Column ops implementation first draft #1696

sfc-gh-heshah · 2024-05-29T16:35:16Z

Added draft implementations for most column expressions (excluding those without corresponding AST entities, ISIN, OVER, and string expression based functions missing IR entities).

sfc-gh-oplaton

Superficially, the changes look good, with one observation: The functionality in src/snowflake/snowpark/column.py is very repetitive, to the point where I probably would have missed anything broken.

I think it would make sense for at least some of the functionality (e.g. binary operators) to be further encapsulated into a higher-order function.

For example, __eq__ and __ne__ could become:

def __eq__(self, other):
    return self.bin_op_impl(lambda ast: ast.sp_column_equal_to, EqualTo, other)

def __ne__(self, other):
    return self.bin_op_impl(lambda ast: ast.sp_column_not_equal, NotEqualTo, other)

def bin_op_impl(self, property, comparator, other):
    ast = proto.SpColumnExpr()
    property(ast).lhs.CopyFrom(self._ast)
    Column._fill_ast(property(ast).rhs, other)
    return Column(comparator(self._expression, Column._to_expr(other)), ast = ast)

sfc-gh-oplaton · 2024-05-30T00:44:27Z

I think the change looks good. Thanks for the updated approach! Of course, I'd recommend getting additional input from Leonhard and/or Arthur, as I might have missed some of the finer points of the design.

sfc-gh-heshah · 2024-05-30T01:20:03Z

Sounds good, thank you for the suggestion!

Originally I was thinking the change you mentioned should correspond to further abstracting the IR for SpColumnExpr operations as Arthur suggested on Slack, but I think this minimizes the code changes required if the AST definitions are changed.

Along those same lines I though there were other repetitive AST building patterns that could be abstracted into a static Column._create_ast method which I added. This should further reduce dependence on the exact protobuf definition, and minimize changes necessary in the code in case the AST is evolved further, but please let me know if its a bit overboard.

sfc-gh-oplaton · 2024-05-30T15:32:20Z

src/snowflake/snowpark/column.py

+        ast = proto.SpColumnExpr()
+        if property is not None:
+            for attr, value in assign_fields.items():
+                setattr(property(ast), attr, value)


This isn't critical, but I wanted to make it explicit:

The point of using a callable to access object properties (lambda o: o.some_property) was to avoid using getattr/setattr, but they're equivalent, and there's no strict technical reason why callables are superior to getattr/setattr, just a personal preference (it looks more like code than stringly-typed code; very superficial preference).

That said, the callable pattern is more verbose at the call site, especially when using it repeatedly. (property = lambda ast: ast.sp_column_regexp, versus property = "sp_column_regexp",).

In summary, I think it would make the code less readable to have both the callable pattern and getattr/setattr. Pick one that you prefer (I don't feel strongly either way - callable looks more like code, but getattr/setattr will result in terser code) and be uniform. Uniformity matters a lot for ease of maintenance.

Aside: I assume performance isn't a consideration here. There might be a difference between using a callable and getattr/setattr, but I didn't look into it, and I don't have an intuition here. If this is going to run billions of times, maybe writing a short benchmark and comparing would also help (if you can do it in <15 minutes, this doesn't need to become a side project of its own).

I don't think we need to be concerned with performance here. I would pick whichever code is easier to read/less to write. Having this function here is super useful throughout the code base!

If you're sure performance won't matter, I recommend using getattr/setattr everywhere. Way fewer lexical tokens at each call site.

Just wrote a short benchmark using timeit, wasn't too much work

import timeit setup = """ import snowflake.snowpark._internal.proto.ast_pb2 as proto test = proto.SpColumnExpr() prop = lambda ast: ast.sp_column_apply__string """ getcode1 = """ t = prop(test) """ getcode2 = """ t = getattr(test, "sp_column_apply__string") """ print("Callable:\t", timeit.timeit(setup=setup, stmt=getcode1)) print("getattr:\t", timeit.timeit(setup=setup, stmt=getcode2)) setcode1 = """ test.sp_column_apply__string.field = "test" """ setcode2 = """ setattr(test.sp_column_apply__string, "field", "test") """ print(".field direct:\t", timeit.timeit(setup=setup, stmt=setcode1)) print("setattr:\t", timeit.timeit(setup=setup, stmt=setcode2))

The results I got for 1 million runs of each line were:

