Changelog¶
Unreleased¶
Added¶
- submit:
--withsearch forms (PKG,^PKG,PKG:) now also acceptpkg.git=tag, finding the task that contains that exact subtask instead of matching by package name only. - gen environment:
-dnow traces the whole decision path — tool availability checks, SSH key lookups,~/.gnupgpermission check, keystore classification (entries found inprivate-keys-v1.d), every gpg/git command with its exit status, and the chosen key-generation path.
Fixed¶
- gen environment: empty
private-keys-v1.d/public-keys.ddirectories left behind by an interrupted gpg2 migration no longer flip a classic (secring.gpg) keyring to "modern" — only a real*.keyfile inprivate-keys-v1.dcounts, so key checks go through plaingpginstead of the hanging gpg2/keyboxd (#86). The stuck-lock hint now also suggests removing stale GnuPG lock files.
Added¶
- submit: autodetect kernel-image repos (
.gear/ruleswithspecsubst: kflavour, specName:containing@kflavour@) and create the tag with-s kflavour=<flavour>, fixinggear-create-tag: specsubst variable "kflavour" is not defined. The flavour comes from-k,[specsubst] kflavourin.gear/version-up, or the branch's<flavour>/<dist>prefix (which also defaults-Bto<dist>).
0.40.0 - 2026-06-09¶
Added¶
- sandbox:
[sandbox] specbrin.gear/version-up. Withspecbr = falsethe hybrid runner skips the package's BuildRequires — the chroot is initialised bare (hsh --initroot-only) instead of building the src.rpm; the configured packages (git,[sandbox.chroot] packages,[sandbox] packages) are still installed. Defaults totrue; supersedes a custom[sandbox.chroot] prepare(warns when ignored). - submit: autodetect kernel-module template repos and create per-flavour specsubst tags (
-kexposed only in such repos, comma-separated and TAB-completing flavours; default flavours come from the module's existing builds).karchis derived from the arches the kernel flavour is actually built for, so a module is not built where its kernel is absent.--dry-runprints each tag's specsubst headers. - submit: universal multi-tag
--replace— for kernel-module flavours and batch specsubst submits, each new tag replaces the subtask carrying the same value (adds it if missing), instead of only the first tag. - task gitclone: new subcommand that clones the git repository of every subtask of a task into the current directory, one per package, at the exact commit built in the task. Fetches from the task's own gears over
git://(resolving archived DONE tasks automatically); skipsdeleteandsrpmsubtasks. - task mkrepo: new (experimental) command — generates a local merged apt repository overlaying a task's RPMs on the base branch mirror, for testing the task against a complete snapshot. The packages to add and remove come from the task's authoritative
plan/bin.list.diff(so additions, version replacements, and deletions — including delete-only tasks — are reflected). Thebase/index is built by splicing pkglists (reuse the mirror'spkglist.classic, drop the removed headers, splice in the task'spkglist.task) instead of runninggenbasedir, so indexing reads only the task's packages and is near-instant regardless of branch size. Writes a single pkglist index — plainpkglist.classicby default, orpkglist.classic.xzwith--compressed(not both).--forceremoves a previously generated repo before rebuilding;--cleanremoves it and exits. A ready apt config (apt.conf/sources.list/priorities) is written to<workdir>/<task_id>/aptconf/, and the generatedreleaseidentifies the snapshot (Suite/Label= branch,Description=<branch> (<branch source task> plus task <id>)). - up:
--reset-to-gearflag — hard-resets the package to its published gears/srpms state before updating, discarding local divergence, uncommitted changes, and untracked files (ignored artifacts kept). Resets the branch planned for the update (current if a known ALT branch, otherwise sisyphus) regardless of the active branch. Closes #81.
