type(x) is not type(y) or not isinstance(x, ne.expressions.ExpressionNode) or not isinstance(y, ne.expressions.ExpressionNode) or x.value != y.value or x.astKind != y.astKind or len(x.children) != len(y.children) ): return False for xchild, ychild in zip(x.children, y.children): if not _equiv_expr_node(xchild, ychild): return False return True def _get_idx_expr_recurse( exprnode: ne.expressions.ExpressionNode, indexedcols: frozenset[str], idxexprs: list, strexpr: list[str], ) -> ( list[tuple[Any, tuple[str], tuple[Any]]] | list[ tuple[ne.expressions.ExpressionNode, tuple[str, str], tuple[Any, Any]] ] | tuple[list, list[str]] ): """Actual implementation of the `get_idx_expr()` wrapper. 'idxexprs' is a list of expressions in the form ``(var, (ops), (limits))``. 'strexpr' is the indexable expression in string format. These parameters will be received empty (i.e. [], ['']) for the first time and populated during the different recursive calls. Finally, they are returned in the last level to the original wrapper. If 'exprnode' is not indexable, it will return the tuple ([], ['']) so as to signal this. """ not_indexable = ([], [""]) op_conv = { "and": "&", "or": "|", "not": "~", } negcmp = { "lt": "ge", "le": "gt", "ge": "lt", "gt": "le", } def fix_invert( idxcmp: tuple[Any, str, Any] | tuple[None, None, None], exprnode: ne.expressions.ExpressionNode, indexedcols: frozenset[str], ) -> tuple[ tuple[Any, str, Any] | tuple[None, None, None], ne.expressions.ExpressionNode, bool, ]: invert = False # Loop until all leading negations have been dealt with while idxcmp[1] == "invert": invert ^= True # The information about the negated node is in first position exprnode = idxcmp[0] idxcmp = _get_indexable_cmp(exprnode, indexedcols) return idxcmp, exprnode, invert # Indexable variable-constant comparison. idxcmp = _get_indexable_cmp(exprnode, indexedcols) idxcmp, exprnode, invert = fix_invert(idxcmp, exprnode, indexedcols) if idxcmp[0]: if invert: var, op, value = idxcmp if op == "eq" and value in [True, False]: # ``var`` must be a boolean index. Flip its value. value ^= True else: op = negcmp[op] expr = (var, (op,), (value,)) invert = False else: expr = (idxcmp[0], (idxcmp[1],), (idxcmp[2],)) return [expr] # For now negations of complex expressions will be not supported as # forming part of an indexable condition. This might be supported in # the future. if invert: return not_indexable # Only conjunctions and disjunctions of comparisons are considered # for the moment. if exprnode.astType != "op" or exprnode.value not in ["and", "or"]: return not_indexable left, right = exprnode.children # Get the expression at left lcolvar, lop, llim = _get_indexable_cmp(left, indexedcols) # Get the expression at right rcolvar, rop, rlim = _get_indexable_cmp(right, indexedcols) # Use conjunction of indexable VC comparisons like # ``(a <[=] x) & (x <[=] b)`` or ``(a >[=] x) & (x >[=] b)`` # as ``a <[=] x <[=] b``, for the moment. op = exprnode.value if ( lcolvar is not None and rcolvar is not None and _equiv_expr_node(lcolvar, rcolvar) and op == "and" ): if lop in ["gt", "ge"] and rop in ["lt", "le"]: # l <= x <= r expr = (lcolvar, (lop, rop), (llim, rlim)) return [expr] if lop in ["lt", "le"] and rop in ["gt", "ge"]: # l >= x >= r expr = (rcolvar, (rop, lop), (rlim, llim)) return [expr] # Recursively get the expressions at the left and the right lexpr = _get_idx_expr_recurse(left, indexedcols, idxexprs, strexpr) rexpr = _get_idx_expr_recurse(right, indexedcols, idxexprs, strexpr) def add_expr(expr, idxexprs: list, strexpr: list[str]) -> None: """Add a single expression to the list.""" if isinstance(expr, list): # expr is a single expression idxexprs.append(expr[0]) lenexprs = len(idxexprs) # Mutate the strexpr string if lenexprs == 1: strexpr[:] = ["e0"] else: strexpr[:] = [ "(%s %s e%d)" % (strexpr[0], op_conv[op], lenexprs - 1) ] # Add expressions to the indexable list when they are and'ed, or # they are both indexable. if lexpr != not_indexable and (op == "and" or rexpr != not_indexable): add_expr(lexpr, idxexprs, strexpr) if rexpr != not_indexable: add_expr(rexpr, idxexprs, strexpr) return (idxexprs, strexpr) if rexpr != not_indexable and op == "and": add_expr(rexpr, idxexprs, strexpr) return (idxexprs, strexpr) # Can not use indexed column. return not_indexable def _get_idx_expr( expr: ne.expressions.ExpressionNode, indexedcols: frozenset[str], ) -> tuple[list[tuple[Any, tuple[str], tuple[Any]]], list[str]]: """Extract an indexable expression out of `exprnode`. Looks for variable-constant comparisons in the expression node `exprnode` involving variables in `indexedcols`. It returns a tuple of (idxexprs, strexpr) where 'idxexprs' is a list of expressions in the form ``(var, (ops), (limits))`` and 'strexpr' is the indexable expression in string format. Expressions such as ``0 < c1 <= 1`` do not work as expected. Right now only some of the *indexable comparisons* are considered: * ``a <[=] x``, ``a == x`` and ``a >[=] x`` * ``(a <[=] x) & (y <[=] b)`` and ``(a == x) | (b == y)`` * ``~(~c_bool)``, ``~~c_bool`` and ``~(~c_bool) & (c_extra != 2)`` (where ``a``, ``b`` and ``c_bool`` are indexed columns, but ``c_extra`` is not) Particularly, the ``!