feat(tests): ckzg functions for peerdas tests #1614

felix314159 · 2025-05-16T09:51:36Z

🗒️ Description

Added an eest wrapper for ckzg functions we will probably use in our tests.
A few open questions:

Their code relies on trusted_setup.txt, I just import in the functions that need it and it assumes it is in the same folder. Is this compatible with our pytest framework or do we need to change how this file is accessed
I currently don't understand why I can only create blobs that hold values 0x73 or lower, otherwise calling compute_cells complains
It seems to me that every cell holds the exact same commitment. So you commit to your blob once and then copy paste the 48 byte commitment into every cell? Seems odd/wasteful to me, but could be the case

Next steps:

Use the functions in here to prepare a few static examples of blob/commitment/cells/proofs and store them as JSON (yml and me don't talk no more) in our repo. Then we can re-use these static examples in our tests instead of having to repeatedly create them at runtime.

🔗 Related Issues

✅ Checklist

All: Set appropriate labels for the changes.
All: Considered squashing commits to improve commit history.
All: Added an entry to CHANGELOG.md.
All: Considered updating the online docs in the ./docs/ directory.
Tests: All converted JSON/YML tests from ethereum/tests have been added to converted-ethereum-tests.txt.
Tests: A PR with removal of converted JSON/YML tests from ethereum/tests have been opened.
Tests: Included the type and version of evm t8n tool used to locally execute test cases: e.g., ref with commit hash or geth 1.13.1-stable-3f40e65.
Tests: Ran mkdocs serve locally and verified the auto-generated docs for new tests in the Test Case Reference are correctly formatted.
Tests: For PRs implementing a missed test case, update the post-mortem document to add an entry the list.

kevaundray · 2025-05-16T15:18:42Z

I currently don't understand why I can only create blobs that hold values 0x73 or lower, otherwise calling compute_cells complains

Blobs are not entirely opaque sequence of bytes. You can think of a blob as being 32 byte chunks where each 32 byte must represent an integer less than q, where q is the bls12-381 scalar field.

The "easier" way to always have it pass is to generate a 31 byte integer and then serialize that into 32 bytes. This is because every 31 byte integer will be less than q since q is roughly a 32 byte integer.

It seems to me that every cell holds the exact same commitment. So you commit to your blob once and then copy paste the 48 byte commitment into every cell? Seems odd/wasteful to me, but could be the case

Yeah, in the original API it was deduplicated, but then you need another index to index the deduplicated blob because the verify function can take in multiple commitments -- I found this to be more confusing though

jtraglia

Hey @felix314159, nice work. I would suggest only loading the trusted setup once and removing the wrapper functions. I don't believe these really add anything, right? Ckzg will do these checks too.

jtraglia · 2025-05-16T15:26:04Z

tests/osaka/eip7594_peerdas/spec.py

+    # below are mine, not sure if they are defined in specs somewhere
+    KZG_PROOF_LENGTH = 48
+    KZG_COMMITMENT_LENGTH = 48
+    CELL_LENGTH = 2048


There are constants for these in the specs! Please see:

BYTES_PER_COMMITMENT

BYTES_PER_PROOF

BYTES_PER_CELL

These are not exposed to the bindings, but maybe they should be. I can do this later if you want.

jtraglia · 2025-05-16T15:34:05Z

tests/osaka/eip7594_peerdas/helper_functions.py

+    if int(hex_byte, 16) > 115:
+        # TODO: figure out why this happens
+        print("For some reason bytes larger than this will lead to errors later in compute_cells(), so it's not allowed for now.")  # noqa: E501
+        return None
+
+    hex_byte = hex_byte.lower()


Yes, this is because of BLS_MODULUS. The uint256 value of each field element must be less than this. Given that Python supports large integers, I think it would be best if this function took a seed value, then generated FIELD_ELEMENTS_PER_BLOB random field elements and joined them together to make the blob. Something like:

import random seed = 42 random.seed(seed) BLS_MODULUS = 0x73eda753299d7d483339d80809a1d80553bda402fffe5bfeffffffff00000001 ints = [random.randrange(BLS_MODULUS) for _ in range(FIELD_ELEMENTS_PER_BLOB)] encoded = [i.to_bytes(BYTES_PER_FIELD_ELEMENT, "big") for i in ints] blob = b"".join(encoded)

