Update sandogasa to v0.20.0 and bump its emojis dependency to 0.9

emojis changelog

0.9.0

June 20th, 2026

• Improve Debug implementation for Emoji struct

• Remove unnecessary PartialEq implementation

• Fix Unicode version metadata and expose emoji versions. Fixes the Unicode version by correctly translating form the emoji version. Adds EmojiVersion and Emoji::emoji_version() for callers that need the UTS #51 emoji version separately from the Unicode Standard version.

Contributed by Mike Fiedler

sandogasa changelog

v0.20.0

A minor bump because two things break: dbranch's --stage build no longer makes the source package, and koji-lag's and dbranch's library types gained fields. On-disk data and reports are unaffected.

• dbranch: making the source package is its own stage. "--stage build" is now only the pbuilder-dist scratch build, so pass "--stage source,build" (or "all", unchanged) where "build" alone used to do both. build and upload now check what they are about to consume: a source package older than the current commit offers a rebuild, a missing one is a sequencing mistake and fails a non-interactive run naming the stage to run. Breaking API: Stages gains "source", plan::debuild_argv takes an argument
• dbranch: a PPA upload no longer risks rejection for a missing tarball
• koji-lag: new backfill subcommand sweeping a long window a day at a time, writing daily/YYYY/MM/DD and collating complete weeks and months, with --pause-at for a clean break point and --if-exists for what to do about work already on disk. New reports subcommand renders a swept tree without fetching, and "report --out DIR" writes report.txt and report.json in one pass
• koji-lag: a backfill walks each creation once. It seeks to the window rather than paging back from today, and consecutive weeks carry their overlap between them - three days that cost 108 pages now cost 47
• koji-lag: requests are paced by how long the hub takes to answer. --duty-cycle names the share of one connection to aim for, so a loaded hub is asked less often and one that recovers is asked more
• koji-lag: the SRPM rebuild is collected and reported as its own stage, keyed by the arch of the host that ran it, and per-arch statistics are broken out by the shape of the build - multi-arch, single-arch, and noarch keyed by the machine that took it. A report now says how many builds its bottleneck count is out of. Breaking API: Dataset gains host_arches, TaskRecord gains method, FetchOpts gains duty_percent
• koji-lag: progress output says where a sweep has reached and roughly what is left, rather than a bare page number
• ebranch: check-wip reports what a build is waiting behind, showing an earlier version's update still in testing and how long it has served, and its summary names the work left on target branches once Rawhide only waits. It no longer misses builds newer than the one it recorded
• koji: calls are bounded, with a separate longer bound for operations meant to block. regen-repo is no longer killed after 30 seconds

Full details: https://github.com/slopfest/sandogasa/blob/v0.20.0/CHANGELOG.md#v0200

v0.19.4

A bugfix release, cut promptly: check-wip in 0.19.2 and 0.19.3 could stop noticing new builds, which is the one thing it exists to do. Nothing has been yanked - the fault is a CLI under-reporting state, the library crates sharing the version are unaffected, and the stale records the older versions wrote are corrected on the next run.

• ebranch check-wip missed builds that were already in Koji. To avoid asking about finished work it skipped any branch whose repositories already carry the version being pushed - but with no COPR behind an effort it used the newest build it had itself recorded, so once that matched the repositories the branch looked finished, was never asked about again, and the record could never change. Only a version staged in a COPR closes a branch now, and a side tag's contents apply to packages already tracked rather than only to new ones
• ebranch check-wip asks Koji in bulk where that is cheaper. A branch was one koji list-tagged per package per tag, so fifteen packages meant thirty subprocesses; a tag can be listed whole and read for every package at once. Measured crossover is about seven packages, and both forms are kept
• ebranch check-wip says where a newer build is waiting, from one template so the cases stay comparable: "newer build in a COPR, not landed in rawhide" and "newer build in a side tag, not landed in rawhide". A build in a release's own tags still reads "not yet in the repos", where a compose really is the only wait, and the branch line names the side tag when no update carries the build yet
• Dev builds stop writing tens of gigabytes of debuginfo. [profile.dev] debug = "line-tables-only" halves a cold build of the workspace and its tests, from 6.3GB to 3.3GB, and backtraces keep file and line
• New make sweep target removing the trees the release gates leave behind - rustdoc for both sides of a semver comparison, and an instrumented copy of the workspace for coverage - which came to 25GB at 0.19.3. It leaves target/debug alone

Full details: https://github.com/slopfest/sandogasa/blob/v0.19.4/CHANGELOG.md#v0194

v0.19.3

Everything here came out of using check-wip through F45's mass branching and the Koji outage that came with it.

