Structlog processor

[apelsynca]

Short summary:

1: changed relative imports to absolute.

2:
added LogTideProcessor based on structlog, and tests for that. (and fixed tests)
does not requires structlog to be installed.

usage example:

from logtide_sdk.structlog import LogTideProcessor

# then somewhere in code:

structlog.configure_once(processors=[
    ...,
    ...,
    LogtideProcessor(client=logtide_client, service="service_name_string")
]

[Polliog]

@
Thanks for this! A few things before merge:

1. Blocker: middleware/__init__.py uses wrong absolute imports — logtide_sdk.flask should be logtide_sdk.middleware.flask (same for django/fastapi/starlette). Since they are in try/except ImportError: pass, all middleware silently fail to import (__all__ becomes []).

2. The pydantic quickfix does not fully work. Adding default=repr only fixes the size calc. _process_value leaves unknown objects untouched, and the actual send (_send_logs/_send_batch) posts with json=payload and no default=, so it still raises TypeError at flush time. I would prefer your other idea: do the repr/serialization inside LogTideProcessor so metadata is JSON-safe before it reaches the SDK. Let us drop the json.dumps(..., default=repr) change.

3. frozenset[str] annotation in structlog.py breaks Python 3.8 (PEP 585 needs 3.9+), and we declare requires-python = ">=3.8". Use typing.FrozenSet or from __future__ import annotations.

The structlog processor itself looks good 👍
@

[apelsynca]

Hi, tomorrow will try to do that.

Also i forgore to update the ci.yml (will do)

[Polliog]

drop Python 3.8 and set requires-python = ">=3.10". 3.9 is EOL too (Oct 2025), and 3.10 covers Ubuntu 22.04 LTS since we are still alpha the compat cost is zero, and it removes the functools.cache / frozenset[str] 3.8 breakage for good.

Needs: requires-python, classifiers (drop 3.8/3.9, add 3.10-3.13), and the CI matrix.

Separately, one real bug: json_encoder.py uses chain(...) in _pandas_data_frame_encoder but never imports it add from itertools import chain. Large DataFrames currently raise a NameError that gets swallowed into a repr() fallback.

[Polliog]

Almost there 👍 Two things left:

1. CI matrix is out of sync .github/workflows/ci.yml still runs ["3.8", "3.9", ...], which now contradicts requires-python = ">=3.10" (pip will refuse to install / import will fail on those runners → red CI). Drop 3.8 and 3.9, and add 3.14 to match the classifiers.

2. chain still unimported json_encoder.py:242 uses chain(...) in _pandas_data_frame_encoder but there is no from itertools import chain. Was not touched by the version bump.

[Polliog]

All green now thanks a lot for the thorough work on this! 🙏 The structlog processor + JSON encoder are a really nice addition, and dropping the EOL Python versions was the right call. Merging.