cad-module was the last red job and now passes. Board updated.
Also records that pdf.yml/fuzz reporting "skipped" is correct -- it is
gated on schedule || workflow_dispatch -- so nobody spends time
investigating it as a failure.
The caveat stays prominent: several jobs are green because their gate
is deliberately loose (the nigig-map unit-test ratchet sits at 9 real
failures, and its fmt/clippy steps are report-only). Those are listed
under Known-not-gated so a full green board is not mistaken for a
healthy codebase.
16 of 17 jobs now pass. Updates the board and adds two sections:
- Known-not-gated: the nigig-map unit-test ratchet (9 real logic
failures), the three test/bench targets that do not compile, and the
report-only fmt/clippy steps. Written down so nobody reads a green
tick as "this crate is healthy".
- Writing a ratchet step: the `bash -e` trap that made this workflow
fail at exactly its own baseline, and the `|| status=$?` fix. Cheap
to record, expensive to rediscover.
Also corrects the cad-module note: `cargo fmt -p nigig-build` is 1,559
diffs across 89 files spanning doc, project_management and
cost_estimator, not just CAD.
Adds the fourth defect the first real runs exposed -- the mapping form
of `on:` not being scheduled on this instance -- and a table of the
latest result for all 17 jobs, so "is CI green" has an answer that is
not someone's memory.
14 pass. The two failures, cad-module formatting and nigig-map, are
pre-existing source problems rather than CI plumbing.
Not in the repo root -- this lives next to the workflows it describes.
Covers registering a forgejo-runner against gitdab (Gitea 1.22),
the ubuntu-latest label every job depends on, the docker-vs-host
tradeoff, and how to read job logs given that this instance's REST
API 404s on the logs endpoint.
The important section is action resolution: Forgejo resolves `uses:`
against data.forgejo.org with no github.com fallback, and a missing
action fails the job in "Set up job" before any step runs -- which
reads like an infrastructure blip rather than a config error. Records
which actions currently resolve and which do not.
Also records the four defects the first real runs exposed, three now
fixed, so the next person understands why these workflows look the way
they do.
