Fixes for latest versions

[jalammar]

Hi @paul-gauthier, I'm trying to get this to work with the current versions and I faced the same issues in #5 . These are some updates to this repo, but some would need to happen on the Aider side.

This installation pins swebench==1.1.5 because report.py requires swebench.metrics.report.get_model_report which was removed from later versions (I'm not clear what they replaced it with).

There is an extra change I had to do to run SWE-Bench Lite that I haven't added here because there might already be a better way to do it. That change is to report.py, and it's about importing LITE_DATASET_FNAME and substituting it in most places that mention FULL_DATASET_FNAME. But there might be a better way to do that or existing functionality. I haven't read the code in detail yet.

On the aider side, what I've had to change was:
1- in repomap.py: in get_scm_fname Change except KeyError to except (KeyError, TypeError)
2- Also in repomap.py:
change

 if not query_scm.exists(): 
            return

to

if query_scm is None or not query_scm.exists(): 
            return

Otherwise, it may fail when facing file extensions not in tree-sitter.

Also need to point out this now needs to be run in python 3.11. SWE-bench-docker's run_evaluations.py needs asyncio.TaskGroup.

So for my setup, I did this before installing the requirements:

conda create -n aider-swe python=3.11

[brandonkachen]

@jalammar this is such good work 🙏 I was so confused when things weren't working but your changes helped me figure it out!

Just want to help bump this again to @paul-gauthier