Broken check_characters.py

[lissyx]

Regression from remote IO work:

DeepSpeech/training/deepspeech_training/util/check_characters.py

Line 22 in 5f566f4

from .io import open_remote

This fails with

Traceback (most recent call last):
  File "training/coqui_stt_training/util/check_characters.py", line 22, in <module>
    from .io import open_remote
ModuleNotFoundError: No module named '__main__.io'; '__main__' is not a package

Repro'd on mozilla/deepspeech-train:v0.9.3 Docker image

[lissyx]

@CatalinVoss Do you mind checking?

[CatalinVoss]

Ah sorry about that! I think two options

1. We update the instructions to call this script as a module like

   python -m deepspeech_training.util.check_characters
   

   That allows you to call the script from the base directory as long as you specify the right relative module path and there are __init__.pys along the way.
2. We update the import to from io import open_remote which assumes that you're calling the script from within the util directory.

Do you have a preference between these two?

[lissyx]

Ah sorry about that! I think two options

1. We update the instructions to call this script as a module like
   ```
   python -m deepspeech_training.util.check_characters
   ```
   
   
   That allows you to call the script from the base directory as long as you specify the right relative module path and there are `__init__.py`s along the way.

2. We update the import to ` from io import open_remote` which assumes that you're calling the script from within the `util` directory.

Do you have a preference between these two?

I'd prefer 1

It could be awesome if you can add GitHub Actions coverage :)

[CatalinVoss]

I updated the docs. It's awesome that you moved to Github Actions -- I love them. However, I couldn't figure out at a quick glance where you'd want to add an invocation of this script in the various workflows. I'm probably not the right person to write a test for this as I've also never used this script…

[lissyx]

where you'd want to add an invocation of this script in the various workflows.

I'd suggest adding a new job for that, next to e.g. the C++ binary tests

[lissyx]

1. `python -m deepspeech_training.util.check_characters`

So after thinking again, it also means we are breaking backward compatility.

[lissyx]

@reuben Do you have an opinion on whether it is sound we break backward compat here?

[lissyx]

We update the import to from io import open_remote which assumes that you're calling the script from within the util directory.

it's even going to break because io is a python provided module, so usually it should take precedence.

[CatalinVoss]

If you prefer, we can also just revert the remote I/O changes for this file, which would mean that you just can’t do this for remote files.

…

On Tue, Apr 6, 2021 at 05:26 lissyx ***@***.***> wrote: We update the import to from io import open_remote which assumes that you're calling the script from within the util directory. it's even going to break because io is a python provided module, so usually it should take precedence. — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#3599 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACRFKYFY647VTVZXPWLOTTTHL4YBANCNFSM42JBTP7A> .

[lissyx]

If you prefer, we can also just revert the remote I/O changes for this file, which would mean that you just can’t do this for remote files.
…
On Tue, Apr 6, 2021 at 05:26 lissyx @.***> wrote: We update the import to from io import open_remote which assumes that you're calling the script from within the util directory. it's even going to break because io is a python provided module, so usually it should take precedence. — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#3599 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AACRFKYFY647VTVZXPWLOTTTHL4YBANCNFSM42JBTP7A .

Either way, current 0.9.3 images and docs are broken, and whatever fix we pick we need to make a new dot release (which makes me unhappy), so I prefer to wwait for reuben's opinion :)

[CatalinVoss]

which makes me unhappy

Sorry 😞

[lissyx]

It was not against you, it's just that it's not the best time to have to do one. It just proves that what is not tested is broken ;)

[lissyx]

@CatalinVoss Do you mind sending the same PR agaisnt r0.9 please?

[lissyx]

Reopening for r0.9

[CatalinVoss]

@lissyx OK. Are there any other docs to update or is this one file change the only one?

[lissyx]

@lissyx OK. Are there any other docs to update or is this one file change the only one?

No it's the same one

[lissyx]

@CatalinVoss Thanks!