Fixed¶
up / spec parsing: the RPM built-in macro
%nilnow expands to the empty string during version computation. A spec composing its version as%base%sublevel%extrawith%define extra_version %nil(e.g. kernel-image) previously yielded a bogus current version like6.18.35%nil, sozoryn upfailed withCannot find any tags for current version '6.18.35%nil'.build (
--section): a short-circuit section build of aspecsubst:template package (kernel-image, kernel-modules) no longer feeds rpmbuild a raw template —rpmbuildfailed withInvalid symbol '@' (0x40) in: Name: kernel-image-@kflavour@. zoryn now expands@var@placeholders before syncing the spec into the chroot, mirroring gear (values fromgit config gear.specsubst.<var>), and errors with a clear hint when a value is unset.- up: the upstream remote is now set up from any forge, not just GitHub/GitLab/codeberg/sourcehut. The cascade (
.gear/upstream/remotes→ specVcs:→ specUrl:) is an explicit upstream declaration, so the first candidate thatgit ls-remotecan reach (probed with atimeoutandGIT_TERMINAL_PROMPT=0) is used — no host allowlist. Self-hosted Gitea/Forgejo/cgit instances (e.g. 0ad'sgitea.wildfiregames.com) work now, and an upstream hosted onaltlinux.space(a Forgejo) is accepted like any other reachable forge — unblocking packages like kitsune-adw. - sandbox: hybrid mode now also bind-mounts the chroot's
/var(read-only) into bwrap, alongside/usr,/lib,/lib64,/etc,/bin,/sbin, with a writable/var/tmptmpfs on top so hooks can use it as scratch. - sandbox: the hybrid chroot is now prepared once per run instead of for every hook. Re-initialising per hook was pure overhead — most visible with
[sandbox] specbr = false, where the bare init (hsh --initroot-only) rebuilt the chroot from scratch before each script. The bare init also streams its output live (its lines aren't recognised as build stages, so the long basesystem step previously looked like a hang). - task mkrepo: a package rebuilt with an identical
epoch:version-releaseis now taken from the task, not the stale base mirror.bin.list.diffomits such rebuilds, so the task's own package set is now authoritative: every task build replaces the same-named base copy in both the symlink farm and the spliced index. - gen environment: GPG key generation no longer hangs silently. zoryn now reads the passphrase itself and feeds it to gpg via
--batch --passphrase-fd(so gpg never blocks on its own/dev/ttyprompt after the migration messages of a legacy~/.gnupg), pre-checks~/.gnupg/$GNUPGHOMEownership and permissions and aborts with a fix command when unsafe, and prints recovery hints on failure. It also detects a classic keyring (secret keys insecring.gpg, no keybox) from disk and drives it withgpginstead ofgpg2—gpg2would migrate it and can hang on keyboxd's lock. Both listing and generation then go throughgpg: a genuine GnuPG 1.4gpggenerates via a batch control file (so the new key lands in the samesecring.gpgthe user reads), while a 2.xgpg(the commongpg→gpg2 symlink) uses the loopback--quick-generate-keypath so it never prompts and re-hangs. The key-listing calls are guarded withtimeout -k, so a wedged agent turns into an actionable "rungpgconf --kill all" error instead of an infinite wait. Closes #86. - commit / changelog parsing: a changelog version containing letters (e.g. AFLplusplus
5.00c) is no longer dropped.zoryn commitproduced a malformed-subject because the header parser only accepted all-numeric versions; it now accepts RPM-legal version characters. Closes #85. - up (scheme detection): a merge of an upstream tag (monorepo
Merge tag 'hatchling-v1.29.0') is no longer misdetected astarball-watch. The detector now defers to the tag-based stages — recognizing prefixedv-version tags like<prefix>-v<version>— and names the actual merged tag in its reason. - up (scheme detection): a merge of an upstream release tag with a non-semver name (e.g. 0ad's
Merge tag 'a27.1') is no longer misdetected astarball-watch. When the merged commit is absent from every fetched upstream branch, the detector now also matches it by SHA against the upstream remote's own tags (git ls-remote --tags), so any upstream tag-naming convention counts as a realgit-merge. - builder shell: now mounts the same paths the build uses (
/proc,/dev/pts,/dev/kvm, …) and validates/auto-fixes hasher-privallowed_mountpointsfirst, so packages whose BuildRequires need/procwork in the interactive shell. - submit:
--replacenow implies--allow-overwrite-tagfor the local tag rewrite. Replacing an in-progress subtask aftergit commit --amendorgit rebaseno longer aborts with "tag points to commit not an ancestor of HEAD" — the whole point of--replaceis to swap in a new commit at the same tag name. A dim notice is printed when the implication is auto-applied.--refreshand the gears-immutability check are unaffected. - submit: brand-new packages (not yet in Sisyphus) no longer get stuck on the "tag points to a commit not an ancestor of HEAD" safety check when a stale local tag remains from a previous attempt. When the gears repo is provably missing (ls-remote fails) and the package is unknown to RDB, the local tag is treated as stale and overwritten automatically with a warning. Existing packages keep the strict ancestor check.
- task test-rebuild: now aborts with an error instead of silently reporting success when the task has no built repository to overlay. Previously a task whose RPMs failed to download, or whose repo was not ready, produced a green run that rebuilt dependents against the clean branch — testing nothing. The download now reports a per-architecture result; an empty listing or any failed RPM aborts the run, as does a run with no available builder architectures. The gate is the task's repository presence, not its state: an early guard only rejects tasks still building (
NEW,AWAITING,PENDING,BUILDING,COMMITTING,FAILING), whoseRPMS.taskwould be empty or partial. Terminal states proceed, includingFAILED— a failed task often still has a usable (if partial) repository, and test-rebuilding it is a normal workflow;DONEwarns the overlay is redundant andFAILEDwarns the RPM set may be partial.--local-rpmsand--without-taskare unaffected. - submit:
--replacewith multi-branch-B sisyphus,p11now finds and replaces the matching task in each branch (one auto-search per branch). Previously the comma-joined repo string was passed as a single filter totask ls, so nothing matched and the command aborted with "no open tasks found". Explicit--replace=TASK_IDis still single-target and errors out when combined with multiple-Brepos.