=`` operator and negations of complex boolean expressions are *not considered* as valid candidates: * ``a != 1`` and ``c_bool != False`` * ``~((a > 0) & (c_bool))`` """ return _get_idx_expr_recurse(expr, indexedcols, [], [""]) class CompiledCondition: """Container for a compiled condition.""" @lazyattr def index_variables(self) -> frozenset: """Columns participating in the index expression.""" idxexprs = self.index_expressions idxvars = [] for expr in idxexprs: idxvar = expr[0] if idxvar not in idxvars: idxvars.append(idxvar) return frozenset(idxvars) def __init__( self, func: ne.interpreter.NumExpr, params: list[str], idxexprs: list[tuple[Any, tuple[str, ...], Any]], strexpr: str, **kwargs, ) -> None: self.function = func """The compiled function object corresponding to this condition.""" self.parameters = params """A list of parameter names for this condition.""" self.index_expressions = idxexprs """A list of expressions in the form ``(var, (ops), (limits))``.""" self.string_expression = strexpr """The indexable expression in string format.""" self.kwargs = kwargs """NumExpr kwargs (used to pass ex_uses_vml to numexpr)""" def __repr__(self) -> str: return f"""idxexprs: {self.index_expressions} strexpr: {self.string_expression} idxvars: {self.index_variables}""" def with_replaced_vars( self, condvars: dict[str, Column | np.ndarray] ) -> CompiledCondition: """Replace index limit variables with their values in-place. A new compiled condition is returned. Values are taken from the `condvars` mapping and converted to Python scalars. """ exprs = self.index_expressions exprs2 = [] for expr in exprs: idxlims = expr[2] # the limits are in third place limit_values = [] for idxlim in idxlims: if isinstance(idxlim, tuple): # variable idxlim = condvars[idxlim[0]] # look up value idxlim = idxlim.tolist() # convert back to Python limit_values.append(idxlim) # Add this replaced entry to the new exprs2 var, ops, _ = expr exprs2.append((var, ops, tuple(limit_values))) # Create a new container for the converted values newcc = CompiledCondition( self.function, self.parameters, exprs2, self.string_expression, **self.kwargs, ) return newcc def _get_variable_names( expression: ne.expressions.ExpressionNode, ) -> list[str]: """Return the list of variable names in the Numexpr `expression`.""" names = [] stack = [expression] while stack: node = stack.pop() if node.astType == "variable": names.append(node.value) elif hasattr(node, "children"): stack.extend(node.children) return list(set(names)) # remove repeated names def compile_condition( condition: str, typemap: dict[str, type], indexedcols: frozenset[str] ) -> CompiledCondition: """Compile a condition and extract usable index conditions. Looks for variable-constant comparisons in the `condition` string involving the indexed columns whose variable names appear in `indexedcols`. The part of `condition` having usable indexes is returned as a compiled condition in a `CompiledCondition` container. Expressions such as '0 < c1 <= 1' do not work as expected. The Numexpr types of *all* variables must be given in the `typemap` mapping. The ``function`` of the resulting `CompiledCondition` instance is a Numexpr function object, and the ``parameters`` list indicates the order of its parameters. """ # Get the expression tree and extract index conditions. expr = ne.necompiler.stringToExpression(condition, typemap, {}) if expr.astKind != "bool": raise TypeError( "condition ``%s`` does not have a boolean type" % condition ) idxexprs = _get_idx_expr(expr, indexedcols) # Post-process the answer if isinstance(idxexprs, list): # Simple expression strexpr = ["e0"] else: # Complex expression idxexprs, strexpr = idxexprs # Get rid of the unnecessary list wrapper for strexpr strexpr = strexpr[0] # Get the variable names used in the condition. # At the same time, build its signature. varnames = _get_variable_names(expr) signature = [(var, typemap[var]) for var in varnames] try: # See the comments in `numexpr.evaluate()` for the # reasons of inserting copy operators for unaligned, # *unidimensional* arrays. func = ne.necompiler.NumExpr(expr, signature) except NotImplementedError as nie: # Try to make this Numexpr error less cryptic. raise _unsupported_operation_error(nie) _, ex_uses_vml = ne.necompiler.getExprNames(condition, {}) kwargs = {"ex_uses_vml": ex_uses_vml} params = varnames # This is more comfortable to handle about than a tuple. return CompiledCondition(func, params, idxexprs, strexpr, **kwargs) def call_on_recarr( func: Callable, params: Iterable, recarr: np.ndarray, param2arg: Callable[[Any], Any] | None = None, **kwargs, ) -> None: """Call `func` with `params` over `recarr`. The `param2arg` function, when specified, is used to get an argument given a parameter name; otherwise, the parameter itself is used as an argument. When the argument is a `Column` object, the proper column from `recarr` is used as its value. """ args = [] for param in params: if param2arg: arg = param2arg(param) else: arg = param if hasattr(arg, "pathname"): # looks like a column arg = get_nested_field(recarr, arg.pathname) args.append(arg) return func(*args, **kwargs)