def new_empty_rvar(ir_set: irast.EmptySet, *,
ctx: context.CompilerContextLevel) -> pgast.BaseRangeVar:
nullref_alias = ctx.env.aliases.get('e')
val = typecomp.cast(pgast.Constant(val=None, nullable=True),
source_type=ir_set.scls,
target_type=ir_set.scls,
force=True,
env=ctx.env)
nullrel = pgast.SelectStmt(
target_list=[pgast.ResTarget(val=val, name=nullref_alias)],
nullable=True)
rvar = dbobj.rvar_for_rel(nullrel, env=ctx.env)
rvar.path_scope.add(ir_set.path_id)
rvar.value_scope.add(ir_set.path_id)
null_ref = pgast.ColumnRef(name=[nullref_alias], nullable=True)
pathctx.put_rvar_path_output(rvar,
ir_set.path_id,
aspect='value',
var=null_ref,
env=ctx.env)
if ir_set.path_id.is_objtype_path():
pathctx.put_rvar_path_output(rvar,
ir_set.path_id,
aspect='identity',
var=null_ref,
env=ctx.env)
return rvar
def tuple_var_as_json_object(tvar, *, env):
if not tvar.named:
return pgast.FuncCall(name=('jsonb_build_array', ),
args=[
serialize_expr(t.val, nested=True, env=env)
for t in tvar.elements
],
null_safe=True,
nullable=tvar.nullable)
else:
keyvals = []
for element in tvar.elements:
rptr = element.path_id.rptr()
if rptr is None:
name = element.path_id[-1].name.name
else:
name = rptr.shortname.name
if rptr.is_link_property():
name = '@' + name
keyvals.append(pgast.Constant(val=name))
if isinstance(element.val, pgast.TupleVar):
val = serialize_expr(element.val, env=env)
else:
val = element.val
keyvals.append(val)
return pgast.FuncCall(name=('jsonb_build_object', ),
args=keyvals,
null_safe=True,
nullable=tvar.nullable)
def _compile_set_in_singleton_mode(
node: irast.Set, *, ctx: context.CompilerContextLevel) -> pgast.Base:
if isinstance(node, irast.EmptySet):
return pgast.Constant(value=None)
elif node.expr is not None:
return dispatch.compile(node.expr, ctx=ctx)
else:
if node.rptr:
ptrcls = node.rptr.ptrcls
source = node.rptr.source
if not ptrcls.is_link_property():
if source.rptr:
raise RuntimeError('unexpectedly long path in simple expr')
colref = pgast.ColumnRef(
name=[common.edgedb_name_to_pg_name(ptrcls.shortname)])
elif isinstance(node.scls, s_scalars.ScalarType):
colref = pgast.ColumnRef(
name=[common.edgedb_name_to_pg_name(node.scls.name)])
else:
colref = pgast.ColumnRef(
name=[common.edgedb_name_to_pg_name(node.scls.name)])
return colref
def compile_Constant(expr: irast.Base, *,
ctx: context.CompilerContextLevel) -> pgast.Base:
result = pgast.Constant(val=expr.value)
result = typecomp.cast(result,
source_type=expr.type,
target_type=expr.type,
force=True,
env=ctx.env)
return result
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 new_static_class_rvar(
ir_set: irast.Set,
*,
lateral: bool = True,
ctx: context.CompilerContextLevel) -> pgast.BaseRangeVar:
set_rvar = new_root_rvar(ir_set, ctx=ctx)
clsname = pgast.Constant(val=ir_set.rptr.source.scls.material_type().name)
nameref = dbobj.get_column(set_rvar,
common.edgedb_name_to_pg_name('schema::name'))
condition = astutils.new_binop(nameref, clsname, op='=')
substmt = pgast.SelectStmt()
include_rvar(substmt, set_rvar, ir_set.path_id, aspect='value', ctx=ctx)
substmt.where_clause = astutils.extend_binop(substmt.where_clause,
condition)
return new_rel_rvar(ir_set, substmt, ctx=ctx)
def get_path_output_or_null(
rel: pgast.Query, path_id: irast.PathId, *,
aspect: str, env: context.Environment) -> \
typing.Tuple[pgast.OutputVar, bool]:
try:
ref = get_path_output(rel, path_id, aspect=aspect, env=env)
is_null = False
except LookupError:
alias = env.aliases.get('null')
restarget = pgast.ResTarget(name=alias, val=pgast.Constant(val=None))
if hasattr(rel, 'returning_list'):
rel.returning_list.append(restarget)
else:
rel.target_list.append(restarget)
is_null = True
ref = pgast.ColumnRef(name=[alias], nullable=True)
return ref, is_null
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_link_update(
ir_stmt: irast.MutatingStmt, ir_expr: irast.Base, props_only: bool,
wrapper: pgast.Query, dml_cte: pgast.CommonTableExpr,
iterator_cte: pgast.CommonTableExpr, *,
ctx: context.CompilerContextLevel) -> typing.Optional[pgast.Query]:
"""Perform updates to a link relation as part of a DML statement.
