Fix constant error observations appearing in the logs

[tofarr]

Now checking that .gitignore exists. before trying to filter based on its contents.

• Include this change in the Release Notes. If checked, you must provide an end-user friendly description for your change below

---

I see the following constantly in the logs:

|21:13:47 - openhands:DEBUG: action_execution_server.py:166 - Running action:
    |FileReadAction(path='.gitignore', start=0, end=-1, thought='', action='read', security_risk=None)
    ...
    |21:13:47 - openhands:DEBUG: action_execution_server.py:168 - Action output:
    |**ErrorObservation**
    |File not found: /workspace/.gitignore. Your current working directory is /workspace.
    ...

The reason is that every time we call /list-files we try to filter based on the contents of .gitignore - so we get an error message every time we try to list files if there is no .gitignore.

This change checks that the .gitignore exists before trying to read it.

---

To run this PR locally, use the following command:

docker run -it --rm   -p 3000:3000   -v /var/run/docker.sock:/var/run/docker.sock   --add-host host.docker.internal:host-gateway   -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:6e252fe-nikolaik   --name openhands-app-6e252fe   docker.all-hands.dev/all-hands-ai/openhands:6e252fe

[enyst]

For some reason, one of the image builds got stuck for over 4h. I restarted them.

[enyst]

I wonder if this is a bit heavy, to do twice, for every click. (first at line 68)

Could we avoid it somehow? Or it doesn't matter?

[rbren]

Yeah I agree we shouldn't do a list_files first--we should just try to access the file, and use error handling for FileNotFoundError

If that's causing logspam let's remove the logs upstream

[enyst]

Just a thought, if there is no addition to the stream on this execution path (and it shouldn't be), could we just get/read the ErrorObs and treat it here directly?

[tofarr]

With this approach, the logs would still contain:

|**ErrorObservation**
    |File not found: /workspace/.gitignore. Your current working directory is /workspace

The only ways around this that I can see are:

1. Live with this oddity.
2. Do not filter files using .gitignore
3. Check that the .gitignore exists. (The approach I took)
4. Update the FileReadAction with an optional flag to permit missing files and not yield an error observation.
5. Update the runtime to allow access to the FileStore and use this directly.

[enyst]

Correct me if wrong, but isn't there also
6. Do not log stuff in the FileReadAction, but in its client code?

Or that sounds really bad? I don't think they are many... but I could be wrong.

[enyst]

Actually, this one isn't doing it, it's the run_action, and that is really useful...

[rbren]

How about we use runtime.read directly instead of run_action

[enyst]

Looks like it also does run_action? We could write one that doesn't, I guess, unless I'm missing something.

OK, but it's late and sorry, it seems I no longer understand this problem in the first place: weren't these all in the sandbox now, both the results of list_files and .gitignore, so why do we have runtime.list_files return all files?

[enyst]

Thanks for doing this, the log spam is annoying and everyone will be better off when it's gone.

We have a good discussion in the comments, warts and confusions and all. I think hashing out the potential issues in comments makes a PR better, and I would love to see the final form here.

So I'll put a Request changes just to make sure... that it's not merged too fast. 😅

(I noticed you merge PRs a little too fast quite often, before comments are addressed or even acknowledged. Sorry, but I think we can do this better, because sometimes that can break things and some of us really don't like to break things in main here.)

[tofarr]

I'm gonna punt on this one now as there does seem to be concerns about the additional overhead of a file existence check. Perhaps a more thorough refactor will allow a better solution