0.39.0 - 2026-05-26¶
Changed¶
- task rebuild: default
--dptypeis nowbinary(wassource). Binary dependencies match the practical "what breaks if this ABI changes" question better for the rebuild workflow. - up (merge-hooks):
.gear/merge-up.d/*hooks now run only when this invocation actually produced a merge commit, or when the user explicitly asked for the stage via--stage merge-hooks/--from merge-hooks. Running the full pipeline on an already-up-to-date tree no longer re-runs hooks that already ran in the previouszoryn up(and that had no merge in this run to react to). The persistent pipeline state grew amerge_performedflag (set bystage_mergeon success, preserved across--continue); explicit--stage merge-hooksinvocations also opt in via a runtime-onlyexplicit_merge_hooksflag that is not persisted, so a re-run never silently inherits yesterday's opt-in. - up: new Step 0.3 syncs the local package repo with the public gears/srpms mirror before detection — fast-forwards when behind, auto-merges on divergence, aborts on conflicts or a refused fast-forward. Only runs when the local branch is a known ALT branch (sisyphus, p11, p10, c10f2, …); topic branches like
masterorwip-*skip silently. Also skipped for packages unknown to RDB, dirty trees, detached HEAD, or fetch failure; fetch is bounded byGIT_HTTP_LOW_SPEED_LIMIT=1000/_TIME=10so a slow mirror can't hang the run. Closes #80.
Added¶
- task copy:
--into <task_id>option to append subtasks to an existing target task instead of creating a new one. Target repo is derived from that task (-Bignored); target-task state is validated (DONE/SWEPT/EPERM rejected, in-progress states warned); dependencies are left untouched. - task copy:
--subtask <SUBTASK>(repeatable, comma-separated) to copy only selected subtasks. Accepts the same syntax aszoryn task delete(numeric ID, package name,pkg.git=tag) with TAB completion from the source task. - task test-rebuild:
--without-taskflag — skips downloading task RPMs and rebuilds dependent packages against the current branch state (baseline). Hasher RPMs are cleared before each individual build to prevent cross-contamination between dependent packages in the same run. Logs get thebaseline.prefix. Mutually exclusive with--verify-failuresand--local-rpms. - task add:
--replaceflag to replace an existing subtask in a single command. Target subtask is derived from the action's package name (matches bothpkg_nameandgit_repo); ambiguous matches error out asking the user to delete the unwanted subtask manually. Cannot be combined with multi-packagedel/rebuild. Closes #76. - task add:
<task_id>^<subtask_id>positional shorthand for--before <subtask_id>— accepts task references in the form printed bytask ls/task show. - task add:
<task_id>/<subtask_id>positional shorthand for--replacetargeting an exact subtask (bypasses auto-find by package name).
Fixed¶
- submit / task batch: tag-overwrite safety rework (Closes #71)
submitrecreates a stale local tag automatically when the existing tag is an ancestor of HEAD (fast-forward case) and not yet published in gears. Refuses with an actionable message otherwise. The sisyphus-onlyAlt_branches.is_rootforce-tag shortcut is gone — uniform algorithm on every branch.--allow-overwrite-tagflag (on bothsubmitandtask batch) bypasses the gears/ancestor checks when you know the stale tag is unrelated to HEAD.--refreshcontinues to overwrite the local tag unconditionally regardless of this flag — it is only relevant on non-refresh submits.- Output distinguishes
Created tag: X(fresh),Reusing existing tag: X (already on HEAD), andRecreated tag: X (was at <sha>)(force-overwrite). Previously every successful tag step printedCreated tageven when no tag was created. --dry-runnow refuses (exit 1) when the gear specfile is missing or lacksVersion/Release(previously synthesized?-?placeholder tags).- Same safety algorithm applies to all batch tag paths (simple, first, remaining) and is consistent with non-batch submit.
