why setting docker image-build's nocache=True?

[zzcwest]

https://github.com/DeepSoftwareAnalytics/swe-factory/blob/main/app/agents/test_analysis_agent/test_analysis_agent.py#L317

As shown in the code, nocache=True — why is it set to True, when setting it to False would allow Docker build to utilize the cached image layers?

[gnohgnailoug]

Thanks, this is a great point. You're right that nocache=False would make builds faster. In early experiments, we use nocache=True to avoid old cached layers masking real setup issues, so the environment is rebuilt cleanly from scratch.

But, considering our current large-scale SWE data collection workflow, I agree that setting no_cache=False by default could be a more practical choice for efficiency. A good next step would be to add a no_cache parameter (default=False), so users can choose between faster builds and stricter validation depending on the scenario. We can also document this clearly in the README.

If you're interested, a PR would be very welcome :)

[zzcwest]

If nocache=True, intermediate images cannot be deleted for cache reuse, which may lead to uncontrollable disk consumption. A feasible solution is to delete all images after the task is completed.

Hence, I think this feature requires a comprehensive discussion before a PR.

[gnohgnailoug]

If nocache=True, intermediate images cannot be deleted for cache reuse, which may lead to uncontrollable disk consumption. A feasible solution is to delete all images after the task is completed.

Hence, I think this feature requires a comprehensive discussion before a PR.

Thanks, this is a fair point. In practice, I usually just handle this with a small script: first run the target tasks, then clean up the Docker images from that run.

So I see this more as an operational issue than a blocker for supporting no_cache. We can document this recommended workflow in the README.

[zzcwest]

Thank you!