:param ir_stmt:
IR of the statement.
:param ir_expr:
IR of the INSERT/UPDATE body element.
:param props_only:
Whether this link update only touches link properties.
:param wrapper:
Top-level SQL query.
:param dml_cte:
CTE representing the SQL INSERT or UPDATE to the main
relation of the Object.
:param iterator_cte:
CTE representing the iterator range in the FOR clause of the
EdgeQL DML statement.
"""
toplevel = ctx.toplevel_stmt
edgedb_ptr_tab = pgast.RangeVar(
relation=pgast.Relation(schemaname='edgedb', name='pointer'),
alias=pgast.Alias(aliasname=ctx.env.aliases.get(hint='ptr')))
ltab_alias = edgedb_ptr_tab.alias.aliasname
rptr = ir_expr.rptr
ptrcls = rptr.ptrcls
target_is_scalar = isinstance(ptrcls.target, s_scalars.ScalarType)
path_id = rptr.source.path_id.extend(ptrcls, rptr.direction,
rptr.target.scls)
# The links in the dml class shape have been derived,
# but we must use the correct specialized link class for the
# base material type.
mptrcls = ptrcls.material_type()
# Lookup link class id by link name.
lname_to_id = pgast.CommonTableExpr(query=pgast.SelectStmt(
from_clause=[edgedb_ptr_tab],
target_list=[
pgast.ResTarget(val=pgast.ColumnRef(name=[ltab_alias, 'id']))
],
where_clause=astutils.new_binop(
lexpr=pgast.ColumnRef(name=[ltab_alias, 'name']),
rexpr=pgast.Constant(val=mptrcls.name),
op=ast.ops.EQ)),
name=ctx.env.aliases.get(hint='lid'))
lname_to_id_rvar = pgast.RangeVar(relation=lname_to_id)
toplevel.ctes.append(lname_to_id)
target_rvar = dbobj.range_for_ptrcls(mptrcls,
'>',
include_overlays=False,
env=ctx.env)
target_alias = target_rvar.alias.aliasname
if target_is_scalar:
target_tab_name = (target_rvar.relation.schemaname,
target_rvar.relation.name)
else:
target_tab_name = common.link_name_to_table_name(mptrcls.shortname,
catenate=False)
tab_cols = \
ctx.env.backend._type_mech.get_cached_table_columns(target_tab_name)
assert tab_cols, "could not get cols for {!r}".format(target_tab_name)
dml_cte_rvar = pgast.RangeVar(
relation=dml_cte,
alias=pgast.Alias(aliasname=ctx.env.aliases.get('m')))
col_data = {
'ptr_item_id':
pgast.ColumnRef(name=[lname_to_id.name, 'id']),
'std::source':
pathctx.get_rvar_path_identity_var(dml_cte_rvar,
ir_stmt.subject.path_id,
env=ctx.env)
}
# Drop all previous link records for this source.
delcte = pgast.CommonTableExpr(query=pgast.DeleteStmt(
relation=target_rvar,
where_clause=astutils.new_binop(
lexpr=col_data['std::source'],
op=ast.ops.EQ,
rexpr=pgast.ColumnRef(name=[target_alias, 'std::source'])),
using_clause=[dml_cte_rvar],
returning_list=[
pgast.ResTarget(val=pgast.ColumnRef(
name=[target_alias, pgast.Star()]))
]),
name=ctx.env.aliases.get(hint='d'))
pathctx.put_path_value_rvar(delcte.query,
path_id.ptr_path(),
target_rvar,
env=ctx.env)
