Allow directly nested arrays of arrays #7419
Conversation
|
Needs lots of tests, especially for the schema. While compiler can handle nested arrays, the protocol can't
|
dnwpark
force-pushed
the
array-of-array
branch
4 times, most recently
from
f861a04 to
693b4cf
Compare
dnwpark
force-pushed
the
array-of-array
branch
3 times, most recently
from
6403e0a to
b8fadd0
Compare
| class Polymorphism(s_enum.StrEnum): | ||
| NotUsed = 'NotUsed' | ||
| Simple = 'Simple' | ||
| Array = 'Array' | ||
| Collection = 'Collection' |
There was a problem hiding this comment.
This new mechanism needs documentation, I think
edb/pgsql/compiler/astutils.py
Outdated
| def array_get_inner_array( | ||
| wrapped_array: pgast.BaseExpr, | ||
| array_typeref: irast.TypeRef, | ||
| ) -> pgast.BaseExpr: | ||
| # Nested arrays wrap their inner arrays in a tuple |
There was a problem hiding this comment.
Make this a docstring comment, and put in the comment the SQL that this generates.
(We don't do this universally, but should!)
| @test.xerror(""" | ||
| edb.errors.InternalServerError: return type record[] is not supported | ||
| for SQL functions | ||
| """) |
There was a problem hiding this comment.
Do some of these work? If so, let's split the test
| @test.xerror(""" | ||
| edb.errors.InternalServerError: return type record[] is not supported | ||
| for SQL functions | ||
| """) |
There was a problem hiding this comment.
I think the eventual fix for this category of problems (which also shows up with arrays of tuples currently) is to inline the body of the function. (Which is defined as SQL)
tests/schemas/cards.esdl
Outdated
| alias AliasArrayOfArrayOfScalar := [[1, 2, 3], [4, 5, 6]]; | ||
|
|
||
| alias CardsByCost := array_agg(( | ||
| for cost in range_unpack(range(0, max(Card.cost))) |
tests/test_edgeql_expr_aliases.py
Outdated
| async def test_edgeql_aliases_array_of_array_02(self): | ||
| await self.assert_query_result( | ||
| r""" | ||
| select CardsByCost; | ||
| """, | ||
| [ | ||
| [ | ||
| [], | ||
| ['Imp', 'Dwarf', 'Sprite'], | ||
| ['Bog monster', 'Giant eagle'], | ||
| ['Giant turtle', 'Golem'], | ||
| ['Djinn'] | ||
| ], | ||
| ], | ||
| ) |
There was a problem hiding this comment.
Oh, though it doesn't work yet :P
tests/test_edgeql_sys.py
Outdated
| @@ -121,7 +123,7 @@ async def _test_sys_query_stats(self): | |||
| async def _test_sys_query_stats_with_tag(self): | |||
| # Test tags are correctly set | |||
| tag = 'test_tag' | |||
| self.con = self.con.with_annotation('tag', tag) | |||
| self.con = self.con.with_query_tag(tag) | |||
There was a problem hiding this comment.
Could you split the gel-python upgrade to a standalone PR updating it to whatever the latest normal version is?
edb/pgsql/types.py
Outdated
| @@ -382,6 +382,8 @@ def pg_type_from_ir_typeref( | |||
| or (irtyputils.is_abstract(ir_typeref.subtypes[0]) | |||
| and irtyputils.is_scalar(ir_typeref.subtypes[0]))): | |||
| return ('anyarray',) | |||
| elif (irtyputils.is_array(ir_typeref.subtypes[0])): | |||
| @@ -237,6 +237,54 @@ def array_as_json_object( | |||
| ) | |||
| ] if needs_unnest else [], | |||
| ) | |||
| elif irtyputils.is_array(el_type): | |||
There was a problem hiding this comment.
Put a comment about why the complication
edb/pgsql/compiler/relgen.py
Outdated
| has_array_polymorphic_arg = any( | ||
| ir_arg.polymorphism == qltypes.Polymorphism.Array | ||
| for ir_arg in expr.args.values() | ||
| ) | ||
| wrap_single_polymorphic_args = ( | ||
| has_array_polymorphic_arg | ||
| or expr.return_polymorphism == qltypes.Polymorphism.Array | ||
| ) | ||
|
|
There was a problem hiding this comment.
Could you add some more comments here? I don't totally understand why we need to wrap arguments
dnwpark
force-pushed
the
array-of-array
branch
5 times, most recently
from
781cd6c to
11d0b2a
Compare
Allows directly nested arrays.
The query
select ([[1],[2]]++[[3]])[1:3][0];returns[2].The query
select find(array_replace([[1],[2],[3]],[2],[4]),[1]);returns0.