Skip to content

[cleanup] Avoid importing pandas at ccflow import time #272

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
timkpaine merged 1 commit into from
Sep 21, 2026
Merged

[cleanup] Avoid importing pandas at ccflow import time #272

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 7 additions & 15 deletions ccflow/utils/tokenize.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,6 +159,13 @@ def _normalize_date(obj):

@normalize_token.register(datetime)
def _normalize_datetime(obj):
cls = type(obj)
# pandas.Timestamp subclasses datetime and must not share a token with an equal plain datetime.
# Matching by name keeps `import ccflow` from paying for pandas just to register one handler.
# DataFrame/Series/Index have no structural handler and fall through to cloudpickle; adding them
# (as dask does) would require real lazy registration rather than this name check.
if cls is not datetime and any(c.__name__ == "Timestamp" and _module_in(c.__module__, ("pandas",)) for c in cls.__mro__):
return ("pd_timestamp", obj.isoformat())
return ("datetime", obj.isoformat())


Expand Down Expand Up @@ -371,22 +378,7 @@ def _normalize_np_scalar(obj):
return ("np_scalar", str(type(obj).__name__), obj.item())


def _register_pandas() -> None:
try:
import pandas as pd
except ImportError: # pragma: no cover
return

# Only Timestamp has a structural handler; DataFrame/Series/Index fall through to the cloudpickle
# fallback. That works today but is fragile across pandas version upgrades — a follow-up could add
# structural handlers (matching dask) for stability and a perf win on large frames.
@normalize_token.register(pd.Timestamp)
def _normalize_pd_timestamp(obj):
return ("pd_timestamp", obj.isoformat())


_register_numpy()
_register_pandas()


def tokenize(*args: Any, **kwargs: Any) -> str:
Expand Down
Loading