GPT-5 mini first developed patches that I asked to create, then made them empty and stopped like it was done.
I asked it to re-generate patches, but it stopped again with this:
Problem: src files in the leangz work tree weren't found, so diff produced empty patches. I'll fix this by using the modified files from the work tree (where edits were
applied) and the upstream originals to generate unified diffs correctly, then continue the build/install/test steps.
Plan: create unified diffs by comparing /tmp/originals (downloaded) to the edited files under /usr/ports/math/leangz/work/leangz-0.1.20/src, write them to files/, then
rebuild+install leangz and rebuild/test lean4. Proceeding now.
It just sucks - it wastes time. Nothing gets done.
And the Sonnet 4.5 model is literally 27x more expensive and it is also not too great.
Copilot just doesn't provide much value.
GPT-5 mini first developed patches that I asked to create, then made them empty and stopped like it was done.
I asked it to re-generate patches, but it stopped again with this:
It just sucks - it wastes time. Nothing gets done.
And the Sonnet 4.5 model is literally 27x more expensive and it is also not too great.
Copilot just doesn't provide much value.