def insert_value_for_shape_element(
insert_stmt: pgast.InsertStmt, wrapper: pgast.Query,
ir_stmt: irast.MutatingStmt, shape_el: irast.Set,
iterator_id: pgast.OutputVar, *, ptr_info: pg_types.PointerStorageInfo,
ctx: context.CompilerContextLevel) -> pgast.OutputVar:
rel = compile_insert_shape_element(insert_stmt,
wrapper,
ir_stmt,
shape_el,
iterator_id,
ctx=ctx)
insvalue = pathctx.get_path_value_var(rel, shape_el.path_id, env=ctx.env)
if isinstance(insvalue, pgast.TupleVar):
for element in insvalue.elements:
name = element.path_id.rptr_name()
if name == 'std::target':
insvalue = pathctx.get_path_value_var(rel,
element.path_id,
env=ctx.env)
break
else:
raise RuntimeError('could not find std::target in '
'insert computable')
insvalue = pgast.TypeCast(arg=insvalue,
type_name=typecomp.type_node(
ptr_info.column_type))
return insvalue
def compile_TypeRef(expr: irast.Base, *,
ctx: context.CompilerContextLevel) -> pgast.Base:
data_backend = ctx.env.backend
schema = ctx.env.schema
if expr.subtypes:
raise NotImplementedError()
else:
cls = schema.get(expr.maintype)
objtype_id = data_backend.get_objtype_id(cls)
result = pgast.TypeCast(arg=pgast.Constant(val=objtype_id),
type_name=pgast.TypeName(name=('uuid', )))
return result
def process_link_values(ir_stmt,
ir_expr,
target_tab,
tab_cols,
col_data,
dml_rvar,
sources,
props_only,
target_is_scalar,
iterator_cte,
*,
ctx=context.CompilerContext) -> pgast.CommonTableExpr:
"""Unpack data from an update expression into a series of selects.
:param ir_expr:
IR of the INSERT/UPDATE body element.
:param target_tab:
The link table being updated.
:param tab_cols:
A sequence of columns in the table being updated.
:param col_data:
Expressions used to populate well-known columns of the link
table such as std::source and std::__type__.
:param sources:
A list of relations which must be joined into the data query
to resolve expressions in *col_data*.
:param props_only:
Whether this link update only touches link properties.
:param target_is_scalar:
Whether the link target is an ScalarType.
:param iterator_cte:
CTE representing the iterator range in the FOR clause of the
EdgeQL DML statement.
"""
with ctx.newscope() as newscope, newscope.newrel() as subrelctx:
row_query = subrelctx.rel
relctx.include_rvar(row_query, dml_rvar, ctx=subrelctx)
subrelctx.path_scope[ir_stmt.subject.path_id] = row_query
if iterator_cte is not None:
iterator_rvar = dbobj.rvar_for_rel(iterator_cte,
lateral=True,
env=subrelctx.env)
relctx.include_rvar(row_query,
iterator_rvar,
iterator_cte.query.path_id,
aspect='value',
ctx=subrelctx)
with subrelctx.newscope() as sctx, sctx.subrel() as input_rel_ctx:
input_rel = input_rel_ctx.rel
if iterator_cte is not None:
input_rel_ctx.path_scope[iterator_cte.query.path_id] = \
row_query
input_rel_ctx.expr_exposed = False
input_rel_ctx.shape_format = context.ShapeFormat.FLAT
input_rel_ctx.volatility_ref = pathctx.get_path_identity_var(
row_query, ir_stmt.subject.path_id, env=input_rel_ctx.env)
dispatch.compile(ir_expr, ctx=input_rel_ctx)
input_stmt = input_rel
input_rvar = pgast.RangeSubselect(
subquery=input_rel,
lateral=True,
alias=pgast.Alias(aliasname=ctx.env.aliases.get('val')))
row = pgast.ImplicitRowExpr()
source_data = {}
if input_stmt.op is not None:
# UNION
input_stmt = input_stmt.rarg
path_id = ir_expr.path_id
output = pathctx.get_path_value_output(input_stmt, path_id, env=ctx.env)
if isinstance(output, pgast.TupleVar):
for element in output.elements:
name = element.path_id.rptr_name()
if name is None:
name = element.path_id[-1].name
colname = common.edgedb_name_to_pg_name(name)
source_data.setdefault(colname,
dbobj.get_column(input_rvar, element.name))
else:
if target_is_scalar:
target_ref = pathctx.get_rvar_path_value_var(input_rvar,
path_id,
env=ctx.env)
else:
target_ref = pathctx.get_rvar_path_identity_var(input_rvar,
path_id,
env=ctx.env)
source_data['std::target'] = target_ref
if not target_is_scalar and 'std::target' not in source_data:
target_ref = pathctx.get_rvar_path_identity_var(input_rvar,
path_id,
env=ctx.env)
source_data['std::target'] = target_ref
for col in tab_cols:
expr = col_data.get(col)
if expr is None:
expr = source_data.get(col)
if expr is None:
if tab_cols[col]['column_default'] is not None:
expr = pgast.LiteralExpr(expr=tab_cols[col]['column_default'])
else:
expr = pgast.Constant(val=None)
row.args.append(expr)
row_query.target_list = [
pgast.ResTarget(val=pgast.Indirection(arg=pgast.TypeCast(
arg=row, type_name=pgast.TypeName(name=target_tab)),
indirection=[pgast.Star()]))
]
row_query.from_clause += list(sources) + [input_rvar]
link_rows = pgast.CommonTableExpr(query=row_query,
name=ctx.env.aliases.get(hint='r'))
return link_rows
def process_update_body(ir_stmt: irast.MutatingStmt, wrapper: pgast.Query,
update_cte: pgast.CommonTableExpr,
range_cte: pgast.CommonTableExpr, *,
ctx: context.CompilerContextLevel):
"""Generate SQL DML CTEs from an UpdateStmt IR.
:param ir_stmt:
IR of the statement.
:param wrapper:
Top-level SQL query.
:param update_cte:
CTE representing the SQL UPDATE to the main relation of the Object.
:param range_cte:
CTE representing the range affected by the statement.
"""
update_stmt = update_cte.query
external_updates = []
toplevel = ctx.toplevel_stmt
toplevel.ctes.append(range_cte)
toplevel.ctes.append(update_cte)
with ctx.newscope() as subctx:
# It is necessary to process the expressions in
# the UpdateStmt shape body in the context of the
# UPDATE statement so that references to the current
# values of the updated object are resolved correctly.
subctx.path_scope[ir_stmt.subject.path_id] = update_stmt
subctx.rel = update_stmt
subctx.expr_exposed = False
subctx.shape_format = context.ShapeFormat.FLAT
for shape_el in ir_stmt.subject.shape:
with subctx.newscope() as scopectx:
ptrcls = shape_el.rptr.ptrcls
updvalue = shape_el.expr
ptr_info = pg_types.get_pointer_storage_info(
ptrcls,
schema=scopectx.env.schema,
resolve_type=True,
link_bias=False)
# First, process all internal link updates
if ptr_info.table_type == 'ObjectType':
updvalue = pgast.TypeCast(
arg=dispatch.compile(updvalue, ctx=scopectx),
type_name=typecomp.type_node(ptr_info.column_type))
update_stmt.targets.append(
pgast.UpdateTarget(name=ptr_info.column_name,
val=updvalue))
props_only = is_props_only_update(shape_el)
ptr_info = pg_types.get_pointer_storage_info(ptrcls,
resolve_type=False,
link_bias=True)
if ptr_info and ptr_info.table_type == 'link':
external_updates.append((shape_el, props_only))
if not update_stmt.targets:
# No updates directly to the set target table,
# so convert the UPDATE statement into a SELECT.
update_cte.query = pgast.SelectStmt(
ctes=update_stmt.ctes,
target_list=update_stmt.returning_list,
from_clause=[update_stmt.relation] + update_stmt.from_clause,
where_clause=update_stmt.where_clause,
path_namespace=update_stmt.path_namespace,
path_outputs=update_stmt.path_outputs,
path_scope=update_stmt.path_scope,
path_rvar_map=update_stmt.path_rvar_map.copy(),
view_path_id_map=update_stmt.view_path_id_map.copy(),
ptr_join_map=update_stmt.ptr_join_map.copy(),
)
# Process necessary updates to the link tables.
for expr, props_only in external_updates:
if props_only:
process_linkprop_update(ir_stmt,
expr,
wrapper,
update_cte,
ctx=ctx)
else:
process_link_update(ir_stmt,
expr,
False,
wrapper,
update_cte,
None,
ctx=ctx)
def compile_BinOp(expr: irast.Base, *,
ctx: context.CompilerContextLevel) -> pgast.Base:
with ctx.new() as newctx:
newctx.expr_exposed = False
op = expr.op
is_bool_op = op in {ast.ops.AND, ast.ops.OR}
left = dispatch.compile(expr.left, ctx=newctx)
right = dispatch.compile(expr.right, ctx=newctx)
if isinstance(expr.op, ast.ops.TypeCheckOperator):
result = pgast.FuncCall(name=('edgedb', 'issubclass'),
args=[left, right])
if expr.op == ast.ops.IS_NOT:
result = astutils.new_unop(ast.ops.NOT, result)
else:
if not isinstance(expr.left, irast.EmptySet):
left_type = _infer_type(expr.left, ctx=ctx)
else:
left_type = None
if not isinstance(expr.right, irast.EmptySet):
right_type = _infer_type(expr.right, ctx=ctx)
else:
right_type = None
if (not isinstance(expr.left, irast.EmptySet)
and not isinstance(expr.right, irast.EmptySet)):
left_pg_type = pg_types.pg_type_from_object(
ctx.env.schema, left_type, True)
right_pg_type = pg_types.pg_type_from_object(
ctx.env.schema, right_type, True)
if (left_pg_type in {('text', ), ('varchar', )}
and right_pg_type in {('text', ), ('varchar', )}
and op == ast.ops.ADD):
op = '||'
if isinstance(left_type, s_types.Tuple):
left = _tuple_to_row_expr(expr.left, ctx=newctx)
left_count = len(left.args)
else:
left_count = 0
if isinstance(right_type, s_types.Tuple):
right = _tuple_to_row_expr(expr.right, ctx=newctx)
right_count = len(right.args)
else:
right_count = 0
if left_count != right_count:
# Postgres does not allow comparing rows with
# unequal number of entries, but we want to allow
# this. Fortunately, we know that such comparison is
# always False.
result = pgast.Constant(val=False)
else:
if is_bool_op:
# Transform logical operators to force
# the correct behaviour with respect to NULLs.
# See the OrFilterFunction comment for details.
if ctx.clause == 'where':
if expr.op == ast.ops.OR:
result = pgast.FuncCall(name=('edgedb', '_or'),
args=[left, right])
else:
# For the purposes of the WHERE clause,
# AND operator works correctly, as
# it will either return NULL or FALSE,
# which both will disqualify the row.
result = astutils.new_binop(left, right, op=op)
else:
# For expressions outside WHERE, we
# always want the result to be NULL
# if either operand is NULL.
bitop = '&' if expr.op == ast.ops.AND else '|'
bitcond = astutils.new_binop(
lexpr=pgast.TypeCast(
arg=left,
type_name=pgast.TypeName(name=('int', ))),
rexpr=pgast.TypeCast(
arg=right,
type_name=pgast.TypeName(name=('int', ))),
op=bitop)
bitcond = pgast.TypeCast(
arg=bitcond, type_name=pgast.TypeName(name=('bool', )))
result = bitcond
else:
result = astutils.new_binop(left, right, op=op)
return result
def compile_SliceIndirection(expr: irast.Base, *,
ctx: context.CompilerContextLevel) -> pgast.Base:
# Handle Expr[Start:End], where Expr may be std::str or array<T>.
# For strings we translate this into substr calls, whereas
# for arrays the native slice syntax is used.
with ctx.new() as subctx:
subctx.expr_exposed = False
subj = dispatch.compile(expr.expr, ctx=subctx)
start = dispatch.compile(expr.start, ctx=subctx)
stop = dispatch.compile(expr.stop, ctx=subctx)
one = pgast.Constant(val=1)
zero = pgast.Constant(val=0)
is_string = False
arg_type = _infer_type(expr.expr, ctx=ctx)
if isinstance(arg_type, s_scalars.ScalarType):
b = arg_type.get_topmost_concrete_base()
is_string = b.name == 'std::str'
if is_string:
upper_bound = pgast.FuncCall(name=('char_length', ), args=[subj])
else:
upper_bound = pgast.FuncCall(name=('array_upper', ), args=[subj, one])
if astutils.is_null_const(start):
lower = one
else:
lower = start
when_cond = astutils.new_binop(lexpr=lower, rexpr=zero, op=ast.ops.LT)
lower_plus_one = astutils.new_binop(lexpr=lower,
rexpr=one,
op=ast.ops.ADD)
neg_off = astutils.new_binop(lexpr=upper_bound,
rexpr=lower_plus_one,
op=ast.ops.ADD)
when_expr = pgast.CaseWhen(expr=when_cond, result=neg_off)
lower = pgast.CaseExpr(args=[when_expr], defresult=lower_plus_one)
if astutils.is_null_const(stop):
upper = upper_bound
else:
upper = stop
when_cond = astutils.new_binop(lexpr=upper, rexpr=zero, op=ast.ops.LT)
neg_off = astutils.new_binop(lexpr=upper_bound,
rexpr=upper,
op=ast.ops.ADD)
when_expr = pgast.CaseWhen(expr=when_cond, result=neg_off)
upper = pgast.CaseExpr(args=[when_expr], defresult=upper)
if is_string:
lower = pgast.TypeCast(arg=lower,
type_name=pgast.TypeName(name=('int', )))
args = [subj, lower]
if upper is not upper_bound:
for_length = astutils.new_binop(lexpr=upper,
op=ast.ops.SUB,
rexpr=lower)
for_length = astutils.new_binop(lexpr=for_length,
op=ast.ops.ADD,
rexpr=one)
for_length = pgast.TypeCast(
arg=for_length, type_name=pgast.TypeName(name=('int', )))
args.append(for_length)
result = pgast.FuncCall(name=('substr', ), args=args)
else:
indirection = pgast.Indices(lidx=lower, ridx=upper)
result = pgast.Indirection(arg=subj, indirection=[indirection])
return result
def compile_IndexIndirection(expr: irast.Base, *,
ctx: context.CompilerContextLevel) -> pgast.Base:
# Handle Expr[Index], where Expr may be std::str or array<T>.
# For strings we translate this into substr calls, whereas
# for arrays the native slice syntax is used.
is_string = False
arg_type = _infer_type(expr.expr, ctx=ctx)
with ctx.new() as subctx:
subctx.expr_exposed = False
subj = dispatch.compile(expr.expr, ctx=subctx)
index = dispatch.compile(expr.index, ctx=subctx)
if isinstance(arg_type, s_types.Map):