- Side-effect: in batch mode with
includes_original_first=true, the first tag is now built viaapply_batch_formatlike the rest of the batch. If you relied on the implicit<project>-prefix, add{name}totag-namein[batch]. - Simple-submit defers to
gear-create-tagdefault naming when no local tag exists. Previously the fix forced<project>-<version>-<release>style, breaking packages that historically used bare<version>-<release>tags (e.g. libva-intel-media-driver). Safety algorithm still runs when an existing tag is found under either naming style. - task copy: dry-run/real run on a NEW source task whose subtasks have not been fetched yet (no
srpm/package/pkgnamein the API payload) no longer aborts with "No valid subtasks found".resolve_pkg_namenow falls back to the git repo basename fromdirso build subtasks are recognised before any artifact has been produced. - builder:
clear_hasher_rpmsnow also wipes{hasher_dir}/repo/SRPMS.hasher/.sync_rpms_to_hasherdoes the same only whendelete = true(its default, used byzoryn task test-rebuild);zoryn task rebuildcalls it withdelete:falseand keeps the additive-sync semantics. Output ofhshpreviously accumulated indefinitely across a test-rebuild batch. - test-rebuild:
fork_single_taskchild now clears the inherited apt-tmpdir cache, installsSIGINT/SIGTERMhandlers that runcleanup_allbefore_exit, and explicitly cleans up on normal exit — closing the per-task/tmp/.private/<user>/zoryn.XXXXXXXX/leak both for completed and interrupted runs. - up (merge):
zoryn up(and--continue) on a branch where the target upstream tag is already merged no longer reports a bogus "Merge failed - conflicts detected". The pipeline now detects that the tag is reachable fromHEADand skips the merge step, letting the remaining stages run. - task manage: when running a task that contains rebuild/delete subtasks (or targets a stable repo), gyle's "please specify a reason" rejection no longer ends the action with an "Error: …" status. The TUI now detects gyle's reason-required errors and re-opens the build-reason input prompt prefilled with the existing task message, preserving the original run mode (default /
--commit/--test-only) so the retry uses the same flags. Also fixes the previously silent path where pressingr(plain run) on a rebuild task would just fail. - gear rules parsing: any
tar.<comp>:directive (where<comp>is a non-empty lowercase-alphanumeric suffix) is now recognised by the.gear/rulesparser, replacing the previous hand-maintained allowlist oftar.gz:/tar.bz2:/tar.xz:. Concretely,tar.zst:was missed: in a package whose.gear/rulesstarted with e.g.tar.zst: thunderbirdfollowed by a secondarytar: l10n name=l10n,find_tar_directoryskipped the first directive andzoryn uptried to update thel10nsubdirectory instead of the main source tree. The four duplicated prefix lists (intar_line_uses_version_template,parse_tar_directory,main_tar_uses_version_template, andrewrite_rules_for_upstream_git) now share a single helper, so future compression suffixes added by gear are picked up automatically rather than producing a silent miss. - up (merge-hooks):
zoryn upon an already-updated project (or--stage merge-hooksafter the fact) no longer feeds the bare spec version into$TAG, which broke hooks doinggit rev-parse "$TAG^{}"withambiguous argument '1.5.0^{}'for upstreams that tag as e.g.v1.5.0(kryoptic). The stage now resolves the spec version to an actual tag viaGau_git.find_all_matching_tags(honoring.gear/rulestag prefixes andpkg_name-patterns), and skips merge-hooks cleanly when no tag matches instead of running with a non-ref string. Closes #78. - task batch: main package no longer fails with
gear-create-tag failed(and a follow-upsrc refspec ... does not match anyon push) when re-running for a second target repo: the per-package tag is now resolved up front and reuses the bare<version>-<release>tag on HEAD when the batch-formatted tag is not present, so push andtask addreference the tag that actually exists. Fresh first runs also create the main package's tag with the batch-formatted name (gear-create-tag -n …) for consistency.
0.38.0 - 2026-05-05¶
Added¶
- task refresh:
--runqueues the task for build right after a successful refresh; pair with--test-only/--commit(mutually exclusive) and-m MESSAGEto forward those flags totask run. - changelog:
web_regexparser for upstream HTML release notes (e.g. Wireshark). Configureparser = "web_regex",url,web-regex-pattern(PCRE with named groups(?<id>...)and(?<desc>...)), and optionalweb-regex-stop-atsubstring marker in.gear/version-up. URL-based parsers (web_regex,markdown,osv-json,html-table) also gain{old_version}/{new_version}placeholder substitution. CVE descriptions are RPM-macro-escaped (%→%%) before being written into the spec%changelog, so upstream prose containing literal%cannot trigger macro expansion.
Removed¶
- config: dropped legacy INI config support — the
zoryn_config_fixmigration utility, the INI parser, and the on-load auto-detection path are gone. INI-style~/.zorynand.gear/version-upnow print a TOML parse warning on stderr and fall back to defaults; INI-style batch configs fail with a parse error. Convert them to TOML manually.
Fixed¶
- task delete/approve/disapprove/log:
pkg.git=tagnow picks the subtask whose tag matches (not the first sharing the git repo); a bare ambiguouspkgerrors out and lists the candidates. Closes #74. - bash completion: TAB after
pkg.git=tagprefix no longer stalls or duplicates thepkg.git=prefix in the reply. - config: invalid TOML in any user-edited config file now fails with
zoryn: <path>:<line>:<col>: invalid TOML: <msg>and exit 1, instead of silently falling back to defaults with only a stderr warning. Previously a typo in.gear/version-upwould silently swallow the entire[changelog]section sozoryn upproduced an entry without CVE references;~/.zoryn,builders.d/*.conf, and the[commands]section had the same silent-fallback hazard.Toml_utils.warn_parse_error(the helper that printed the warning) is replaced byfail_parse_errorthat raisesFailure, caught bymain.ml's top-level handler. Batch config files (which already returnedErrorfrom a result-typed API) keep their result-based behavior; only the redundant stderr warning is dropped there. - tests:
test_cmd_gen_watch's "fails for nonexistent package" case now runs withZORYN_TEST_FIXTURESset, so the binary uses fixture-mode HTTP instead of hitting live Debian/Gentoo/Arch endpoints. Without it the test was flaky on CI — a transparent proxy or stray upstream hit could mark a bogus package as "found" and the expected non-zero exit never came. - up: scheme detection now falls back to
tarball-watchwhen every structural git-merge signal is absent but a.gear/watch(or acopy: *.watchrule /Source: *.watch) is present. Previously the scheme stayedunknownandzoryn upfailed with "no upstream URL found" for packages whose upstream is a plain HTTP tarball site (e.g.dhcp_probefromwww.net.princeton.edu). - up: for tarball-watch packages whose spec
Url:points at PyPI, the version now comes from PyPI and the watch file is consulted only to find the matching tarball download URL. Previously the pipeline searched for a git tag matching the PyPI version (fails without an upstream git remote, e.g.python3-module-grpcio-tools), and the pure-watch fallback was prone to picking upstream pre-release tags like1.80.0-pre1over the real release1.80.0.