# Record the effect of this removal in the relation overlay
# context to ensure that the RETURNING clause potentially
# referencing this link yields the expected results.
overlays = ctx.env.rel_overlays[ptrcls.shortname]
overlays.append(('except', delcte))
toplevel.ctes.append(delcte)
# Turn the IR of the expression on the right side of :=
# into a subquery returning records for the link table.
data_cte = process_link_values(ir_stmt,
ir_expr,
target_tab_name,
tab_cols,
col_data,
dml_cte_rvar, [lname_to_id_rvar],
props_only,
target_is_scalar,
iterator_cte,
ctx=ctx)
toplevel.ctes.append(data_cte)
data_select = pgast.SelectStmt(
target_list=[
pgast.ResTarget(val=pgast.ColumnRef(
name=[data_cte.name, pgast.Star()]))
],
from_clause=[pgast.RangeVar(relation=data_cte)])
# Inserting rows into the link table may produce cardinality
# constraint violations, since the INSERT into the link table
# is executed in the snapshot where the above DELETE from
# the link table is not visible. Hence, we need to use
# the ON CONFLICT clause to resolve this.
conflict_cols = ['std::source', 'std::target', 'ptr_item_id']
conflict_inference = []
conflict_exc_row = []
for col in conflict_cols:
conflict_inference.append(pgast.ColumnRef(name=[col]))
conflict_exc_row.append(pgast.ColumnRef(name=['excluded', col]))
conflict_data = pgast.SelectStmt(
target_list=[
pgast.ResTarget(val=pgast.ColumnRef(
name=[data_cte.name, pgast.Star()]))
],
from_clause=[pgast.RangeVar(relation=data_cte)],
where_clause=astutils.new_binop(
lexpr=pgast.ImplicitRowExpr(args=conflict_inference),
rexpr=pgast.ImplicitRowExpr(args=conflict_exc_row),
op='='))
cols = [pgast.ColumnRef(name=[col]) for col in tab_cols]
updcte = pgast.CommonTableExpr(
name=ctx.env.aliases.get(hint='i'),
query=pgast.InsertStmt(
relation=target_rvar,
select_stmt=data_select,
cols=cols,
on_conflict=pgast.OnConflictClause(
action='update',
infer=pgast.InferClause(index_elems=conflict_inference),
target_list=[
pgast.MultiAssignRef(columns=cols, source=conflict_data)
]),
returning_list=[
pgast.ResTarget(val=pgast.ColumnRef(name=[pgast.Star()]))
]))
pathctx.put_path_value_rvar(updcte.query,
path_id.ptr_path(),
target_rvar,
env=ctx.env)
# Record the effect of this insertion in the relation overlay
# context to ensure that the RETURNING clause potentially
# referencing this link yields the expected results.
overlays = ctx.env.rel_overlays[ptrcls.shortname]
overlays.append(('union', updcte))
toplevel.ctes.append(updcte)
return data_cte
def process_insert_body(ir_stmt: irast.MutatingStmt, wrapper: pgast.Query,
insert_cte: pgast.CommonTableExpr,
insert_rvar: pgast.BaseRangeVar, *,
ctx: context.CompilerContextLevel) -> None:
"""Generate SQL DML CTEs from an InsertStmt IR.
:param ir_stmt:
IR of the statement.
:param wrapper:
Top-level SQL query.
:param insert_cte:
CTE representing the SQL INSERT to the main relation of the Object.
"""
cols = [pgast.ColumnRef(name=['std::__type__'])]
select = pgast.SelectStmt(target_list=[])