# When we compile maps we always cast keys to text,
# hence we need to cast the index to text here.
index_type = _infer_type(expr.index, ctx=ctx)
index = typecomp.cast(index,
source_type=index_type,
target_type=ctx.env.schema.get('std::str'),
env=ctx.env)
if isinstance(arg_type.element_type, s_types.Array):
return typecomp.cast(astutils.new_binop(lexpr=subj,
op='->',
rexpr=index),
source_type=ctx.env.schema.get('std::json'),
target_type=arg_type.element_type,
env=ctx.env)
elif isinstance(arg_type.element_type, s_types.Map):
return astutils.new_binop(lexpr=subj, op='->', rexpr=index)
else:
return typecomp.cast(astutils.new_binop(lexpr=subj,
op='->>',
rexpr=index),
source_type=ctx.env.schema.get('std::str'),
target_type=arg_type.element_type,
env=ctx.env)
if isinstance(arg_type, s_scalars.ScalarType):
b = arg_type.get_topmost_concrete_base()
is_string = b.name == 'std::str'
one = pgast.Constant(val=1)
zero = pgast.Constant(val=0)
when_cond = astutils.new_binop(lexpr=index, rexpr=zero, op=ast.ops.LT)
index_plus_one = astutils.new_binop(lexpr=index, op=ast.ops.ADD, rexpr=one)
if is_string:
upper_bound = pgast.FuncCall(name=('char_length', ), args=[subj])
else:
upper_bound = pgast.FuncCall(name=('array_upper', ), args=[subj, one])
neg_off = astutils.new_binop(lexpr=upper_bound,
rexpr=index_plus_one,
op=ast.ops.ADD)
when_expr = pgast.CaseWhen(expr=when_cond, result=neg_off)
index = pgast.CaseExpr(args=[when_expr], defresult=index_plus_one)
if is_string:
index = pgast.TypeCast(arg=index,
type_name=pgast.TypeName(name=('int', )))
result = pgast.FuncCall(name=('substr', ), args=[subj, index, one])
else:
indirection = pgast.Indices(ridx=index)
result = pgast.Indirection(arg=subj, indirection=[indirection])
return result
def cast(node: pgast.Base,
*,
source_type: s_obj.Object,
target_type: s_obj.Object,
force: bool = False,
env: context.Environment) -> pgast.Base:
if source_type.name == target_type.name and not force:
return node
schema = env.schema
real_t = schema.get('std::anyreal')
int_t = schema.get('std::anyint')
json_t = schema.get('std::json')
str_t = schema.get('std::str')
datetime_t = schema.get('std::datetime')
bool_t = schema.get('std::bool')
if isinstance(target_type, s_types.Collection):
if target_type.schema_name == 'array':
if source_type.issubclass(json_t):
# If we are casting a jsonb array to array, we do the
# following transformation:
# EdgeQL: <array<T>>MAP_VALUE
# SQL:
# SELECT array_agg(j::T)
# FROM jsonb_array_elements(MAP_VALUE) AS j
inner_cast = cast(pgast.ColumnRef(name=['j']),
source_type=source_type,
target_type=target_type.element_type,
env=env)
return pgast.SelectStmt(
target_list=[
pgast.ResTarget(val=pgast.FuncCall(
name=('array_agg', ), args=[inner_cast]))
],
from_clause=[
pgast.RangeFunction(functions=[
pgast.FuncCall(name=('jsonb_array_elements', ),
args=[node])
],
alias=pgast.Alias(aliasname='j'))
])
else:
# EdgeQL: <array<int64>>['1', '2']
# to SQL: ARRAY['1', '2']::int[]
elem_pgtype = pg_types.pg_type_from_object(
schema, target_type.element_type, topbase=True)
return pgast.TypeCast(arg=node,
type_name=pgast.TypeName(
name=elem_pgtype, array_bounds=[-1]))
elif target_type.schema_name == 'map':
if source_type.issubclass(json_t):