Agreed, great suggestion, and the seed can be an int to replace the hex_byte: str in the original function.

jtraglia · 2025-05-16T15:35:20Z

tests/osaka/eip7594_peerdas/helper_functions.py

+    expected_blob_length = 2**int(math.log2(Spec.BYTES_PER_BLOB))
+    assert len(blob) == expected_blob_length, f"Expected blob of length {expected_blob_length} but got blob of length {len(blob)}"  # noqa: E501


This feels a strange. Why don't we just do assert len(blob) == Spec.BYTES_PER_BLOB?

jtraglia · 2025-05-16T15:36:06Z

tests/osaka/eip7594_peerdas/helper_functions.py

+    assert len(blob) == expected_blob_length, f"Expected blob of length {expected_blob_length} but got blob of length {len(blob)}"  # noqa: E501
+
+    # calculate commitment
+    ts = ckzg.load_trusted_setup("trusted_setup.txt", 0)


This is a slow operation. It would be best if this were only done once. Maybe in some setup function?

We should add the trusted setup as a module constant at the beginning of the file:

from os.path import realpath from pathlib import Path TRUSTED_SETUP_FILE_NAME = "trusted_setup.txt" TRUSTED_SETUP_PATH = Path(realpath(__file__)).parent / TRUSTED_SETUP_FILE_NAME TRUSTED_SETUP = ckzg.load_trusted_setup(str(TRUSTED_SETUP_PATH), 0)

jtraglia · 2025-05-16T15:37:30Z

tests/osaka/eip7594_peerdas/helper_functions.py

+def eest_compute_cells(blob: bytes) -> list[int]:
+    """Take a blob and returns a list of cells that are derived from this blob."""
+    ts = ckzg.load_trusted_setup("trusted_setup.txt", 0)
+
+    cells = ckzg.compute_cells(blob, ts)
+
+    assert len(cells) == 128
+
+    return cells


Several of these wrapper functions don't appear to be very useful. Could we not just call ckzg directly?

jtraglia · 2025-05-16T15:39:07Z

tests/osaka/eip7594_peerdas/helper_functions.py

+    return commitment
+
+
+def eest_compute_cells(blob: bytes) -> list[int]:


I believe this return type isn't correct. It doesn't return a list of ints. It returns a list of bytes.

marioevz

This is excellent, thanks for putting this together!

Just a couple of comments to make some parts of it more pythonic, just overall we should try to avoid to use strings when other more appropriate primitive python types can be used.

marioevz · 2025-05-16T17:33:31Z

tests/osaka/eip7594_peerdas/helper_functions.py

+    if int(hex_byte, 16) > 115:
+        # TODO: figure out why this happens
+        print("For some reason bytes larger than this will lead to errors later in compute_cells(), so it's not allowed for now.")  # noqa: E501
+        return None
+
+    hex_byte = hex_byte.lower()


Agreed, great suggestion, and the seed can be an int to replace the hex_byte: str in the original function.

marioevz · 2025-05-16T17:36:45Z

tests/osaka/eip7594_peerdas/helper_functions.py

+class PersistentBlobGenerator:
+    # PersistentBlobGenerator("4a") creates blob_4a.json in cwd and contains blob, commitment, cells and proofs
+    encoding: str = "utf-8"
+


Pydantic can be of excellent help here:

from typing import List from pydantic import BaseModel from ethereum_test_base_types import Bytes class Blob(BaseModel): name: str blob: Bytes commitments: List[Bytes] cells: List[Bytes] proofs: List[Bytes]

And this can be then converted to json using model_dump or model_dump_json, and the to_json and from_json methods can be removed.

