Retire legacy code in torchtext

[zhangguanheng66]

❓ Questions and Help

As the maintainer of the torchtext library, we plan to retire torchtext.data.batch.Batch and torchtext.data.example.Example in the next release (by the end of October). Those two components will be still available in torchtext.legacy. To handle the compatibility issue, should we send a PR and fix this on GitHub first, or we could just make some changes in fbcode?

Here are the proposed changes:

# old API
from torchtext.data.batch import Batch
from torchtext.data.example import Example

# new API
from torchtext.legacy.data.batch import Batch
from torchtext.legacy.data.example import Example

Relevant issue: pytorch/text#985

[github-actions]

Hi! thanks for your contribution!, great first issue!

[Borda]

I think when you know from which version you are going to change it we can implement it right away on one of the following ways (also if you have another proposal it is welcome)

1. with known version:

   if torchtext.__version__ < X.Y:
       from torchtext.data.batch import Batch
       from torchtext.data.example import Example
   else:
       from torchtext.legacy.data.batch import Batch
       from torchtext.legacy.data.example import Example
   

2. generic ways:

   try:
       from torchtext.legacy.data.batch import Batch
       from torchtext.legacy.data.example import Example
   except ImportError:
       from torchtext.data.batch import Batch
       from torchtext.data.example import Example
   

[zhangguanheng66]

I think when you know from which version you are going to change it we can implement it right away on one of the following ways (also if you have another proposal it is welcome)

1. with known version:

   if torchtext.__version__ < X.Y:
       from torchtext.data.batch import Batch
       from torchtext.data.example import Example
   else:
       from torchtext.legacy.data.batch import Batch
       from torchtext.legacy.data.example import Example
   

2. generic ways:

   try:
       from torchtext.legacy.data.batch import Batch
       from torchtext.legacy.data.example import Example
   except ImportError:
       from torchtext.data.batch import Batch
       from torchtext.data.example import Example
   

Thanks. We will know which the version we will change. I'm just wondering if I should change this on Github first, or I can send a Diff internally to fbcode and later on sync with the master branch. I just want to know the mechanism to sync PyTorchLightening between fbcode and github.

[Borda]

you can prepare a PR and later ping us when it actually, just not to forget it...
even throw we would find it as soon as you release it that some our tests are failing :]

[zhangguanheng66]

Sounds good. For the timeline, we will drop those legacy code from the master branch early October and eventually in 0.8.0 release (10/27/2020).

[Borda]

so I think that if you decided to go with version check you can send it now...

[ananthsub]

Thanks. We will know which the version we will change. I'm just wondering if I should change this on Github first, or I can send a Diff internally to fbcode and later on sync with the master branch. I just want to know the mechanism to sync PyTorchLightening between fbcode and github.

@zhangguanheng66 There is none. PyTorch Lightning isn't linked to internal pytorch

[zhangguanheng66]

Thanks. We will know which the version we will change. I'm just wondering if I should change this on Github first, or I can send a Diff internally to fbcode and later on sync with the master branch. I just want to know the mechanism to sync PyTorchLightening between fbcode and github.

@zhangguanheng66 There is none. PyTorch Lightning isn't linked to internal pytorch

@ananthsub Thanks. So I have to manually fix the internal version by myself, right? Is there anyone importing OSS PyTorch Lightning to fbcode periodically?

[ananthsub]

That’s me + our team. We have scripts to do this and update whenever there are new releases. I’ll ping you to follow up

[stale]

This issue has been automatically marked as stale because it hasn't had any recent activity. This issue will be closed in 7 days if no further activity occurs. Thank you for your contributions, Pytorch Lightning Team!