Callable: 0.04975545800000003 getattr: 0.040397291999999974 .field direct: 0.10505558299999995 setattr: 0.12402554099999996

Overall it seems getattr is actually faster, but setting the field directly is faster. I do think a refactor with using getattr will be cleaner so I will proceed with that.

sfc-gh-lspiegelberg · 2024-05-30T16:03:05Z

src/snowflake/snowpark/column.py

@@ -226,6 +228,7 @@ class Column:
    This class has methods for the most frequently used column transformations and operators. Module :mod:`snowflake.snowpark.functions` defines many functions to transform columns.
    """

+    # NOTE: For now assume Expression instances can be safely ignored when building AST


can you add an example that's problematic here? Not sure what to do to resolve this note.

This won't be an issue once all API functions that return Column instances also populate the ast parameter in the constructor.

For example currently because the isin method is not implemented to build the AST before returning, doing df.filter(col("A").isin(1, 2, 3) & col("B")) would fail since the boolean operator & would try to construct an AST using that of the new col("A").isin(1, 2, 3) column (which we currently don't fill if the only argument provided in the Column constructor is an expr1 of type Expression)

great! let's just add what you explained as comment.

can we add the example?

is there a TODO post phase 0?

sfc-gh-lspiegelberg · 2024-05-30T16:04:00Z

src/snowflake/snowpark/column.py

            if isinstance(expr1, str) and isinstance(expr2, str):
                if expr2 == "*":
                    self._expression = Star([], df_alias=expr1)
                else:
                    self._expression = UnresolvedAttribute(
                        quote_name(expr2), df_alias=expr1
                    )
+


What would the alias field here be? is this for expr.as_(alias)? Can we maybe add an example what would need to get captured here?

The alias field would be something that the user provides to the DataFrame.alias(self, name: str) method.

The Snowpark API then allows users to create columns that reference this DataFrame instance by its alias using a line like col(<df_alias>, <column_name>)

In the IR we'll need the alias to resolve which DataFrame instance the user is referring to, but there might be a better way to go about enabling this functionality.

prob. would be good to include your explanation as comment here

sfc-gh-lspiegelberg · 2024-05-30T16:04:28Z

src/snowflake/snowpark/column.py

                if expr1 == "*":
-                    self._ast.sp_column_sql_expr.sql = "*"
+                    self._ast = Column._create_ast(


As soon as #1701 is in , can we add tests for this?

sfc-gh-lspiegelberg · 2024-05-30T16:04:57Z

src/snowflake/snowpark/column.py

+                copy_messages = {"col": self._ast},
+                assign_fields = {"field": field},
+            )
+            return Column(SubfieldString(self._expression, field), ast = ast)


let's use the precommit to format everything :)

sfc-gh-lspiegelberg · 2024-05-30T16:05:11Z

src/snowflake/snowpark/column.py

        else:
            raise TypeError(f"Unexpected item type: {type(field)}")

    # overload operators
+    def _bin_op_impl(self, property: Callable, operator: BinaryExpression, other: ColumnOrLiteral) -> "Column":


pls. add docstrings to this

sfc-gh-lspiegelberg · 2024-05-30T16:05:36Z

src/snowflake/snowpark/column.py

    def __eq__(self, other: Union[ColumnOrLiteral, Expression]) -> "Column":
        """Equal to."""
-        right = Column._to_expr(other)
-        return Column(EqualTo(self._expression, right))
+        return self._bin_op_impl(lambda ast: ast.sp_column_equal_to, EqualTo, other)


big fan of the refactor here, so much cleaner!

sfc-gh-lspiegelberg · 2024-05-30T16:06:02Z

src/snowflake/snowpark/column.py


    def __rtruediv__(self, other: Union[ColumnOrLiteral, Expression]) -> "Column":
-        return Column(Divide(Column._to_expr(other), self._expression))
+        return self._bin_op_rimpl(lambda ast: ast.sp_column_divide, Divide, other)

    def __mod__(self, other: Union[ColumnOrLiteral, Expression]) -> "Column":
        """Reminder."""


typo: Remainder

sfc-gh-lspiegelberg · 2024-05-30T16:07:19Z

src/snowflake/snowpark/column.py


    def __neg__(self) -> "Column":
        """Unary minus."""
+        # TODO: Need SpColumnNeg IR entity


I feel like we should start filing JIRAs with all of these TODOs. @sfc-gh-azwiegincew shall we maybe create an epic ThinClient AST/IR translation and track those under it?