marioevz · 2025-05-16T17:38:59Z

tests/osaka/eip7594_peerdas/helper_functions.py

+    def __init__(self, singular_byte: str):
+        # safely derive blob from input
+        blob: bytes | None = generate_blob_from_hex_byte(singular_byte)
+        assert blob is not None, f"PersistentBlobGenerator received invalid input: {singular_byte}"
+
+        if "0x" in singular_byte:
+            singular_byte = singular_byte.replace("0x", "", 1)
+
+        commitment: bytes = eest_blob_to_kzg_commitment(blob)
+        cells, proofs = eest_compute_cells_and_kzg_proofs(blob)
+
+        # populate instance
+        self.name: str = "blob_" + singular_byte
+        self.blob: bytes = blob
+        self.commitments: list[bytes] = [commitment] * len(cells)
+        self.cells: list[bytes] = cells
+        self.proofs: list[bytes] = proofs


If we change this class to pydantic, the only other change is that this init function needs to be a class method instead.

@classmethod def from_index(cls, index: int): ... return cls(blob, proofs, ...)

or similar.

danceratopz · 2025-05-19T08:44:02Z

uv.lock

@@ -1,10 +1,6 @@
 version = 1


Hey @felix314159, as we discussed, can you please revert the updates to all packages and only modify the required packages.

Instructions available here:
https://eest.ethereum.org/main/dev/deps_and_packaging/

marioevz

Left a lot of comments but I think these are necessary to future-proof the new code introduced 👍

Thanks!

marioevz · 2025-05-26T18:20:06Z

src/ethereum_test_types/blob.py

+
+TRUSTED_SETUP_FILE_NAME = "blob_trusted_setup.txt"
+TRUSTED_SETUP_PATH = Path(realpath(__file__)).parent / TRUSTED_SETUP_FILE_NAME
+TRUSTED_SETUP = ckzg.load_trusted_setup(str(TRUSTED_SETUP_PATH), 0)


Convert to ClassVar inside of Blob that is loaded upon first use:

class Blob(CamelModel): ... _trusted_setup: ClassVar[Any | None] = None ... @classmethod def trusted_setup(cls): ... if cls._trusted_setup is None: # load trusted setup return cls._trusted_setup

Good idea but I will make this function void, no need to return the trusted setup if it is already accessible

marioevz · 2025-05-26T18:41:15Z

src/ethereum_test_types/blob.py

+        return would_be_static_blob_path
+
+    @staticmethod
+    def NewBlob(fork: Fork, seed: int = 0, timestamp: int = 0) -> "Blob":  # noqa: N802


Suggested change

def NewBlob(fork: Fork, seed: int = 0, timestamp: int = 0) -> "Blob": # noqa: N802

def from_seed(fork: Fork, seed: int = 0, timestamp: int = 0) -> "Blob":

I think this name is a bit more accurate and follows python method naming better.

Yes we should use snake case I agree. However, I strongly disagree calling it from_seed as the seed is just an optional parameter. You probably wanted to name it from_fork.

marioevz · 2025-05-26T18:49:37Z

src/ethereum_test_types/blob.py

+        fork_str: str = fork.name().lower()
+
+        # if this blob already exists then load from file
+        blob_location: Path = Blob.get_filepath(fork_str, seed, timestamp)


Passing the timestamp is a good idea, but we could improve it a bit further here since I think using the timestamp to get the filename could result in too many duplicates:

Suggested change

fork_str: str = fork.name().lower()

# if this blob already exists then load from file

blob_location: Path = Blob.get_filepath(fork_str, seed, timestamp)

fork = fork.fork_at(timestamp=timestamp) # This resolves to the correct fork in fork transition tests, but it's no-op in normal forks

fork_str: str = fork.name().lower()

# if this blob already exists then load from file

blob_location: Path = Blob.get_filepath(fork_str, seed)

And then we just remove the timestamp parameter from get_filepath.

marioevz · 2025-05-26T18:54:29Z

src/ethereum_test_types/blob.py

+        def generate_blob_data(rng_seed: int = 0) -> Bytes:
+            """Calculate blob data deterministically via provided seed."""
+            # apply RNG seed
+            random.seed(rng_seed)


We can change this to use a local thread-safe instance:

Suggested change

random.seed(rng_seed)

rng = random.Random(rng_seed)

Very good catch, thanks

marioevz · 2025-05-26T19:00:52Z

src/ethereum_test_types/blob.py

+        assert fork.engine_get_blobs_version() is not None, (
+            f"You provided fork {fork.name()} but it does not support blobs!"
+        )  # TODO: why does mypy say '"type[BaseFork]" has no attribute "engine_get_blobs_version"'?


