llamastash / llamastash / 26514649134

87%

Build Type

push

github

Committed by deepu105

Commit Message

fix(discovery): honor full HF env precedence and Ollama Linux system path

Discovery scanned only ~/.cache/huggingface/hub, so `llamastash list` missed
models that `llamastash pull` had written to a relocated cache via HF_HOME.
Move HF/Ollama/LM Studio path resolution into `util::model_caches` so writes
(init::download) and scans (discovery::known_caches) share one source of
truth.

- HF precedence now mirrors hf-hub 1.0.0-rc.1: HF_HUB_CACHE →
  HUGGINGFACE_HUB_CACHE → \$HF_HOME/hub → \$XDG_CACHE_HOME/huggingface/hub →
  \$HOME/.cache/huggingface/hub.
- Ollama scan adds /usr/share/ollama/.ollama/models on Linux (official
  systemd installer location).
- UAT isolation also sets HF_HUB_CACHE so a maintainer's exported value
  can't override the sandbox HF_HOME.
- Drop dead macOS ~/Library/Caches/huggingface/hub scan; hf-hub never
  writes there on any platform.

Coverage Stats

293 of 303 new or added lines in 4 files covered. (96.7%)

142 existing lines in 5 files now uncovered.

38976 of 45024 relevant lines covered (86.57%)

144779.87 hits per line