0.37.0 - 2026-04-22¶
Security¶
- up (hooks):
zoryn upnow lists every.gear/{up,merge-up}.d/*script by name before the[y/N]prompt when hooks would run unsandboxed (via--no-sandbox,--sandbox direct, a missing runner binary, orrunner=directin~/.zoryn), and strips control bytes and Unicode bidi/zero-width/line-separator/BOM hazards from the displayed names (CVE-2021-42574 "Trojan Source" class). The announcement, rejection notices, and prompt all share stdout so stderr redirection cannot silence part of the picture;--continueno longer inherits--no-sandboxor cached consent across processes.
Changed¶
- up (hooks) — BREAKING: symlinks under
.gear/up.d/and.gear/merge-up.d/(and the hook directories themselves if they are symlinks) are now refused: the target may differ from the displayed name, so a.gear/up.d/innocent -> /bin/shtrick would defeat the new list-before-prompt safeguard. Replace such entries with regular files (copy, include, or use a helper script that invokes the real target).
Fixed¶
- task approve/disapprove all: delegate the
alltarget to gyle in a single SSH call so stale client-side subtask state no longer yieldssubtask #N not foundcascades before the real subtasks are reached. Also warn when--revokeis combined with-m(gyle ignores the message). Closes #68. - task refresh: stop picking stale repo snapshots on girar —
get_last_done_task_idsorted candidate tasks bytask_iddescending and picked the firstDONEone, but girar does not commit tasks intask_idorder (a task with a higher id can finish before one with a lower id). The reference snapshot could therefore predate already-committed newer versions, sotask refreshreported "all packages up to date" even when subtasks were clearly behind repo. Now sort bytask_changeddescending (withtask_iddesc as a tiebreaker) to pick the latest-committedDONEtask. - task manage: no more "ghost symbols" on the subtask-detail and dependencies headers — the colored
#task_id / Sub #sub_id -- STATE -- repo -- owneroverlay on the subtask-detail screen was drawn one column left of the plain header text (" / Sub "is 7 chars, not 6), so each value's trailing char bled through from the uncolored underlay (Sub#1000,try 11,sisyphuss,nickell). The-- Dependencies of #idoverlay was off by one the other way (the prefix is 19, not 20 chars), rendering a doubled##id. Closes #70. - task manage: mouse selection no longer leaks across screen boundaries — clicking on TaskList, then entering the log viewer (or the other way around) used to leave the old
(row, col)selection in place; the log view then re-interpreted it as logical(screen_line, col)and painted reverse-video in the wrong spot, and the idle-release copy grabbed whatever happened to live at those coordinates. Selection overlays and the copy path are now gated on the screen the selection was started on. - task manage: mouse selection is cleared on every data refresh (manual
R/F5, SSE push, auto-poll fallback). The absolute(row, col)coordinates became stale as soon as the task list / detail was reloaded, so the copy-on-release path risked returning text from an unrelated row. - task manage:
Copied: …preview truncation is UTF-8 safe and no longer slices a multi-byte character mid-sequence (the status line usedString.subon byte offsets). - Shell mock layer (tests): SSH commands now require a fixture when
ZORYN_TEST_FIXTURESis set — previously the test harness silently fell back to realsshwhen the fixture file was missing, so tests could accidentally connect to a live build server against the developer's credentials (and pass or fail depending on their real task state). Missing fixtures now raise a clearShell fixture missingerror. Fall-through behaviour is preserved for git/gear/hsh because many tests rely on real local-only git operations on fixture-built repos.