Yes. I'll spend some quality time with Jira today.

sfc-gh-lspiegelberg · 2024-05-30T16:08:11Z

src/snowflake/snowpark/column.py

        return Column(Not(self._expression))

    def _cast(self, to: Union[str, DataType], try_: bool = False) -> "Column":
+        # TODO: Update SpColumnCast IR entity with new field "try_", then uncomment


how difficult would this be? I would be also fine to defer this work to another PR here. Ok, to have only simple column expr supported for now.

Not very difficult, I have already made the changes and compiled the new protobuf in the PR in the monorepo with the modifications necessary: https://github.com/snowflakedb/snowflake/pull/183083

However, I will get rid of these TODOs in favor of a flexible approach in the _create_ast method which can silently skip IR entities and fields or raise an error depending on a parameter.

sfc-gh-lspiegelberg · 2024-05-30T16:09:07Z

src/snowflake/snowpark/column.py

@@ -591,11 +699,18 @@ def regexp(self, pattern: ColumnOrLiteralStr) -> "Column":
        :meth:`rlike` is an alias of :meth:`regexp`.

        """
+        # TODO: Should SpColumnRegexp.pattern be an Expr in the IR, could sbe SpColumnExpr as in SpColumnLike?


There seem to be a few TODOs which require IR changes, maybe collect in doc, and then go through in meeting? We prob. can resolve those quickly.

sfc-gh-lspiegelberg · 2024-05-30T16:09:56Z

src/snowflake/snowpark/column.py

-                    for col in parse_positional_args_to_list(*cols)
-                ],
-            )
+        prop = lambda ast: ast.sp_column_within_group


The code here looks long - can we add a comment to explain the logic?

sfc-gh-lspiegelberg · 2024-05-30T16:10:45Z

src/snowflake/snowpark/column.py

    def _named(self) -> NamedExpression:
        if isinstance(self._expression, NamedExpression):
            return self._expression
        else:
            return UnresolvedAlias(self._expression)
+
+    @staticmethod
+    def _create_ast(property: Optional[Callable] = None,


This looks like a super helpful function. Can we add a docstring to explain it? This seems to be used throughout - so a good idea to have a detailed docstring and also whichever asserts/checks are needed within it.

sfc-gh-lspiegelberg · 2024-05-30T16:12:11Z

src/snowflake/snowpark/column.py

+    @classmethod
+    def _fill_ast(cls, ast: proto.SpColumnExpr, value: ColumnOrLiteral) -> None:
+        """
+        Copy from a Column object's AST, or copy a literal value into an AST expression.


please explain params, use Google Style (you can configure in PyCharm) to explain params.

src/snowflake/snowpark/types.py

sfc-gh-lspiegelberg · 2024-05-30T16:14:06Z

src/snowflake/snowpark/types.py

-
+
+    def fill_ast(self, ast: proto.SpDataType) -> None:
+        raise NotImplementedError(f"{self.__class__.__name__} does not have an equivalent SpDataType yet")


this is a base method I assume? Maybe let's not use NotImplementedError, but instead ValueError and what to implement.

you can also use https://docs.python.org/3/library/abc.html

There are some intermediary types like _AtomicType(DataType) that should not be required to implement fill_ast I think so I will raise a ValueError instead.

sfc-gh-lspiegelberg · 2024-05-30T16:15:15Z

src/snowflake/snowpark/types.py

    pass


+# Ignoring pandas types in AST for now


you prob. can remove the comment, it will give a failure

sfc-gh-lspiegelberg · 2024-05-30T16:15:32Z

src/snowflake/snowpark/types.py



 class GeometryType(DataType):
    """Geometry data type. This maps to the GEOMETRY data type in Snowflake."""
-
+    # TODO: Create SpGeometryType IR enttiy, then uncomment


typo: enttiy -> entity

sfc-gh-lspiegelberg · 2024-05-30T16:17:01Z

src/snowflake/snowpark/types.py

@@ -115,6 +121,11 @@ def __hash__(self):
        if self.length == StringType._MAX_LENGTH:
            return StringType().__hash__()
        return super().__hash__()
+
+    def fill_ast(self, ast: proto.SpDataType) -> None:
+        # TODO: Create SpStringType IR entity with field "length: Int", and update


instead of all the TODOs as comment, could also use NotImplementedError with a comment on what to do. You could also define an error message

ErrorMessage.missing_ir('SpStringType IR needs field "length: Int"')

that would help a lot!