• koji calls are bounded - 30 seconds by default, SANDOGASA_KOJI_TIMEOUT to override, 0 to wait forever. The koji CLI waits indefinitely, so a hub outage hung every caller with no output at all. A hub that does not answer is asked once: the first timeout latches per profile, and hub_unresponsive lets a caller with queries left report from what it already has. check-wip against a down hub now finishes in under a minute with one warning instead of hanging
• ebranch check-wip seeds its ledger from a registered side tag, the way it already did from a COPR, recording each package as built for that branch rather than staged. An effort built straight into a side tag no longer needs --add per package
• ebranch check-wip --prune names what to forget: packages, side-tags, or both. A bare --prune still means packages. Neither list is pruned on a question that went unanswered - a timeout, an unreachable hub or a failed COPR leaves both alone and says which check established nothing
• ebranch check-wip --forget NAME,... drops packages and records the refusal, so neither a COPR nor a side tag takes them up again. --add reverses it
• ebranch check-wip no longer deletes packages it has no evidence about: pruning acts only on packages a COPR the ledger follows once staged, so a package added by hand or found in a side tag is left alone
• ebranch check-wip follows the distro through mass branching. Bodhi's F45 had branch rawhide and now has f45; EPEL-10.3's is epel10; dist-git has an epel10 branch but never an epel10.3. Releases are matched by branch or name, two targets naming one release collapse to one, and a target a side tag implies comes back when that branch becomes a release of its own
• ebranch check-wip reports the age of what it shows. A line dated itself by the oldest fact known about the branch, including facts it was not printing, so a current branch read as a day stale

Full details: https://github.com/slopfest/sandogasa/blob/v0.19.3/CHANGELOG.md#v0193

v0.19.2

• New ebranch check-wip subcommand - tracks a multi-package effort on its way into the distro from a TOML ledger: where each package sits across COPR, dist-git, its Review Request, Koji and Bodhi, per target release. Flags: --copr, --target, --side-tag, --add, --set, --package, --prune, --rescan-reviews, --offline, --json
• check-wip reports retirement properly - the dead.package reason, the retirement date when it can be established from dist-git, and whether anyone has a review open, rather than treating a dead package as work waiting to be done
• check-wip distinguishes a branch, a target and a release - Bodhi's F45 has branch rawhide and EPEL-10.3 has branch epel10, so releases are matched by branch or name, duplicate targets collapse into one, and a package shipped for epel10.3 is no longer told it needs a branch of that name
• ebranch check-update judges FTBFS and FailsToInstall bugs, and handles Review Request bugs for new packages, so karma feedback for those no longer has to be entered by hand
• ebranch submit proposes the bug list for an update, discovered from the changelog and from open bugs on the components being shipped
• Config defaults can no longer authorize a write - apply, claim, give-karma, prune, submit and yes are rejected as [defaults] keys, and a --no-<flag> can override a default that switched something on
• sandogasa-bugzilla asks for the fields it deserializes, so an approved fedora-review+ flag is actually seen
• sandogasa-bugclass no longer treats a package name as a version prefix, which had let rust-ctor answer for rust-ctor-proc-macro
• poi-tracker searches Bugzilla by release version rather than by branch name, which found nothing on numbered branches
• Release tooling verifies a publish landed (make check-published) and documents how to compute --publish-interval
• Dependency: emojis 0.8 to 0.9

Full details: https://github.com/slopfest/sandogasa/blob/v0.19.2/CHANGELOG.md#v0192