Use `FutureWarning` instead of `DeprecationWarning` for deprecation warning

akihironitta commented 2 years ago

basically a copy of https://github.com/PyTorchLightning/metrics/issues/744 which was addressed in https://github.com/PyTorchLightning/metrics/pull/749

🚀 Feature

see suggestion in https://github.com/PyTorchLightning/metrics/pull/740#discussion_r782088021

Motivation

most of the deprecations in TM are meant to users not developers

Pitch

Replace DeprecationWarning with FutureWarning defined at: https://github.com/PyTorchLightning/pytorch-lightning/blob/033dba1494a177954e8ca59bc74b1635e83b9efa/pytorch_lightning/utilities/warnings.py#L44 and remove: https://github.com/PyTorchLightning/pytorch-lightning/blob/033dba1494a177954e8ca59bc74b1635e83b9efa/pytorch_lightning/utilities/warnings.py#L48-L49

Alternatives

Keep using DeprecationWarning.

Additional context

exception DeprecationWarning Base class for warnings about deprecated features when those warnings are intended for other Python developers. Ignored by the default warning filters, except in the __main__ module (PEP 565). Enabling the Python Development Mode shows this warning.
exception FutureWarning Base class for warnings about deprecated features when those warnings are intended for end users of applications that are written in Python.

If you enjoy Lightning, check out our other projects! ⚡

Metrics: Machine learning metrics for distributed, scalable PyTorch applications.
Lite: enables pure PyTorch users to scale their existing code on any kind of device while retaining full control over their own loops and optimization logic.
Flash: The fastest way to get a Lightning baseline! A collection of tasks for fast prototyping, baselining, fine-tuning, and solving problems with deep learning.
Bolts: Pretrained SOTA Deep Learning models, callbacks, and more for research and production with PyTorch Lightning and PyTorch.
Lightning Transformers: Flexible interface for high-performance research using SOTA Transformers leveraging Pytorch Lightning, Transformers, and Hydra.

cc @borda

puhuk commented 2 years ago

Can I take this

edward-io commented 2 years ago

edit: added this as a comment to the PR directly.

Can we subclass FutureWarning to create a LightningFutureWarning class? We use the base class internally to log user deprecations, e.g.

class LightningFutureWarning(FutureWarning):
  def __init__(self, ...):
     super().__init_(...)
     log_event(UserDeprecationEvent(msg))

We'll be able to add these logging events to PyTorch Lightning OSS soon, torch.monitor is going to be released in the next version of PyTorch.

akihironitta commented 2 years ago

@edward-io Sounds good to me. @carmocca Shall we?

carmocca commented 2 years ago

@edward-io Why not add that log_event call to rank_zero_deprecation instead?

carmocca commented 2 years ago

One big downside of using FutureWarning is that we would need to replace all pytest.deprecated_call uses. https://github.com/PyTorchLightning/pytorch-lightning/pull/11527 is currently failing for this reason.

ananthsub commented 2 years ago

One big downside of using FutureWarning is that we would need to replace all pytest.deprecated_call uses. https://github.com/PyTorchLightning/pytorch-lightning/pull/11527 is currently failing for this reason.

Would we need to find/replace these with pytest.warns(FutureWarning) ?

Lightning-AI / pytorch-lightning