AxelNoun opened a new issue, #921:
URL: https://github.com/apache/tsfile/issues/921

   # Python SDK: can pyarrow>=18,<20 on 3.10–3.13 be widened?
   
   **Status: hypothesis.** We do not know why this pin exists; asking whether 
it can be relaxed.
   
   Measured on **2026-08-25** from `https://pypi.org/pypi/tsfile/json` (see
   `pypi_tsfile_matrix.md`). Latest SDK: **2.4.0**. Pins identical for 2.3.0 / 
2.3.1 / 2.4.0.
   
   `datasets` (since 4.8.5) requires `pyarrow>=21`. The Apache TsFile Python 
SDK declares:
   
   | Python | SDK `pyarrow` | Intersection with `datasets` |
   |---|---|---|
   | 3.9 | `>=16,<18` | n/a (`datasets` needs ≥ 3.10) |
   | 3.10–3.13 | `>=18,<20` | empty → `ResolutionImpossible` |
   | 3.14+ | `>=22,<24` | `[22, 24)` |
   
   Empirical (2026-08-25): `pip install --dry-run "datasets==5.0.1" 
"tsfile==2.4.0"` on
   Python 3.12 → `ResolutionImpossible`. Wheels exist for cp39–cp314; the hole 
is the pin.
   
   The same SDK already accepts pyarrow 22/23 on 3.14, which *suggests* (does 
not prove)
   that the `<20` bound on 3.10–3.13 may be leftover rather than a live ABI/API 
break.
   
   **If** that bound can be widened to overlap `pyarrow>=21` on 3.10–3.13, then
   `datasets[tsfile]` and dataset-viewer images on those Pythons become 
co-installable
   with no change required in either Hugging Face repository. That is an effect
   description, not an attribution of fault.
   
   Question for upstream: is `pyarrow>=18,<20` on 3.10–3.13 motivated by a known
   incompatibility, or can it be relaxed?
   
   Context: co-install with Hugging Face datasets (requires pyarrow>=21) fails 
on Python 3.10–3.13 with ResolutionImpossible. Related discussion: 
https://github.com/huggingface/datasets/issues/8499
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to