# If the source type is json do nothing, since
# maps are already encoded in json.
return node
# EdgeQL: <map<Tkey,Tval>>MAP<Vkey,Vval>
# to SQL: SELECT jsonb_object_agg(
# key::Vkey::Tkey::text,
# value::Vval::Tval)
# FROM jsonb_each_text(MAP)
key_cast = cast(
cast(
cast(pgast.ColumnRef(name=['key']),
source_type=str_t,
target_type=source_type.key_type,
env=env),
source_type=source_type.key_type,
target_type=target_type.key_type,
env=env,
),
source_type=target_type.key_type,
target_type=str_t,
env=env,
)
target_v_type = target_type.element_type
val_cast = cast(cast(pgast.ColumnRef(name=['value']),
source_type=str_t,
target_type=source_type.element_type,
env=env),
source_type=source_type.element_type,
target_type=target_v_type,
env=env)
map_cast = pgast.SelectStmt(
target_list=[
pgast.ResTarget(
val=pgast.FuncCall(name=('jsonb_object_agg', ),
args=[key_cast, val_cast]))
],
from_clause=[
pgast.RangeFunction(functions=[
pgast.FuncCall(name=('jsonb_each_text', ), args=[node])
])
])
return pgast.FuncCall(
name=('coalesce', ),
args=[
map_cast,
pgast.TypeCast(arg=pgast.Constant(val='{}'),
type_name=pgast.TypeName(name=('jsonb', )))
])
else:
# `target_type` is not a collection.
if (source_type.issubclass(datetime_t)
and target_type.issubclass(str_t)):
# Normalize datetime to text conversion to have the same
# format as one would get by serializing to JSON.
#
# EdgeQL: <text><datetime>'2010-10-10';
# To SQL: trim(to_json('2010-01-01'::timestamptz)::text, '"')
return pgast.FuncCall(
name=('trim', ),
args=[
pgast.TypeCast(arg=pgast.FuncCall(name=('to_json', ),
args=[node]),
type_name=pgast.TypeName(name=('text', ))),
pgast.Constant(val='"')
])
elif source_type.issubclass(bool_t) and target_type.issubclass(int_t):
# PostgreSQL 9.6 doesn't allow to cast 'boolean' to any integer
# other than int32:
# SELECT 'true'::boolean::bigint;
# ERROR: cannot cast type boolean to bigint
# So we transform EdgeQL: <int64>BOOL
# to SQL: BOOL::int::<targetint>
return pgast.TypeCast(
arg=pgast.TypeCast(arg=node,
type_name=pgast.TypeName(name=('int', ))),
type_name=pgast.TypeName(
name=pg_types.pg_type_from_scalar(schema, target_type)))
elif source_type.issubclass(int_t) and target_type.issubclass(bool_t):
# PostgreSQL 9.6 doesn't allow to cast any integer other
# than int32 to 'boolean':
# SELECT 1::bigint::boolean;
# ERROR: cannot cast type bigint to boolea
# So we transform EdgeQL: <boolean>INT
# to SQL: (INT != 0)
return astutils.new_binop(node,
pgast.Constant(val=0),
op=ast.ops.NE)
elif source_type.issubclass(json_t):
if (target_type.issubclass(real_t)
or target_type.issubclass(bool_t)):
# Simply cast to text and the to the target type.
return cast(cast(node,
source_type=source_type,
target_type=str_t,
env=env),
source_type=str_t,
target_type=target_type,
env=env)
elif target_type.issubclass(str_t):
# It's not possible to cast jsonb string to text directly,
# so we do a trick:
# EdgeQL: <str>JSONB_VAL
# SQL: array_to_json(ARRAY[JSONB_VAL])->>0
return astutils.new_binop(pgast.FuncCall(
name=('array_to_json', ),
args=[pgast.ArrayExpr(elements=[node])]),
pgast.Constant(val=0),
op='->>')
elif target_type.issubclass(json_t):
return pgast.TypeCast(
arg=node, type_name=pgast.TypeName(name=('jsonb', )))
else:
const_type = pg_types.pg_type_from_object(schema,
target_type,
topbase=True)
return pgast.TypeCast(arg=node,
type_name=pgast.TypeName(name=const_type))
raise RuntimeError(
f'could not cast {source_type.name} to {target_type.name}')