I agree, raising an error on the client side would probably be the best approach to raise the alarm on any Snowpark API elements with no representation in the IR.

I was afraid of blocking the client from working while we evaluate the IR changes, but I can turn the TODOs into NotImplementedError if that is okay.

Please see the draft pull request (still setting up the protobuf compiler to make sure the new AST is valid) I created addressing all of the TODOs here: https://github.com/snowflakedb/snowflake/pull/183083

sfc-gh-lspiegelberg

Great work! Left a few comments, I think it would be nice to have some testing added here.

src/snowflake/snowpark/_internal/ast_utils.py

sfc-gh-lspiegelberg · 2024-06-11T16:34:43Z

src/snowflake/snowpark/_internal/ast_utils.py

+
+
+# TODO: Instead of regexp, grab assign statments using Python ast library
+RE_SYMBOL_NAME = re.compile(r'^\s*([a-zA-Z_]\w*)\s*=.*$', re.DOTALL)


you prob. can move this constant within the get_symbol function

src/snowflake/snowpark/_internal/ast_utils.py

sfc-gh-lspiegelberg · 2024-06-11T16:35:14Z

src/snowflake/snowpark/_internal/ast_utils.py

+    ast.file = curr_frame.filename
+    ast.start_line = curr_frame.lineno
+
+    if sys.version_info[1] >= 11:


should we fill for the older Python versions with None or so?

The protobuf message created will default to None for these fields if not provided. Trying to assign them with None will raise an error due to the type mismatch.

src/snowflake/snowpark/_internal/proto/update-from-devvm.sh

sfc-gh-lspiegelberg · 2024-06-11T16:37:05Z

src/snowflake/snowpark/column.py

@@ -363,6 +431,7 @@ def __round__(self, n=None):
    def __hash__(self):
        return hash(self._expression)

+    # TODO: Implement AST generation for in_ / isin (relies on MultipleExpression and ScalarSubquery)


Let's file a JIRA under https://snowflakecomputing.atlassian.net/browse/SNOW-1473690, then TODO SNOW-12345

src/snowflake/snowpark/column.py

sfc-gh-lspiegelberg · 2024-06-11T16:38:33Z

src/snowflake/snowpark/types.py

+        Raises:
+            ValueError: If corresponding SpDataType IR entity is not available, raise an error
+        """
+        raise ValueError(


why not NotImplementedError?

sfc-gh-lspiegelberg · 2024-06-11T16:39:15Z

src/snowflake/snowpark/types.py


 # Data types
 class NullType(DataType):
    """Represents a null type."""

-    pass
+    def fill_ast(self, ast: proto.SpDataType) -> None:


should these be private methods as well? _fill_ast?

sfc-gh-lspiegelberg · 2024-06-11T16:40:01Z

tests/ast/run-unparser.sh

@@ -1,2 +1,2 @@
 #!/bin/bash


I think @sfc-gh-azwiegincew recently updated this, pls reflect!

Yes, this file is now deleted.

pls delete this file

sfc-gh-lspiegelberg

some nits, great work! Let's get this in.

sfc-gh-azwiegincew

Gorgeous tests! Just a few nits.

sfc-gh-azwiegincew · 2024-06-11T21:40:44Z

src/snowflake/snowpark/_internal/ast_utils.py

+import snowflake.snowpark._internal.proto.ast_pb2 as proto
+
+NoneType = type(None)
+PYTHON_TO_AST_CONST_MAPPINGS = {


empty line to separate 13 and 14

I'm not a huge fan of PYTHON_TO_AST_CONST_MAPPINGS. It's not reused anywhere, and you have complications such as the bytearray case. I would suggest removing this and just adding the cases directly in your one function that uses this global.

src/snowflake/snowpark/_internal/ast_utils.py

sfc-gh-azwiegincew · 2024-06-11T21:49:02Z

src/snowflake/snowpark/dataframe.py

@@ -85,6 +85,11 @@
    Unpivot,
    ViewType,
 )
+from snowflake.snowpark._internal.ast_utils import (
+    setattr_if_not_none,


sfc-gh-azwiegincew · 2024-06-11T21:51:19Z

tests/ast/data/col_alias.test

@@ -0,0 +1,20 @@
+## TEST CASE


empty line after this for symmetry with EXPECTED OUTPUT – in all the new tests

sfc-gh-azwiegincew · 2024-06-11T21:52:26Z

tests/ast/data/col_alias.test

@@ -0,0 +1,20 @@
+## TEST CASE
+from snowflake.snowpark.functions import col


I would suggest adding import snowflake.snowpark.functions in the test driver and using functions.col in your tests.