Can instead use supports_blobs here:

Suggested change

assert fork.engine_get_blobs_version() is not None, (

f"You provided fork {fork.name()} but it does not support blobs!"

) # TODO: why does mypy say '"type[BaseFork]" has no attribute "engine_get_blobs_version"'?

assert fork.supports_blobs(), (

f"Provided fork {fork.name()} but it does not support blobs!"

)

marioevz · 2025-05-26T19:37:41Z

src/ethereum_test_types/blob.py

+            detected_fork = parts[1]
+            detected_seed = parts[2]
+            detected_timestamp = parts[3].removesuffix(".json")
+            assert detected_seed.isdigit(), (
+                f"Failed to extract seed from filename. Ended up with seed {detected_seed} given filename {file_path}"  # noqa: E501
+            )
+            assert detected_timestamp.isdigit(), (
+                f"Failed to extract timestamp from filename. Ended up with timestamp {detected_timestamp} given filename {file_path}"  # noqa: E501
+            )
+            file_path = Blob.get_filepath(
+                detected_fork, int(detected_seed), int(detected_timestamp)
+            )


Feels unnecessary, is it just for debugging purposes?

If the test writer tries to import an existing blob via filename this code ensures that the name of the blob file is valid (as in matches the expected format). I would prefer to keep this, if someone tries to import an invalidly named blob this way we at least get a descriptive error message

Edit: it was removed because entire function was reworked due to suggested filename change

marioevz · 2025-05-26T19:40:30Z

src/ethereum_test_types/blob.py

+        assert self.fork in ["osaka"], (
+            f"verify_cell_kzg_proof_batch() is not available for fork: {self.fork}"
+        )


Suggested change

assert self.fork in ["osaka"], (

f"verify_cell_kzg_proof_batch() is not available for fork: {self.fork}"

)

assert self.fork.blob_kzg_cell_proofs() > 0, (

f"verify_cell_kzg_proof_batch() is not available for fork: {self.fork}"

)

marioevz · 2025-05-26T19:42:25Z

src/ethereum_test_types/blob.py

+        assert len(self.cells) == 128, (
+            f"You are supposed to pass a full cell list with 128 elements to this function, but got list of length {len(self.cells)}"  # noqa: E501
+        )


Suggested change

assert len(self.cells) == 128, (

f"You are supposed to pass a full cell list with 128 elements to this function, but got list of length {len(self.cells)}" # noqa: E501

)

assert len(self.cells) == self.fork.blob_kzg_cell_proofs(), (

f"You are supposed to pass a full cell list with {self.fork.blob_kzg_cell_proofs()} elements to this function, but got list of length {len(self.cells)}" # noqa: E501

)

Same for the rest of the function, it can be adapted in the same way.

marioevz · 2025-05-26T19:45:08Z

src/ethereum_test_types/eest_types.py

I really don't think this rename is necessary, were you getting an error somehow with the original name? If so, could you share so we can help debug pls?

We are shadowing the stdlib "types" when we have types.py. This means any import that implictly requires stdlib types will fail with ImportError: cannot import name 'GenericAlias' from partially initialized module 'types' (most likely due to a circular import). Try the following once with types.py and then again with eest_types.py:

echo 'from enum import Enum' > ./src/ethereum_test_types/abcd.py uv run python ./src/ethereum_test_types/abcd.py

Good catch :D I think this is particular to using uv run python ./src/... but still I think it should be ok to finally split types.py into different files.

