Alter alias even from ref propagation

[aljazerzen]

• uncomment the hack
• make commands produced by affected refs execute as causes instead of innards
• fix for reflection writer when we create a type and drop it in the same migration

[aljazerzen]

@msullivan Yesterday, we changed affected refs to be applied as "caused" instead of "innards". You said that you are scared what this will break. I think we have exactly one test that breaks:

create type X {
                create property bar -> int64 {
                    create constraint std::exclusive
                }
            };

alter type X alter property bar using ('1');

The alter produces the following delta root:

<DeltaRoot canonical=True at 0x7fe853cc6b90> (
    <AlterSchemaVersion classname=__schema_version__ at 0x7fe8691f8fd0> (
        version = UUID('d95d5c03-d7e2-11ee-ad61-d18567326038') | UUID('1cacc186-d7e3-11ee-a9b2-a180ec730ca2')
    ),
    <CreateMigration classname=m15cipax5pdr45grvgdjw5ejtq3twiran5gkn6pqy2txlskmnohrna at 0x7fe853c88510> (
        script       = None | "SET generated_by := (schema::MigrationGeneratedBy.DDLStatement);\nALTER TYPE X {\n    ALTER PROPERTY bar {\n        USING ('1');\n    };\n};",
        builtin      = None | False,
        internal     = None | False,
        parents      = None | [<ObjectShell Migration(<UnqualName m1kackz3nz3oketzgvnbqdbul3w546c7li46ydofri37piyt6kluna>) at 0x7fe8538966d0>],
        generated_by = None | 'DDLStatement',
        name         = None | <UnqualName m15cipax5pdr45grvgdjw5ejtq3twiran5gkn6pqy2txlskmnohrna>,
        <AlterObjectType source_context=<edb.common.context.ParserContext object at 0x7fe85382a5d0>, classname=default::X at 0x7fe853cd4150> (
            <AlterProperty source_context=<edb.common.context.ParserContext object at 0x7fe853b3dd10>, classname=default::__|bar@default|X at 0x7fe868fabe90> (
                main ,
                <AlterPropertyOwned classname=default::__|bar@default|X at 0x7fe8539f31d0>(owned=True | True),
                <SetPropertyType classname=default::__|bar@default|X at 0x7fe8539f0990> (
                    target          = <ScalarType 00000000-0000-0000-0000-000000000105 at 0x0x7fe8691860d0> | <TypeShell ScalarType(<QualName std::str>) at 0x7fe85388f950> # computed,
                    computed_fields = None | frozenset({'target'})
                ),
                computed_link_alias  = None | None,
                computed_link_alias_is_backward = None | None,
                expr                 = None | <CompiledExpression text="'1'", refs={00000000-0000-0000-0000-000000000101}, origin=None at 0x7fe853a119d0>,
                <AlterPropertyUpperCardinality classname=default::__|bar@default|X at 0x7fe853a10e10> (
                    cardinality     = None | <SchemaCardinality.One: 'One'> # computed,
                    computed_fields = edb.common.checked.FrozenCheckedSet[str]({'target'}) | frozenset({'target', 'cardinality'})
                ),
                <AlterPropertyLowerCardinality classname=default::__|bar@default|X at 0x7fe853a11450> (
                    required        = None | True # computed,
                    computed_fields = edb.common.checked.FrozenCheckedSet[str]({'target', 'cardinality'}) | frozenset({'target', 'required', 'cardinality'})
                ),
                computable           = None | True,
                caused ,
                <AlterConstraint classname=default::std|exclusive@default|__||bar&default||X@da39a3ee5e6b4b0d3255bfef95601890afd80709, ddl_identity={'args': edb.schema.expr.ExpressionList([])}, from_expr_propagation=True at 0x7fe853820090> (
                    finalexpr = <Expression text='SELECT false', refs={00000000-0000-0000-0000-000000000100, 00000000-0000-0000-0000-000000000101, 00000000-0000-0000-0000-000000000105, 00000000-0000-0000-0000-000000000109, 00000000-0000-0000-0000-00000000010f}, origin=None at 0x7fe853972410> | <CompiledExpression text='select std::_is_exclusive(__subject__)', refs={00000000-0000-0000-0000-000000000101, a58ae470-e626-5d1c-90ee-c73125d07418, 00000000-0000-0000-0000-000000000109}, origin=None at 0x7fe853972890>,
                    args      = edb.schema.expr.ExpressionList([]) | None,
                    params    = [b0d7669e-476d-5d5a-b1e3-87566e023465] | [b0d7669e-476d-5d5a-b1e3-87566e023465],
                    abstract  = False | False
                )
            )
        ),
        id           = None | UUID('1ca55b59-d7e3-11ee-a0c5-a9fbe2800863')
    )
)

... which fails in postgres, because we try to drop the column while constraint still refers to it.

Solution that I'm seeing is to move the "column drop" to the end of the "alter property" command.

[aljazerzen]

The remaining error is this:

# from test_schema_migrations_computed_optionality_01
# schema 1:
            abstract type Removable {
                optional single property removed := false;
            };

            type Topic extending Removable { };
            type Definition { };

            alias VisibleTopic := (
                SELECT Topic {
                    defs := (SELECT Definition),
                }
                FILTER NOT .removed
            );

# schema 2:
            abstract type Removable {
                property removed := false;
            };

            type Topic extending Removable { };
            type Definition { };

            alias VisibleTopic := (
                SELECT Topic {
                    defs := (SELECT Definition),
                }
                FILTER NOT .removed
            );

We have an "alter property removed", which propagates via two paths:

• via Topic, because extends Removable, which propagates into VisibleTopic, which drops alias types and recreates,
• via reference to property removed in the VisibleTopic, which triggers alter alias again.

This fails because the property VisibleTopic.removed is found to be descendant of Removable.removed and is scheduled to be updated here:
https://github.com/edgedb/edgedb/blob/main/edb/schema/inheriting.py#L958
... but because one alter alias runs in the first iteration, the property does not exists anymore. Well it does exist, but with a different id.

[msullivan]

get will raise an error by itself, right?
If you want it deferred to this assert, pass a default

[msullivan]

We need to enable ruff's prefix whitespace linting!