Or because col is so special (i.e. common), adding that import as well (keeping .functions for other functions in the future).

sfc-gh-azwiegincew · 2024-06-11T21:55:34Z

tests/ast/run-unparser.sh

@@ -1,2 +1,2 @@
 #!/bin/bash


Yes, this file is now deleted.

sfc-gh-azwiegincew · 2024-06-11T21:55:54Z

tests/ast/test_ast_driver.py

@@ -23,29 +25,33 @@ class TestCase:

 def parse_file(file):
    """Parses a test case file."""
-    with open(file, "r", encoding="utf-8") as f:
+    with open(file, encoding="utf-8") as f:


Why did you remove r?

sfc-gh-lspiegelberg · 2024-06-12T17:17:46Z

src/snowflake/snowpark/_internal/ast_utils.py

+    return curr_frame
+
+
+# TODO: Instead of regexp, grab assign statments using Python ast library


can we add a JIRA?

sfc-gh-lspiegelberg · 2024-06-12T17:18:16Z

src/snowflake/snowpark/_internal/proto/ast_pb2.py

@@ -11,582 +11,589 @@
 _sym_db = _symbol_database.Default()



has this been updated with the most recent one from the server / mono repo branch?

Yes it is the most recent one

sfc-gh-lspiegelberg · 2024-06-12T17:19:31Z

src/snowflake/snowpark/_internal/server_connection.py

@@ -724,7 +724,7 @@ def ast_query(self, request_id__ast) -> Any:
        return self._conn._rest.request(
            f"/queries/v1/query-request?requestId={request_id}", req
        )


remove change here (use git restore -s origin/server-side-snowpark src/snowflake/snowpark/_internal/server_connection.py or so)

sfc-gh-lspiegelberg · 2024-06-12T17:22:14Z

tests/ast/update-unparser.sh

+# I.e., add the following lines to ~/.ssh/config
+# Host devvm
+#	  HostName sdp-devvm-<ldap user>
+


reuse same logic for devvm / workspace incl. example from other script here?

sfc-gh-lspiegelberg

one more pass and then this should be good to go

sfc-gh-vbudati · 2024-06-12T17:23:08Z

src/snowflake/snowpark/mock/_connection.py

@@ -662,7 +662,6 @@ def get_result_query_id(self, plan: SnowflakePlan, **kwargs) -> str:
        result_set, _ = self.get_result_set(plan, to_iter=True, **kwargs)
        return result_set["sfqid"]

-


I think this was a side effect of running the precommit fix_lint

sfc-gh-vbudati · 2024-06-12T17:34:20Z

src/snowflake/snowpark/column.py

+
+        Args:
+            property (str): IR entity representing the unary operation in the Column AST
+            operator (UnaryExpression): Snowpark unary operator to buidl the Expression instance


Suggested change

operator (UnaryExpression): Snowpark unary operator to buidl the Expression instance

operator (UnaryExpression): Snowpark unary operator to build the Expression instance

nit

sfc-gh-vbudati

Looks great! I think it should be good to merge once you delete the file Arthur and Leonhard mentioned - tests/ast/run-unparser.sh

sfc-gh-heshah requested review from sfc-gh-lspiegelberg, sfc-gh-azwiegincew and sfc-gh-oplaton May 29, 2024 16:35

sfc-gh-heshah force-pushed the server-side-snowpark-colops branch from fa728c0 to 646f000 Compare May 29, 2024 16:39

sfc-gh-oplaton reviewed May 29, 2024

View reviewed changes

sfc-gh-oplaton approved these changes May 30, 2024

View reviewed changes

sfc-gh-oplaton reviewed May 30, 2024

View reviewed changes

sfc-gh-lspiegelberg reviewed May 30, 2024

View reviewed changes

src/snowflake/snowpark/types.py Outdated Show resolved Hide resolved

sfc-gh-lspiegelberg reviewed May 30, 2024

View reviewed changes

sfc-gh-heshah force-pushed the server-side-snowpark-colops branch from 5a6e8da to 54a0bee Compare May 31, 2024 16:21

