@hexed saddle 👋 There seems to be a problem with enums. When running codegen from inside a ExperimentalPrivilegedNesting container the TypeDefKind enum only has the STRING_KIND, INTEGER_KIND, etc, values and not STRING, INTEGER, etc.
#Enum issue inside nested connection
late wigeon
late wigeon
It's the same version, and dev build: v0.18.15-010101000000-dev-5efecbf8f2e5
Hmmm... this seems to happen only on the new self calls PR, specifically in the new moduleTypeDefs function so it's in a phase where the module isn't fully loaded yet. Maybe that view isn't being applied yet at this point? Can we move that needle? Or maybe it's just missing the thing that applies the view?
It's getting called at this point in core/schema/modulesource.go: https://github.com/dagger/dagger/pull/10584/files#diff-35a72c75f37c6dcddd3e1df88cccf5a8b22ac1f6304301c66c3a1c7394ff0eafR2274
late wigeon
I made a repro and it's ok in v0.18.14. Will test against main now, but I suspect it only happens in the self calls PR since it changes how modules are loaded.
Btw, when you changed the codegen to refer to other members like:
class TypeDefKind(Enum):
STRING = "STRING_KIND"
STRING_KIND = STRING
Was the motivation simply to make an alias or is there a reason not to:
class TypeDefKind(Enum):
STRING = "STRING"
STRING_KIND = "STRING_KIND"
hexed saddle
no real reason, it can just be changed back if you like
but go/typescript did it - i found it much clearer as to which ones were aliases
oh wait wait. yes there is a reason.
the values should be the same
otherwise, we lose the property that STRING == STRING_KIND
late wigeon
Thing is in the first case the type of STRING is a string, while the type of STRING_KIND is an enum member.
The proper way to make an alias in a Python enum is just to have the same value.
hexed saddle
😢 right okay, sorry, that's a my bad then
the codegen result should be:
class TypeDefKind(Enum):
STRING = "STRING_KIND"
STRING_KIND = "STRING_KIND"
A simple example:
import enum
class Test(enum.Enum):
ONE = "ONE"
UNO = "ONE"
print("By name:", Test["UNO"].name)
# Output: By name: ONE
lemme revert that change
it used to be like this, i did it in https://github.com/dagger/dagger/pull/10632/commits/caac8b23db2452d67d83dec0ae9af72a67c45528
late wigeon
Just reverting will produce:
class TypeDefKind(Enum):
STRING = "STRING"
STRING_KIND = "STRING_KIND"
Instead of this though:
class TypeDefKind(Enum):
STRING = "STRING_KIND"
STRING_KIND = "STRING_KIND"
No issue on main too. Going to test that PR the same way or see if it only happens within moduleTypeDefs
late wigeon
@hexed saddle, is there a way to confirm enum values with a GraphQL query (with dagger listen)?
I've created this repro:
package main
import (
"context"
"dagger/nested/internal/dagger"
)
type Nested struct{}
func (m *Nested) Test(ctx context.Context) (string, error) {
var code string = `
import pprint
import anyio
import graphql
from gql.dsl import DSLSchema
import dagger
from dagger import dag
def get_value(value, schema) -> str:
if value.ast_node and (directive := schema.get_directive("enumValue")):
args = graphql.get_directive_values(directive, value.ast_node)
if args:
return args["value"]
return value.value
async def main():
async with await dagger.connect():
print("Version:", await dag.version())
schema = DSLSchema(await dag._ctx.conn.session.get_schema())
pprint.pprint(
{
name: get_value(value, schema)
for name, value in schema.TypeDef.withKind._get_argument(
"kind"
).type.of_type.values.items()
}
)
anyio.run(main)
`
return dag.Container().
From("ghcr.io/astral-sh/uv:debian-slim").
WithNewFile("/script.py", code).
WithExec([]string{"uv", "run", "--with", "dagger-io", "/script.py"}, dagger.ContainerWithExecOpts{
ExperimentalPrivilegedNesting: true,
}).
Stdout(ctx)
}
But it's returning this:
{'BOOLEAN': 'BOOLEAN',
'BOOLEAN_KIND': 'BOOLEAN_KIND',
'ENUM': 'ENUM',
'ENUM_KIND': 'ENUM_KIND',
'FLOAT': 'FLOAT',
'FLOAT_KIND': 'FLOAT_KIND',
'INPUT': 'INPUT',
'INPUT_KIND': 'INPUT_KIND',
'INTEGER': 'INTEGER',
'INTEGER_KIND': 'INTEGER_KIND',
'INTERFACE': 'INTERFACE',
'INTERFACE_KIND': 'INTERFACE_KIND',
'LIST': 'LIST',
'LIST_KIND': 'LIST_KIND',
'OBJECT': 'OBJECT',
'OBJECT_KIND': 'OBJECT_KIND',
'SCALAR': 'SCALAR',
'SCALAR_KIND': 'SCALAR_KIND',
'STRING': 'STRING',
'STRING_KIND': 'STRING_KIND',
'VOID': 'VOID',
'VOID_KIND': 'VOID_KIND'}
I expected 'STRING': 'STRING_KIND' for example.
hexed saddle
it potentially looks like you're always following the return value.value path
late wigeon
Oh wait, I may need the insert_stubs thing?
hexed saddle
and not looking at the enumValue
late wigeon
https://github.com/dagger/dagger/pull/10802\
The order may still be wrong, right?
class TypeDefKind(Enum):
STRING = "STRING_KIND"
STRING_KIND = "STRING_KIND"
It's declared like so:
TypeDefKindString = TypeDefKinds.Register("STRING_KIND", "A string value.")
_ = TypeDefKinds.AliasView("STRING", "STRING_KIND", enumView)
Which means that it should have this order:
class TypeDefKind(Enum):
STRING_KIND = "STRING_KIND"
STRING = "STRING_KIND"
Otherwise dagger.TypeDefKind.STRING_KIND becomes an alias of dagger.TypeDefKind.STRING instead, meaning the member's name will be "STRING" instead of "STRING_KIND".
hexed saddle
mmmm, indeed. the name should be the original STRING_KIND
this is definitely true in go already
late wigeon
So you can't sort those values. How can you tell in codegen which one is the alias or not? In this case I can check if name == value, but what about user defined ones?
hexed saddle
they're not sorted, they're bucketed by value first
then the first value in the bucket is the "original"
i think the only fix for python needed is to avoid the sorting step?
late wigeon
Yep, I think so.
late wigeon
@hexed saddle, indeed this fixes my problem. I've sorted by value after grouping which effectively keeps the same overall sort but flips the original value vs alias. I can push the change to your branch.