Added¶
- task manage: full mouse text selection on the task list, detail, deps and subtask detail screens — drag to select a range, double-click to select a word. Triple-click on a task list entry's wrapped package block selects all
pkg=versionitems of that task (for other triple-click targets the clicked screen line is selected). Multi-row copies are joined with spaces (no line breaks) so a block of packages lands on a single paste-ready line. Copy fires automatically on mouse release (after ~0.4 s idle) via OSC 52. Single click on a#NNNtoken still selects just the digits and copies the bare number. - docs/site: project website at https://rider.altlinux.team/zoryn/ — a bilingual landing page plus full MkDocs Material documentation (Installation, Quick start, Configuration, Build farm, Hooks, Sandbox, Batch configs, per-command reference). Sources under
docs/{en,ru}/; thisCHANGELOG.mdat the repo root is the single source for the "Changelog" page (symlinked fromdocs/en/changelog.md; RU falls back to EN viamkdocs-static-i18n). - ci:
.forgejo/workflows/deploy-site.yaml— builds the MkDocs site and publishessite/to Forgejo Pages on every push tomastertouchingsite/,docs/,mkdocs.yml,overrides/, ordune-project. The landing's version badge is baked fromdune-projectat build time byoverrides/hooks.py— one source of truth for the displayed version. Closes #28. - up: new
[tarball] subdirconfig field — subdirectory name inside the tarball to extract (passed asgear-update --subdir=<value>). Supports{version}and{name}placeholders, e.g.subdir = "thunderbird-{version}". Validated against shell metacharacters, path separators (/) and directory-traversal values (.,.., or a..prefix) before use so it cannot escape the intended extraction directory. Useful for tarballs with multiple top-level entries like Mozilla source tarballs - up: new
parser = "mozilla"in[changelog]— fetches CVEs from mozilla/foundation-security-advisories for Firefox, Firefox ESR, Thunderbird (override withmozilla-product).
0.36.0 - 2026-04-15¶
Added¶
- task manage: show approval/disapproval status in the task list — aggregated
✓/✗line with usernames between task header and package list - Persistent search history in TUI search inputs (test-rebuild, up, task manage). Use Up/Down (or Ctrl+P/Ctrl+N) to walk previous patterns and Ctrl+R for reverse incremental search, readline-style. History is kept in
~/.local/state/zoryn/<tui>.search.history(max 500 entries per TUI). Closes #61. - build --section: new
bbandbavalues (with long-form aliasespkgbinaryandpkgall) — runrpmbuild -bb --short-circuit/-ba --short-circuitin the existing chroot to repackage only (Processing files→ rpm, or src.rpm + rpm), reusing the%prep/%build/%installartifacts from the previous build. Useful when iterating on%files, packaging macros, or metadata without rerunning the full pipeline throughhsh.
Changed¶
test-rebuildlog-viewer: pressing Enter on an empty search now re-runs the most recent history entry instead of clearing the match highlight.- build (python auto-deps): auto-fix retry now uses
hsh-rebuildto reuse the existing chroot instead of a cleanhshbuild, significantly reducing rebuild time since all BuildRequires are already installed. Falls back to clean build if chroot is unavailable. (#56)
Fixed¶
- task rebuild:
--skip <pkg>now filters the dependency list in--deps-by-pkg/--from-log/default modes as well (previously silently ignored outside--all-subtasks); numeric subtask IDs in--skipstill apply only to--all-subtasksand now emit a warning elsewhere. - up: upstream fetch no longer aborts when a local tag would clobber an upstream tag — the fetch is split into a mandatory branches step (
--no-tags) and a non-fatal tags step. Errors now name the spec file and the real reason instead of the generic "No upstream remote available". - submit: no longer attaches bogus
--depson tasks from a different build chain.Girar.is_repo_fresherused a flatbranch_sorting_orderindex, so p11 (p-chain) was ranked "fresher" than c10f2 (c-chain) andfind_upstream_tasksfed those tasks into--deps. It now consultsAlt_branches.get_build_predecessors, which understands the real chainssisyphus → p11 → p10 → p9 → p8andsisyphus → c10f2 → c9f2. For c10f2 only sisyphus is considered upstream; p11/p10 tasks are ignored. (#65) - build --section install: drop
--nocheckfrom therpmbuildarguments. ALT Linuxrpmbuilddoes not accept--nocheck(it's ahsh-level option), sozoryn build --section installfailed immediately with--nocheck: unknown option. The flag was also redundant:--short-circuittellsrpmbuildto jump straight to the requested stage, so-bi --short-circuitruns only%installand%checkis never reached. - build (python auto-deps): auto-fix now detects when upstream has changed the check dependencies source (e.g., switching from setuptools to hatchling) causing a source's deps to become empty in
pyproject_deps.json. Previously the build would silently succeed with cached chroot packages but fail on fresh builds. Now zoryn aborts with a clear message asking to update the%pyproject_deps_resync_check_*macros in the spec file. (#62)
0.35.0 - 2026-04-09¶
Added¶
- commit: new
zoryn commitcommand — native replacement forgear-commit, extracts changelog from spec file as commit message; supports--spec,-m,--no-edit,--amend,-aflags. By default opens the message in the same editorgit commituses (resolved viagit var GIT_EDITOR). For vim-family editors (vim/vi/nvim/gvim) the prefilled message is piped through stdin and the buffer is renamed via:file, which makes vim mark the buffer as modified — soZZ,:wq,:xon an unchanged prefilled buffer all commit (unlike opening vim directly on a file, where these commands are no-ops on unchanged content). Save/abort detection is by file existence::q!aborts, empty buffer aborts ("commit message is empty"),:cqaborts with non-zero exit. For non-vim editors falls back togit commit -F file --edit, which has standard git semantics. ExportsGEAR_COMMITandGEAR_COMMIT_AMENDenv vars for git hooks compatibility
Changed¶
- task rebuild / batch rebuild: local SRPM lookup is now stricter — only files whose release starts with
altare considered. Previously the legacyHasher.find_srpmaccepted any release suffix, so a directory containing third-party rebuilds (e.g.foo-1.0-1.fc38.src.rpm,foo-1.0-1.mga9.src.rpm, or vendor packages with noaltprefix at all) could be matched by name. After the unification ontoBuilder.find_srpm_in_path(which shares its matcher with the existingBuilder.find_srpmused everywhere else), such files are silently ignored andtask rebuildwill print "No src.rpm found" for them. If you mirror non-ALT SRPMs into your[sources] srpms_path, keep them in a separate directory;task rebuildonly consumes ALT-tagged SRPMs (this matches the behaviourBuilder.find_srpmhas always used for remote builders). Package names are now also validated throughGau_common.Validation.validate_package_namebefore lookup, mirroring the rest of the builder API.