values = select.target_list
# The main INSERT query of this statement will always be
# present to insert at least the std::id and std::__type__
# links.
insert_stmt = insert_cte.query
insert_stmt.cols = cols
insert_stmt.select_stmt = select
if ir_stmt.parent_stmt is not None:
iterator_set = ir_stmt.parent_stmt.iterator_stmt
else:
iterator_set = None
if iterator_set is not None:
with ctx.substmt() as ictx:
ictx.path_scope = ictx.path_scope.new_child()
ictx.path_scope[iterator_set.path_id] = ictx.rel
clauses.compile_iterator_expr(ictx.rel, iterator_set, ctx=ictx)
ictx.rel.path_id = iterator_set.path_id
pathctx.put_path_bond(ictx.rel, iterator_set.path_id)
iterator_cte = pgast.CommonTableExpr(
query=ictx.rel, name=ctx.env.aliases.get('iter'))
ictx.toplevel_stmt.ctes.append(iterator_cte)
iterator_rvar = dbobj.rvar_for_rel(iterator_cte, env=ctx.env)
relctx.include_rvar(select,
iterator_rvar,
path_id=ictx.rel.path_id,
ctx=ctx)
iterator_id = pathctx.get_path_identity_var(select,
iterator_set.path_id,
env=ctx.env)
else:
iterator_cte = None
iterator_id = None
values.append(
pgast.ResTarget(val=pgast.SelectStmt(
target_list=[pgast.ResTarget(val=pgast.ColumnRef(name=['id']))],
from_clause=[
pgast.RangeVar(relation=pgast.Relation(name='objecttype',
schemaname='edgedb'))
],
where_clause=astutils.new_binop(
op=ast.ops.EQ,
lexpr=pgast.ColumnRef(name=['name']),
rexpr=pgast.Constant(val=ir_stmt.subject.scls.shortname)))))
external_inserts = []
tuple_elements = []
parent_link_props = []
with ctx.newrel() as subctx:
subctx.rel = select
subctx.rel_hierarchy[select] = insert_stmt
subctx.expr_exposed = False
subctx.shape_format = context.ShapeFormat.FLAT
if iterator_cte is not None:
subctx.path_scope = ctx.path_scope.new_child()
subctx.path_scope[iterator_cte.query.path_id] = select
# Process the Insert IR and separate links that go
# into the main table from links that are inserted into
# a separate link table.
for shape_el in ir_stmt.subject.shape:
rptr = shape_el.rptr
ptrcls = rptr.ptrcls.material_type()
if (ptrcls.is_link_property()
and rptr.source.path_id != ir_stmt.subject.path_id):
parent_link_props.append(shape_el)
continue
ptr_info = pg_types.get_pointer_storage_info(
ptrcls,
schema=subctx.env.schema,
resolve_type=True,
link_bias=False)
props_only = False
# First, process all local link inserts.
if ptr_info.table_type == 'ObjectType':
props_only = True
field = pgast.ColumnRef(name=[ptr_info.column_name])
cols.append(field)
insvalue = insert_value_for_shape_element(insert_stmt,
wrapper,
ir_stmt,
shape_el,
iterator_id,
ptr_info=ptr_info,
ctx=subctx)
tuple_el = astutils.tuple_element_for_shape_el(shape_el, field)
tuple_elements.append(tuple_el)
values.append(pgast.ResTarget(val=insvalue))
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_inserts.append((shape_el, props_only))
if iterator_cte is not None:
cols.append(pgast.ColumnRef(name=['__edb_token']))
values.append(pgast.ResTarget(val=iterator_id))
pathctx.put_path_identity_var(insert_stmt,
iterator_set.path_id,
cols[-1],
force=True,
env=subctx.env)
pathctx.put_path_bond(insert_stmt, iterator_set.path_id)
toplevel = ctx.toplevel_stmt
toplevel.ctes.append(insert_cte)
# Process necessary updates to the link tables.
for shape_el, props_only in external_inserts:
process_link_update(ir_stmt,
shape_el,
props_only,
wrapper,
insert_cte,
iterator_cte,
ctx=ctx)
if parent_link_props:
prop_elements = []
with ctx.newscope() as scopectx:
scopectx.rel = wrapper
for shape_el in parent_link_props:
rptr = shape_el.rptr
scopectx.path_scope[rptr.source.path_id] = wrapper
pathctx.put_path_rvar_if_not_exists(wrapper,
rptr.source.path_id,
insert_rvar,
aspect='value',
env=scopectx.env)
dispatch.compile(shape_el, ctx=scopectx)
tuple_el = astutils.tuple_element_for_shape_el(shape_el, None)
prop_elements.append(tuple_el)
valtuple = pgast.TupleVar(elements=prop_elements, named=True)
pathctx.put_path_value_var(wrapper,
ir_stmt.subject.path_id,
valtuple,
force=True,
env=ctx.env)
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}')