2026-09-11 - selfish-cli stands alone again: the local commit stamp, and the logging that logged nothing

Took the oops-libs cleanup BUILDING.md had been flagging as a regression: selfish-cli depended on oops-build and oops-log from the sibling oops-libs, so a clone of only SELFish did not build. Both dependencies are gone (D104).

The commit stamp came home in a few lines. oops_build::emit() → emit_commit() in build.rs (git short SHA into OOPS_COMMIT, a CI override winning, -dirty when the tree is modified); oops_build::line!() → version_line() in main.rs. The macro was only a macro because of the crate boundary - env!/option_env! have to expand at the consumer's call site - and inside selfish-cli that boundary is gone, so it is a plain function now. --version is byte-for-byte what it was: selfish v0.1.0 - <sha>-dirty.

The surprise: the logging logged nothing. oops-log installs a tracing subscriber, and I went in expecting to reimplement it. Grep found no tracing:: event anywhere in the workspace, and no library crate depends on tracing at all - the subscriber had been wrapping a facade nobody called. So it is deleted, not reimplemented, and tracing left with it. Cargo.lock shed 162 lines: oops-build, oops-log, tracing, tracing-core, tracing-subscriber and their transitive deps.

Docs and CI followed the code. README, BUILDING, the library feature doc, the bin/selfish header and the CI workflow all described the dependency as a live regression; all now say the property holds, and CI drops its bootstrap step because there is nothing left to bootstrap.

One circumstance worth recording: this landed on top of a concurrent session's in-flight PS4/PS5 → Orbis/Prospero rename, at the operator's direction - the two changes share a commit rather than the cleanup waiting for the working tree to clear. The rename touches selfish-container, selfish-title and data/sdk-versions.toml; the cleanup touches selfish-cli, the workspace manifest and the docs, so they do not overlap in substance.