def range_for_material_objtype(objtype: s_objtypes.ObjectType,
path_id: irast.PathId,
*,
include_overlays: bool = True,
env: context.Environment) -> pgast.BaseRangeVar:
from . import pathctx # XXX: fix cycle
objtype = objtype.material_type(env.schema)
objtype_name = objtype.get_name(env.schema)
table_schema_name, table_name = common.get_backend_name(env.schema,
objtype,
catenate=False)
if objtype_name.module == 'schema':
# Redirect all queries to schema tables to edgedbss
table_schema_name = 'edgedbss'
relation = pgast.Relation(
schemaname=table_schema_name,
name=table_name,
path_id=path_id,
)
rvar = pgast.RangeVar(
relation=relation,
alias=pgast.Alias(aliasname=env.aliases.get(objtype_name.name)))
overlays = env.rel_overlays.get(objtype_name)
if overlays and include_overlays:
set_ops = []
qry = pgast.SelectStmt()
qry.from_clause.append(rvar)
pathctx.put_path_value_rvar(qry, path_id, rvar, env=env)
pathctx.put_path_bond(qry, path_id)
set_ops.append(('union', qry))
for op, cte in overlays:
rvar = pgast.RangeVar(
relation=cte,
alias=pgast.Alias(aliasname=env.aliases.get(hint=cte.name)))
qry = pgast.SelectStmt(from_clause=[rvar], )
pathctx.put_path_value_rvar(qry, path_id, rvar, env=env)
pathctx.put_path_bond(qry, path_id)
if op == 'replace':
op = 'union'
set_ops = []
set_ops.append((op, qry))
rvar = range_from_queryset(set_ops, objtype, env=env)
return rvar
def process_linkprop_update(ir_stmt: irast.MutatingStmt, ir_expr: irast.Base,
wrapper: pgast.Query,
dml_cte: pgast.CommonTableExpr, *,
ctx: context.CompilerContextLevel) -> None:
"""Perform link property updates to a link relation.
:param ir_stmt:
IR of the statement.
:param ir_expr:
IR of the UPDATE body element.
:param wrapper:
Top-level SQL query.
:param dml_cte:
CTE representing the SQL UPDATE to the main relation of the Object.
"""
toplevel = ctx.toplevel_stmt
rptr = ir_expr.rptr
ptrcls = rptr.ptrcls
target_tab = dbobj.range_for_ptrcls(ptrcls,
'>',
include_overlays=False,
env=ctx.env)
dml_cte_rvar = pgast.RangeVar(
relation=dml_cte,
alias=pgast.Alias(aliasname=ctx.env.aliases.get('m')))
cond = astutils.new_binop(
pathctx.get_rvar_path_identity_var(dml_cte_rvar,
ir_stmt.subject.path_id,
env=ctx.env),
dbobj.get_column(target_tab, 'source', nullable=False),
op='=',
)
targets = []
for prop_el in ir_expr.shape:
ptrname = prop_el.rptr.ptrcls.get_shortname(ctx.env.schema)
with ctx.new() as input_rel_ctx:
input_rel_ctx.expr_exposed = False
input_rel = dispatch.compile(prop_el.expr, ctx=input_rel_ctx)
targets.append(pgast.UpdateTarget(name=ptrname.name,
val=input_rel))
updstmt = pgast.UpdateStmt(relation=target_tab,
where_clause=cond,
targets=targets,
from_clause=[dml_cte_rvar])
updcte = pgast.CommonTableExpr(
query=updstmt,
name=ctx.env.aliases.get(ptrcls.get_shortname(ctx.env.schema).name))
toplevel.ctes.append(updcte)
def table_from_ptrcls(ptrcls: s_links.Link, *,
env: context.Environment) -> pgast.RangeVar:
"""Return a Table corresponding to a given Link."""
table_schema_name, table_name = common.get_table_name(ptrcls,
catenate=False)
pname = ptrcls.shortname
if pname.module == 'schema':
# Redirect all queries to schema tables to edgedbss
table_schema_name = 'edgedbss'
relation = pgast.Relation(schemaname=table_schema_name, name=table_name)
rvar = pgast.RangeVar(
relation=relation,
alias=pgast.Alias(aliasname=env.aliases.get(pname.name)))
return rvar
def rvar_for_rel(
rel: pgast.BaseRelation, *,
lateral: bool=False, colnames: typing.List[str]=[],
env: context.Environment) -> pgast.BaseRangeVar:
if isinstance(rel, pgast.Query):
alias = env.aliases.get(rel.name or 'q')
rvar = pgast.RangeSubselect(
subquery=rel,
alias=pgast.Alias(aliasname=alias, colnames=colnames),
lateral=lateral,
)
else:
alias = env.aliases.get(rel.name)
rvar = pgast.RangeVar(
relation=rel,
alias=pgast.Alias(aliasname=alias, colnames=colnames)
)
return rvar
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
target_tab_name = (target_rvar.relation.schemaname,
target_rvar.relation.name)
tab_cols = dbobj.cols_for_pointer(mptrcls, env=ctx.env)
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, specified_cols = 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 specified_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 init_dml_stmt(
ir_stmt: irast.MutatingStmt, dml_stmt: pgast.DML, *,
ctx: context.CompilerContextLevel,
parent_ctx: context.CompilerContextLevel) \
-> typing.Tuple[pgast.Query, pgast.CommonTableExpr,
pgast.CommonTableExpr]:
"""Prepare the common structure of the query representing a DML stmt.
