[REQUEST] environmental variable to disable rich traceback

[matthewfranglen]

How would you improve Rich?

I am using huggingface transformers to train a model in a project with a bunch of dependencies. One of those dependencies is installing the rich traceback.

When there is an error then over two thousand lines of output are printed because the model is serialized on almost every frame (>100 lines of output) along with large tensors, dataframes etc. This makes the stack trace unusable.

I cannot control the fact that rich.traceback.install has been called somewhere. It would be nice if I could set an environmental variable that would prevent the installation from taking effect.

What problem does it solve for you?

I would be able to read and understand the stack trace that is produced.

[github-actions]

We found the following entry in the FAQ which you may find helpful:

• How do I log a renderable?

Feel free to close this issue if you found an answer in the FAQ. Otherwise, please give us a little time to review.

This is an automated reply, generated by FAQtory

[TomJGooding]

Sorry if this is stating the obvious, but wouldn't it make sense if this was configurable in whatever dependency has installed Rich tracebacks?

[matthewfranglen]

How can I reliably find that?

[matthewfranglen]

This seems to be an issue that others have encountered and have proposed truncating locals for: #1810

Broadly I like rich and this is an edge case. The length of the output has made debugging an issue extremely difficult, which is the opposite of the intention of this package. This is my suggestion to fix that.

[TomJGooding]

I cannot control the fact that rich.traceback.install has been called somewhere.

But why should the onus be on Rich rather than whatever dependency has installed the traceback handler?

How can I reliably find that?

I have no idea what dependencies you have installed. I'm just suggesting that perhaps your first port of call should be whatever dependency that's using Rich, rather than going straight to Rich with the issue.

[matthewfranglen]

Rich is the dependency that is producing the problematic output.

[willmcgugan]

I'm reluctant to impose non-standard environment variables on apps. It really isn't the place of a library to do that.

I would petition the maintainer of the library that enabled the tracebacks to add that env var.

Or, you could disable the exception hook. This will restore the Python default:

sys.excepthook = sys.__excepthook__

Alternatively, call traceback.install(...) again with new settings.

[github-actions]

I hope we solved your problem.

If you like using Rich, you might also enjoy Textual

[max-sixty]

I'm a huge fan of rich and everything you've built @willmcgugan

but I would say — having one app mangle all python tracebacks because they used rich badly is frustrating. it means the majority of the time that I encounter rich, it's trying to search within a dozen monitors of arrays because of the rich traceback

I realize it's on the maintainer to use rich correctly, but that response feels a bit like we're running Windows ME, where any process can disrupt everything else. a simple RICH_TRACEBACK=0 would fix those problems. it doesn't prevent libraries from adding their own; it would a safety addition

even a RICH_TRACEBACK_RAISE would be very helpful for finding which is the offending library

fwiw sys.excepthook = sys.__excepthook__ doesn't work for me

thanks again!

[matthewfranglen]

It's worth pointing out that DISABLE_RICH_TRACEBACK appears to be unused across github except for one python AI framework which has it as a setting but doesn't reference it: https://github.com/search?q=DISABLE_RICH_TRACEBACK&type=code

The impact of adding this environmental variable would seem to be minimal.

[paulvinell]

Please implement something like this. Sifting through error logs from a dependency that doesn't allow disabling rich is a pain I wouldn't want to inflict on my worst enemy.