Fixed¶
- task rebuild / batch rebuild: SRPM lookup no longer matches files whose name shares a suffix with the requested package. Previously the matcher in
Hasher.find_srpmused an unanchored regex (Re.eosonly, noRe.bos), so a search forqt6-webenginewould substring-match insidedqt6-webengine-6.10.2-alt0.dde.1.src.rpmand return the wrong file. The duplicate path-based finder has been removed: bothbin/zoryn/cmd_task_rebuild.mlandlib/batch_pipeline/batch_pipeline.mlnow use the newBuilder.find_srpm_in_path, which shares its fully-anchored matcher (^pkg-…$) withBuilder.find_srpm. The "no src.rpm found" warning intask rebuildwas also updated to describe the new pattern (%s-{version}-alt{release}.src.rpm) instead of the old permissive one. - up: state file location is now resolved via
git rev-parse --absolute-git-dirinstead of the hardcoded.git/zoryn-up-state.jsonpath, sozoryn up(and the build pipeline it drives) works inside agit worktree. Previously the top-level.gitin a worktree is a regular file (agitdir:pointer), not a directory, so writing the state file failed withENOTDIR. The new path lives in the per-worktree git directory (.git/worktrees/<name>/zoryn-up-state.json), which is also naturally isolated — two parallelzoryn upruns in two worktrees of the same package no longer collide. (#60) - up: scheme detection on packages whose
.gear/upstream/remotesexists (or whose spec carries only aUrl:to a public forge, noVCS:) no longer requires the maintainer to add the localupstreamremote by hand and no longer takes minutes on repos with hundreds of merge commits.stage_detectnow resolves the upstream URL from.gear/upstream/remotes→ specVcs:→ specUrl:(withGit_url.normalize+is_github_or_gitlab_urlfilter) and fetches BEFORE running scheme detection. The remote-discovery cascade is conservative and never destroys maintainer state: (1) it first walks all local remotes and reuses the one whose canonical URL matches the determined upstream URL — regardless of its name, sovendor/tools/phobos/etc. conventions are honored; (2) failing that, it respects an existingupstreamremote even when its URL differs from the spec (the maintainer may have pointed it at a mirror or fork deliberately); (3) only when neither match exists does itgit remote add upstream <url>from the determined URL. Fetches use--tagswithout--force, so packaging-side tags whose name coincides with upstream tags are preserved. With upstream fetched, scheme detection runs a new structural fast path: pick the upstream tag matching the current spec version (v<v>,<v>,release-<v>,release_<v>,rel-<v>,<pkg>-<v>), VERIFY it actually came from upstream by checking that its commit is reachable fromrefs/remotes/<upstream>/*(so a maintainer-side packaging tag with the same name does not poison the result), and run a singlegit merge-base HEAD <tag>— non-empty result means HEAD shares SHAs with upstream →git-merge; empty means tarball-imported synthetic SHAs →tarball-watch. Tarball imports cannot fake content-hash identity. Falls through to the legacy Tier 1 / Tier 2 cascade when no matching upstream tag exists locally OR when no candidate is verified as upstream.stage_fetchnow reuses the same safe helper as a fallback whenstage_detectcould not determine a URL, and skips the redundant remote-setup step when the URL is already instate.upstream_url. (e.g.microsoft/onnxruntime: scheme detection went from ~30s of slow tag walking to a singlemerge-basecall.) - up: Tier 1 (changelog-history scheme detector) now short-circuits on the first upstream-shaped merge instead of eagerly classifying every merge in the version-bump range — for packages with hundreds of merges between two version bumps and no configured upstream remote, this turns an O(N) walk into O(1) on the common case. The "all merges proven non-upstream → Tarball" branch still requires walking the whole range and remains correct.