:param ir_stmt:
IR of the statement.
:param dml_stmt:
SQL DML node instance.
:return:
A (*wrapper*, *dml_cte*, *range_cte*) tuple, where *wrapper* the
the wrapping SQL statement, *dml_cte* is the CTE representing the
SQL DML operation in the main relation of the Object, and
*range_cte* is the CTE for the subset affected by the statement.
*range_cte* is None for INSERT statmenets.
"""
wrapper = ctx.rel
clauses.init_stmt(ir_stmt, ctx, parent_ctx)
target_ir_set = ir_stmt.subject
dml_stmt.relation = dbobj.range_for_set(ir_stmt.subject,
include_overlays=False,
env=ctx.env)
pathctx.put_path_value_rvar(dml_stmt,
target_ir_set.path_id,
dml_stmt.relation,
env=ctx.env)
pathctx.put_path_source_rvar(dml_stmt,
target_ir_set.path_id,
dml_stmt.relation,
env=ctx.env)
dml_stmt.path_scope.add(target_ir_set.path_id)
dml_cte = pgast.CommonTableExpr(query=dml_stmt,
name=ctx.env.aliases.get(hint='m'))
if isinstance(ir_stmt, (irast.UpdateStmt, irast.DeleteStmt)):
# UPDATE and DELETE operate over a range, so generate
# the corresponding CTE and connect it to the DML query.
range_cte = get_dml_range(ir_stmt, dml_stmt, ctx=ctx)
range_rvar = pgast.RangeVar(
relation=range_cte,
alias=pgast.Alias(aliasname=ctx.env.aliases.get(hint='range')))
relctx.pull_path_namespace(target=dml_stmt, source=range_rvar, ctx=ctx)
# Auxillary relations are always joined via the WHERE
# clause due to the structure of the UPDATE/DELETE SQL statments.
id_col = common.edgedb_name_to_pg_name('std::id')
dml_stmt.where_clause = astutils.new_binop(
lexpr=pgast.ColumnRef(
name=[dml_stmt.relation.alias.aliasname, id_col]),
op=ast.ops.EQ,
rexpr=pathctx.get_rvar_path_identity_var(range_rvar,
target_ir_set.path_id,
env=ctx.env))
# UPDATE has "FROM", while DELETE has "USING".
if hasattr(dml_stmt, 'from_clause'):
dml_stmt.from_clause.append(range_rvar)
else:
dml_stmt.using_clause.append(range_rvar)
else:
range_cte = None
# Due to the fact that DML statements are structured
# as a flat list of CTEs instead of nested range vars,
# the top level path scope must be empty. The necessary
# range vars will be injected explicitly in all rels that
# need them.
ctx.path_scope.clear()
pathctx.put_path_value_rvar(dml_stmt,
ir_stmt.subject.path_id,
dml_stmt.relation,
env=ctx.env)
pathctx.put_path_source_rvar(dml_stmt,
ir_stmt.subject.path_id,
dml_stmt.relation,
env=ctx.env)
dml_rvar = pgast.RangeVar(
relation=dml_cte,
alias=pgast.Alias(aliasname=parent_ctx.env.aliases.get('d')))
relctx.include_rvar(wrapper, dml_rvar, ir_stmt.subject.path_id, ctx=ctx)
pathctx.put_path_bond(wrapper, ir_stmt.subject.path_id)
return wrapper, dml_cte, dml_rvar, range_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 range_for_ptrcls(
ptrcls: s_links.Link, direction: s_pointers.PointerDirection, *,
include_overlays: bool=True,
env: context.Environment) -> pgast.BaseRangeVar:
""""Return a Range subclass corresponding to a given ptr step.
If `ptrcls` is a generic link, then a simple RangeVar is returned,
otherwise the return value may potentially be a UNION of all tables
corresponding to a set of specialized links computed from the given
`ptrcls` taking source inheritance into account.
"""
linkname = ptrcls.shortname
endpoint = ptrcls.source
tgt_col = pgtypes.get_pointer_storage_info(
ptrcls, resolve_type=False, link_bias=True).column_name
cols = [
'std::source',
tgt_col
]
set_ops = []
ptrclses = set()
for source in {endpoint} | set(endpoint.descendants(env.schema)):
# Sift through the descendants to see who has this link
try:
src_ptrcls = source.pointers[linkname].material_type()
except KeyError:
# This source has no such link, skip it
continue
else:
if src_ptrcls in ptrclses:
# Seen this link already
continue
ptrclses.add(src_ptrcls)
table = table_from_ptrcls(src_ptrcls, env=env)
qry = pgast.SelectStmt()
qry.from_clause.append(table)
qry.rptr_rvar = table
# Make sure all property references are pulled up properly
for colname in cols:
selexpr = pgast.ColumnRef(
name=[table.alias.aliasname, colname])
qry.target_list.append(
pgast.ResTarget(val=selexpr, name=colname))
set_ops.append(('union', qry))
overlays = env.rel_overlays.get(src_ptrcls.shortname)
if overlays and include_overlays:
for op, cte in overlays:
rvar = pgast.RangeVar(
relation=cte,
alias=pgast.Alias(
aliasname=env.aliases.get(cte.name)
)
)
qry = pgast.SelectStmt(
target_list=[
pgast.ResTarget(
val=pgast.ColumnRef(
name=[col]
)
)
for col in cols
],
from_clause=[rvar],
)
set_ops.append((op, qry))
rvar = range_from_queryset(set_ops, ptrcls, env=env)
return rvar
