Integration with RLlib (v2.0)

[pseudo-rnd-thoughts]

A repeat of #249 for the 2.0 branch

[greptile-apps]

Greptile Summary

This PR backports RLlib integration improvements to the 2.0 branch, including dependency updates and enhanced error handling.

• Updated gymnasium from 0.29.1 to 1.1.1 and removed numpy version constraint for RLlib compatibility
• Changed INI file path from hardcoded "pufferlib/config/ocean/drive.ini" to dynamically computed path using __file__, improving robustness for different installation scenarios
• Enhanced C binding error handling by checking INI file existence before parsing and properly raising Python exceptions (FileNotFoundError for missing files, ValueError for parse failures)
• Modified observation space upper bound from 1 to 2 in single_observation_space, which could affect trained models

Confidence Score: 3/5

• This PR is moderately safe but requires verification of the observation space change
• The dependency updates and error handling improvements are solid, but the observation space bound change from 1 to 2 is undocumented and could break compatibility with existing trained models or cause unexpected behavior if observations don't actually exceed 1.0
• pufferlib/ocean/drive/drive.py requires attention due to the unexplained observation space change

Important Files Changed

Filename	Overview
pufferlib/ocean/drive/binding.c	Improved error handling by adding INI file existence check and better error messages with PyErr_SetString
pufferlib/ocean/drive/drive.py	Changed INI file path from hardcoded string to computed path, and expanded observation space upper bound from 1 to 2
setup.py	Updated dependency versions: removed numpy version constraint and upgraded gymnasium from 0.29.1 to 1.1.1

Sequence Diagram

sequenceDiagram
    participant User
    participant Drive as Drive Environment
    participant Binding as C Binding Layer
    participant FileSystem as File System
    
    User->>Drive: Initialize Drive(...)
    Note over Drive: Calculate INI file path:<br/>_PACKAGE_DIR/config/ocean/drive.ini
    Drive->>Binding: env_init(ini_file=_INI_FILE, ...)
    Binding->>FileSystem: fopen(ini_file, "r")
    alt File Not Found
        FileSystem-->>Binding: NULL
        Binding-->>Drive: PyErr_SetString(FileNotFoundError)
        Drive-->>User: Exception raised
    else File Exists
        FileSystem-->>Binding: FILE* handle
        Binding->>Binding: fclose(ini_check)
        Binding->>Binding: ini_parse(ini_file, handler, &conf)
        alt Parse Failure
            Binding-->>Drive: PyErr_SetString(ValueError)
            Drive-->>User: Exception raised
        else Parse Success
            Binding->>Drive: Initialize environment
            Drive->>Drive: Set observation space<br/>(low=-1, high=2)
            Drive-->>User: Environment ready
        end
    end

Loading

[greptile-apps]

_{3 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

logic: observation space upper bound increased from 1 to 2 without explanation

Suggested change

	self.single_observation_space = gymnasium.spaces.Box(low=-1, high=2, shape=(self.num_obs,), dtype=np.float32)
	self.single_observation_space = gymnasium.spaces.Box(low=-1, high=1, shape=(self.num_obs,), dtype=np.float32)

Does this change reflect actual observation values that can exceed 1, or is this related to RLlib compatibility requirements?

Prompt To Fix With AI

This is a comment left during a code review.
Path: pufferlib/ocean/drive/drive.py
Line: 87:87

Comment:
**logic:** observation space upper bound increased from 1 to 2 without explanation

```suggestion
        self.single_observation_space = gymnasium.spaces.Box(low=-1, high=1, shape=(self.num_obs,), dtype=np.float32)
```

 Does this change reflect actual observation values that can exceed 1, or is this related to RLlib compatibility requirements?

How can I resolve this? If you propose a fix, please make it concise.

[eugenevinitsky]

2.0 is frozen so we're not really taking substantive PRs to it anymore. So I'm going to close this, re-open if you disagree strongly