But I think that one deserves its own PR, wdyt?

marioevz · 2025-05-26T19:46:06Z

tests/cancun/eip4844_blobs/static_blobs/blob_cancun_0_0.json

When we take the local cache approach, we should remove these static blob files from the branch.

Yes, let's remove them

marioevz

This is taking shape and I think it's closer to being done! 🚀

I've left a couple more minor comments.

Thanks for putting the effort into this so far! Let's try to get it through the finish line :)

marioevz · 2025-05-30T17:05:11Z

src/ethereum_test_forks/__init__.py

+from typing import Literal
+


Seems unnecessary.

Suggested change

from typing import Literal

marioevz · 2025-05-30T17:06:20Z

src/ethereum_test_types/eest_types.py

File is no longer necessary.

marioevz · 2025-05-30T17:07:05Z

tests/cancun/eip4844_blobs/test_blob_txs_full.py

@@ -22,6 +23,8 @@
    Transaction,
    TransactionException,
 )
+from ethereum_test_types import Blob
+from tests.cancun.eip4844_blobs.common import blobs_to_transaction_input


Suggested change

from tests.cancun.eip4844_blobs.common import blobs_to_transaction_input

from .common import blobs_to_transaction_input

marioevz · 2025-05-30T17:09:52Z

whitelist.txt

I think this file got sorted at some point? Could you try to reset the file to the version in the base branch (should be main) and then just add the lines that are required for this PR?

marioevz · 2025-05-30T17:14:11Z

tests/cancun/eip4844_blobs/test_blob_txs_full.py

-                        kzg_proof=INF_POINT,
-                    ),
+                    # you can provide the blob via file name
+                    Blob.from_file("blob_cancun_0_0"),


We should only use the more-abstract from_fork method in tests.

marioevz · 2025-05-30T17:15:51Z

src/ethereum_test_types/blob_trusted_setup.txt

Minor nit-pick but could we rename this file to "kzg_trusted_setup.txt"?

marioevz · 2025-05-30T19:17:24Z

src/ethereum_test_types/blob_types.py

+
+    versioned_hash: Hash
+    name: str  # blob_<fork>_<seed>
+    fork: str


Suggested change

fork: str

fork: Fork

After #1686, and can also delete all str<->Fork conversions everywhere in this file.

A side-effect of this change is that pydantic now struggles to serialize and deserialize Blob object (when we sticked to basic python types like fork: str sth like Blob.model_validate_json(json_str) and self.model_dump_json() 'just worked' but now we have to add hard-to-read logic to tell pydantic how to handle the Fork type at serialization and deserialization:

@field_validator("fork", mode="before") @classmethod def validate_fork(cls, v): """ When reading JSON file and trying to go back from fork string to fork object we must tell pydantic how to do this. """ if isinstance(v, str): return fork_string_to_object(v) return v @field_serializer("fork") def serialize_fork(self, fork: Fork) -> str: """ When trying to serialize a Blob object into a JSON file we must tell pydantic how to do this. """ return fork.name()

felix314159 · 2025-06-02T10:28:40Z

@marioevz Please look at the latest commit, I added fork_at() like you described.

Edit: I think everything is working now, can be reviewed & merged

marioevz

LGTM!

Just a couple of comments but we can merge after those 👍

marioevz · 2025-06-04T11:39:51Z

tests/cancun/eip4844_blobs/test_blob_txs_full.py

-                        kzg_commitment=INF_POINT,
-                        kzg_proof=INF_POINT,
-                    ),
+                    Blob.from_fork(Cancun),


Suggested change

Blob.from_fork(Cancun),

Blob.from_fork(fork),

We can just pass the fork here to avoid constraining the test to a hard-coded fork.

marioevz · 2025-06-04T11:41:35Z

tests/cancun/eip4844_blobs/test_blob_txs_full.py

-                        kzg_commitment=INF_POINT,
-                        kzg_proof=INF_POINT,
-                    )
+                    Blob.from_fork(Cancun),


Suggested change

Blob.from_fork(Cancun),

Blob.from_fork(fork),

Same here.

