When I called Trainer.test, some hooks fit are also called

[dalek-who]

🐛 Bug

When I called Trainer.test, some hooks fit are also called, which including: on_fit_start, on_fit_end, and teardown(stage=fit)
I found a similar issue of setup(stage=fit), that bug has been fixed, but seems these hooks also have this bug
#2715 (comment)

Please reproduce using the BoringModel

https://colab.research.google.com/drive/1GA2Uo8cBaziqoBUI5K2r5x5BjjaZQRJe?usp=sharing

Expected behavior

during trainer.test, on_fit_start, on_fit_end, teardown(stage=fit) should not be called
（also could you please check if other hooks have this bug?)

Environment

• CUDA:
  • GPU:
    • Tesla T4
  • available: True
  • version: 10.1
• Packages:
  • numpy: 1.19.5
  • pyTorch_debug: False
  • pyTorch_version: 1.7.1+cu101
  • pytorch-lightning: 1.2.2
  • tqdm: 4.41.1
• System:
  • OS: Linux
  • architecture:
    • 64bit
  • processor: x86_64
  • python: 3.7.10
  • version: Proposal for help #1 SMP Thu Jul 23 08:00:38 PDT 2020

Additional context

[awaelchli]

While it is an undesired behavior, it is indeed expected given the current flow in Trainer.
We test these cases here:
https://github.com/PyTorchLightning/pytorch-lightning/blob/85c8074beefd81cce12823c5bca7e5f6e93f237d/tests/callbacks/test_callbacks.py#L109

and because Trainer.test calls Trainer.fit internally, then it makes sense that these hooks get called.

To overcome this, a larger refactor/redesign is required.
We currently have to use the stage name in the hooks to avoid repeated execution.

[dalek-who]

@awaelchli However，if we run fit and test, teardown would be called three times:

• teardown("fit") in trainer.fit
• teardown("fit") in trainer.test
• teardown("test") in trainer.test

using stage, operations in teardown("fit") will still be called twice, because we can't recognize who call teardown("fit")

[awaelchli]

Well, turns out @carmocca is already working on that #6386