Column ops implementation first draft

1bd82b8

sfc-gh-lspiegelberg reviewed Jun 11, 2024

View reviewed changes

src/snowflake/snowpark/_internal/ast_utils.py Show resolved Hide resolved

sfc-gh-lspiegelberg reviewed Jun 11, 2024

View reviewed changes

src/snowflake/snowpark/_internal/ast_utils.py Show resolved Hide resolved

sfc-gh-lspiegelberg reviewed Jun 11, 2024

View reviewed changes

src/snowflake/snowpark/_internal/proto/update-from-devvm.sh Show resolved Hide resolved

sfc-gh-lspiegelberg reviewed Jun 11, 2024

View reviewed changes

src/snowflake/snowpark/column.py Show resolved Hide resolved

sfc-gh-lspiegelberg reviewed Jun 11, 2024

View reviewed changes

sfc-gh-azwiegincew approved these changes Jun 11, 2024

View reviewed changes

sfc-gh-heshah added 4 commits June 11, 2024 17:32

Address PR comments (no datetime IR changes yet)

3359272

merge server side branch

9bdee20

new datetime ASTs, fill out literal tests

2db418d

fix linting

70327b8

sfc-gh-heshah requested a review from sfc-gh-vbudati June 12, 2024 15:37

sfc-gh-lspiegelberg reviewed Jun 12, 2024

View reviewed changes

sfc-gh-vbudati reviewed Jun 12, 2024

View reviewed changes

sfc-gh-vbudati approved these changes Jun 12, 2024

View reviewed changes

PR fixes

6c98afa

sfc-gh-lspiegelberg approved these changes Jun 12, 2024

View reviewed changes

sfc-gh-heshah merged commit bce97c0 into server-side-snowpark Jun 12, 2024
14 of 16 checks passed

sfc-gh-heshah deleted the server-side-snowpark-colops branch June 12, 2024 17:52

github-actions bot locked and limited conversation to collaborators Jun 12, 2024



		def fill_ast(self, ast: proto.SpDataType) -> None:
		raise NotImplementedError(f"{self.__class__.__name__} does not have an equivalent SpDataType yet")



		# TODO: Instead of regexp, grab assign statments using Python ast library
		RE_SYMBOL_NAME = re.compile(r'^\s([a-zA-Z_]\w)\s=.$', re.DOTALL)

		@@ -0,0 +1,20 @@
		## TEST CASE
		from snowflake.snowpark.functions import col

		return curr_frame


		# TODO: Instead of regexp, grab assign statments using Python ast library

		@@ -662,7 +662,6 @@ def get_result_query_id(self, plan: SnowflakePlan, **kwargs) -> str:
		result_set, _ = self.get_result_set(plan, to_iter=True, **kwargs)
		return result_set["sfqid"]

	operator (UnaryExpression): Snowpark unary operator to buidl the Expression instance
	operator (UnaryExpression): Snowpark unary operator to build the Expression instance

Column ops implementation first draft #1696

Column ops implementation first draft #1696

Conversation

sfc-gh-heshah commented May 29, 2024

sfc-gh-oplaton left a comment • edited Loading

Choose a reason for hiding this comment

sfc-gh-oplaton commented May 30, 2024

sfc-gh-heshah commented May 30, 2024

sfc-gh-oplaton May 30, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfc-gh-heshah May 30, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfc-gh-heshah May 31, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfc-gh-lspiegelberg May 30, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfc-gh-heshah May 30, 2024 • edited Loading

Choose a reason for hiding this comment

sfc-gh-lspiegelberg left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfc-gh-lspiegelberg left a comment

Choose a reason for hiding this comment

sfc-gh-azwiegincew left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfc-gh-lspiegelberg Jun 12, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfc-gh-lspiegelberg left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfc-gh-oplaton left a comment •

edited

Loading

sfc-gh-oplaton May 30, 2024 •

edited

Loading

sfc-gh-heshah May 30, 2024 •

edited

Loading

sfc-gh-heshah May 31, 2024 •

edited

Loading

sfc-gh-lspiegelberg May 30, 2024 •

edited

Loading

sfc-gh-heshah May 30, 2024 •

edited

Loading

sfc-gh-lspiegelberg Jun 12, 2024 •

edited

Loading