marioevz · 2025-06-04T11:41:43Z

tests/cancun/eip4844_blobs/test_blob_txs_full.py

-                        kzg_commitment=INF_POINT,
-                        kzg_proof=INF_POINT,
-                    )
+                    Blob.from_fork(Cancun),


Suggested change

Blob.from_fork(Cancun),

Blob.from_fork(fork),

marioevz · 2025-06-04T11:42:03Z

tests/cancun/eip4844_blobs/test_blob_txs_full.py

 from ethereum_test_forks import Fork
+from ethereum_test_forks.forks.forks import Cancun


Suggested change

from ethereum_test_forks.forks.forks import Cancun

If the rest of the comments are applied.

…many blobs (7vs9, geth)

jtraglia reviewed May 16, 2025

View reviewed changes

marioevz reviewed May 16, 2025

View reviewed changes

danceratopz reviewed May 19, 2025

View reviewed changes

danceratopz changed the title ~~test(add): ckzg functions for peerdas tests~~ feat(tests): ckzg functions for peerdas tests May 19, 2025

danceratopz added scope:tests Scope: Changes EL client test cases in `./tests` type:feat type: Feature fork:osaka Osaka hardfork labels May 19, 2025

felix314159 force-pushed the peerdas-kzg-added branch from d33b90d to ccfedf8 Compare May 26, 2025 10:44

marioevz reviewed May 26, 2025

View reviewed changes

felix314159 force-pushed the peerdas-kzg-added branch from a7861a5 to 35ffb69 Compare May 30, 2025 08:28

marioevz reviewed May 30, 2025

View reviewed changes

felix314159 force-pushed the peerdas-kzg-added branch from 4bf0f25 to f07306e Compare June 4, 2025 09:53

felix314159 requested a review from marioevz June 4, 2025 10:20

marioevz approved these changes Jun 4, 2025

View reviewed changes

felix314159 force-pushed the peerdas-kzg-added branch from 96fb60d to 31a541c Compare June 4, 2025 13:48

felix314159 force-pushed the peerdas-kzg-added branch from 4a23956 to 1e350f2 Compare June 12, 2025 09:28

felix314159 and others added 2 commits June 13, 2025 10:57

added kzg functions for peerdas tests

d58f486

fix: use proofs only instead of full cells

d7aa193

felix314159 force-pushed the peerdas-kzg-added branch from 841673b to 329e19f Compare June 13, 2025 10:57

felix314159 requested a review from marioevz June 13, 2025 12:23

tests passing except for third one, too much gas (nethermind) or too …

abab88c

…many blobs (7vs9, geth)

felix314159 force-pushed the peerdas-kzg-added branch from 18393da to abab88c Compare June 13, 2025 13:01

		expected_blob_length = 2**int(math.log2(Spec.BYTES_PER_BLOB))
		assert len(blob) == expected_blob_length, f"Expected blob of length {expected_blob_length} but got blob of length {len(blob)}" # noqa: E501

		return commitment


		def eest_compute_cells(blob: bytes) -> list[int]:

	def NewBlob(fork: Fork, seed: int = 0, timestamp: int = 0) -> "Blob": # noqa: N802
	def from_seed(fork: Fork, seed: int = 0, timestamp: int = 0) -> "Blob":

	from tests.cancun.eip4844_blobs.common import blobs_to_transaction_input
	from .common import blobs_to_transaction_input

		from ethereum_test_forks import Fork
		from ethereum_test_forks.forks.forks import Cancun

feat(tests): ckzg functions for peerdas tests #1614

Are you sure you want to change the base?

feat(tests): ckzg functions for peerdas tests #1614

Uh oh!

Conversation

felix314159 commented May 16, 2025

🗒️ Description

🔗 Related Issues

✅ Checklist

Uh oh!

kevaundray commented May 16, 2025

Uh oh!

jtraglia left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

marioevz left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

marioevz left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

felix314159 May 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

felix314159 May 27, 2025 •

edited

Loading

felix314159 Jun 2, 2025 •

edited

Loading

felix314159 commented Jun 2, 2025 •

edited

Loading