oglyad-0.1.6.0: CHANGELOG.md
# changelog
## 0.1.6.0
- new `submodules` check (`-u`): update.rhai scripts that call
`rpm.submodules()` run under anda, and a submodule upstream added without a
`%global submodule_*`, or one the spec still has after upstream dropped it,
is a warning. needs an anda with `rpm.submodules()`. the first run after
upgrading reruns every script under anda, since its cache now keeps the
submodule changes too
- scripts run under anda are read from `anda update --report` instead of
anda's log, so a failing helper shows up as `gh(): status code 404 (line 3)`
instead of the first line of the traceback. this needs an anda with
`--report`. an older one is a `collect` warning and those scripts are
skipped
- builds with GHC 9.6 and the libraries EPEL 10 ships, so the el10 package
no longer has to strip version bounds. CI builds against both Fedora and
EPEL 10
## 0.1.5.0
- redo comments
- checks Terra EL too: `-d`/`--distro terra-el` for terrapkg/packages-el,
picked automatically when the remote's url names packages-el, or
`-r el10` on a terrapkg/packages checkout until the split is done. specs
see EL's macros (`%rhel`, `%epel`, `%el10`, `.el10`), builds are read
from `terrael10`, and `--base` compares against AlmaLinux and EPEL.
terra-el checks every `elN` branch by default, terra only checks EL when
asked
- checks terrapkg/tuatara too: `-d`/`--distro tuatara`, or picked
automatically from a checkout with a `tuatara/` directory. its one
release is `tumbleweed`, specs see openSUSE's macros instead of Fedora's,
and builds are read from `tuatara-tumbleweed` on repos.fyralabs.com
- `--fedora` and `--fedora-repo` are now `--base` and `--base-repo`, since
the overlap check uses whichever distro the monorepo builds on. the old
names still work
- cache files are named per distro and release instead of per dist tag, so
the first run after upgrading reevaluates every spec
- releases are checked in parallel instead of one after another. a cold run
over f43, f44, f45 and rawhide went from about 4 minutes to 1.5. progress
messages say which release they're about when checking more than one
- update.rhai scripts without an `updbranch` label only run under anda on
rawhide, like Terra's update workflows, which update rawhide and copy the
result to the release branches. `unparsed-rhai` only reports them on
rawhide too
- upstream lookups and anda runs happen at the same time
- an upstream that several releases ask about is looked up once, even when
the lookup fails
- rpmspec and anda share one pool of `--jobs` slots across all releases
- the missing `GITHUB_TOKEN` warning is printed once per run
- `no-autoupdate` doesn't run on Ultramarine, which updates most packages by
hand on purpose
- rpmspec refusing a spec's `BuildArch` isn't reported when the package's
anda.hcl `arches` leave out this machine's arch anyway, like a
`BuildArch: aarch64` package checked from x86_64. other rpmspec failures
are still reported, including for noarch packages pinned to one builder
## 0.1.4.0
- check Bodhi multiple times and cache response
## 0.1.3.0
- checks Ultramarine-Linux/packages too: `-d`/`--distro ultramarine` reads
`ultramarine/`, the `um43` branches and the `um43` repos with `.um43`
dist tags. without `--distro`, oglyad picks whichever distro's package
directory the checkout has
- repos are found through tetsudou's metalinks, like dnf, instead of
`repos.fyralabs.com` directly. a mirror whose repomd.xml doesn't match the
metalink's checksums is skipped
- `--repo` takes `{repo}` and `{release}` placeholders, so one url can
cover every release and subrepo. it can be a metalink or a base url
- json and markdown reports name the distro
## 0.1.2.0
- checks every maintained release by default (today f43, f44, f45 and
rawhide), as listed by Fedora's Bodhi. `-r`/`--release` takes `43`, `f43`
or `rawhide` and can be repeated. the default used to be just 44
- each release reads its own terrapkg/packages branch (`origin/f43`,
`origin/frawhide`) through `git archive`, without touching the checkout.
`--worktree` checks the checkout as it is against one release, which is
what oglyad used to do. `--remote` picks the remote
- rawhide uses `terrarawhide`, `.fcrawhide` and Fedora's rawhide repo
- findings are merged across releases and tagged with the releases they
came up in. messages that only differ by dist tag show `%{dist}`
- json schema 2: `releases` replaces `monorepo_head` and `terra_repo`, and
each finding lists its `releases`
- an ignore file entry is only reported unused when it matched nothing in
every release
- upstream answers fetched earlier in the same run are reused, so checking
four releases asks each upstream once
- `unparsed-rhai` only lists scripts that really look up a version oglyad
can't follow. kmod packages copying Version from their kmod-common spec,
scripts that only bump Release or write files, and scripts tracking a
commit from `get()` are classified as such instead. whether a script sets
the version is judged by the macros the spec's `Version:` line uses, so
`%global tag` feeding `Version: %(echo %tag | ...)` still counts
- an update.rhai with only comments in it counts as `no-autoupdate`
- `drift` and `stale` don't compare packages with a `nightly` or
`quaterdaily` label in anda.hcl against the latest release, since they
build from a branch head.
a mangled version is still flagged
- anda.hcl values can be bare (`nightly = 1`) as well as quoted
- with `-u`, update.rhai scripts oglyad can't follow (scraping with `get()`,
`gh_tags()` filters, `gh_rawfile()`, ...) run under the real anda on a
throwaway copy of the tree, and the Version left in the spec is compared.
failing scripts show up as `upstream-error`, and answers are cached like
upstream ones. anda gets `GITHUB_TOKEN`, since it needs it for GitHub
helpers
- `drift` replays the version-setting part of update.rhai over the upstream
value, so guards (`if v.starts_with("cli-v")`, `if v < "5"`), edits
(`crop`, `pop`, `truncate`, `replace`), `terminate()` and try/catch apply
before comparing. tags a script would skip no longer show up as drift
- andax's regex helpers (`find`, `find_all`, `sub`) work in that replay,
through a small engine for Rust's regex dialect that gives the same
matches and captures as the regex crate
- `drift` treats a snapshot suffix (`^20260805git...`, `.gitfc62ae7`) as part
of the spec's spelling, so `2.0.0-alpha-1-20250926` matches
`2.0.0~alpha~1~20250926.gitfc62ae7`
- `drift` warns when update.rhai would produce a version that doesn't start
with a letter or digit, like `crop(1)` turning `6.0.1` into `.0.1`
- `unparsed-rhai` warns about scripts with unbalanced brackets, which anda
can't run at all
- `opam()` lookups skip versions that aren't semver, like anda does, so old
Jane Street versions like `113.24.00` no longer win
- `--help` and `--version` print to stdout instead of stderr, so tools like
help2man can read them. bad arguments still go to stderr and exit 2
## 0.1.1.0
prepare for release on Hackage
## 0.1.0.0
first release.
- checks: `build-lag`, `drift`, `duplicate-name`, `fedora-overlap`,
`no-autoupdate`, `stale`, `unparsed-rhai`, `upstream-error`
- text, markdown and json output, with a versioned json envelope
- `--fail-on` and distinct exit codes for automation
- ignore files and `oglyad_ignore` labels in anda.hcl
- caching of spec evaluations, repodata, commit times and upstream answers