- up: scheme detection no longer misclassifies git-merge packages as
tarball-watchwhen the maintainer uses non-standard merge commit messages such asMerge upstream tag 'X'(e.g. oneDNN). The detector now classifies the last update structurally. The primary signal is commit-identity via the upstream remote: a merge is upstream-shaped iff any non-first parent is reachable fromrefs/remotes/<upstream>/*, where the upstream remote is resolved from.gear/upstream/remotes(proper git-config parsing with[remote "name"]section headers, case-insensitive keys,#/;comments), specVcs:/Url:, or a local remote namedupstream/vendor. SHA1 identity is cryptographic — tarball imports cannot fake it. When a remote is configured, its negative answer is authoritative (no fall-through to weaker tag heuristics), so packaging-branch merges with coincidental tags cannot be misclassified. When no remote is configured, the detector falls back to tag-based heuristics (tag pointing at merge commit, tag in parent diff). Tier 1 walks the spec'sVersion:-bump history with--follow --topo-order, verifies ancestry to skip meaningless ranges after rebase, and handles octopus merges.is_github_or_gitlab_urlexplicitly excludes ALT-internal hosts (*.altlinux.org,*.basealt.ru, etc.) so ALT's gitlab-based infra is never misclassified as upstream. Multi-subproject packages (dmd-style with[remote "tools"]+[remote "phobos"]) are handled correctly by matching gear section names against local remote names. - check version-up: spec file lookup now respects
spec:directive in.gear/rules. Previously the command only scanned the repository root and printed "No spec file found, use --from for simulation" for packages whose.speclives inside.gear/(e.g.python3-module-hypothesis). - up: PyPI tag lookup now finds tags whose
.gear/version-uppattern carries a decorative prefix (e.g.hypothesis-python-{major:+}.{minor:+}.{patch:+}) without needing an explicittemplate;[version] filteris still respected. (#57) - gen version-up: when run from
.gear/or any subdirectory of a git repository, write.gear/version-upto the repository root instead of creating a nested.gear/.gear/version-up. The-C/--chdiroption still defines the working directory explicitly. (#59) - builder add:
--hasher-dirwas ignored in--multi-addmode; now supports{hasher_number}template (e.g.,--hasher-dir '/tmp/h_{hasher_number}'). Without placeholder, template is derived automatically with a warning. Priority:--hasher-dir>--based> default~/hasher_{hasher_number} - task add: TAB completion for package names now searches across all branches instead of filtering by the task's target branch — previously e.g.
zoryn task add 411104 copy ocaml-l<TAB>(task for p11) hidocaml-lzmabecause it does not yet exist in p11 - gen pypi2spec: emit
VCS:tag when the repository URL comes only fromhome_page(noproject_urls.Repository), e.g.fasttext. - gen pypi2spec:
--urlnow overrides any URL discovered in PyPI metadata (was: only used as a fallback) and is reflected in the spec'sVCS:line. - gen pypi2spec: normalise PyPI repository URLs through a shared
Gau_common.Git_urlhelper — handlesgit+https:/git:prefixes, trailing slashes, and.gitsuffix uniformly across all code paths.
0.34.0 - 2026-04-06¶
Added¶
- task refresh:
--from <branch>flag for cross-branch sync — detect stale subtasks and replace with copy from source branch - task refresh:
--types <types>flag to control which subtask types are checked (copy, rebuild, build=gear+srpm, ALL) - task test-rebuild: warn about packages already FTBFS on beehive before building; add
--skip-ftbfsto skip per-architecture build tasks where the package is already FTBFS - TAB completion for package names from Repoteka API in
clone,check,task, andgen watchcommands
Fixed¶
- task ls/show: fixed missing package names for delete/copy/rebuild subtasks — used
pkgnamefield from Tasks API instead of absentpackage
Changed¶
- build: unified single-builder and multi-builder batch build code paths — removed ~300 lines of duplicated logic and the first-package special case
0.33.0 - 2026-04-05¶
Added¶
- tui: mouse text selection in log viewer — click+drag to select,
yto copy original text (without wrapping) to clipboard via OSC 52 (#52) - tui:
Space(page down) andb(page up) keys in log viewer — matching standardlesskeybindings (#50) - Syntax highlighting for log output using sublime-syntax engine (#46)
- build:
--sectionflag to run a specific rpmbuild section (prep,build,install,check) in an existing chroot with automatic spec file sync,--rpmbuild-argsfor passing extra rpmbuild arguments — for iterative build development and debugging - build: configurable log filename template
log_filenamein[build]config section (~/.zoryn). Supported variables:{builder},{batch},{pkgname}. Default:build.{batch}.{builder}.log - build, up:
--skip-checknow supportsrpmbuildvalue to skip%checksection via--rpmbuild-args='--without=check'; bare--skip-checkwithout value skips everything (rpmbuild + post-build checks) (#45) - gen pypi2spec:
%pyproject_runtimedeps_metadatafor upstream runtime dependencies (#48) - gen pypi2spec:
AutoReq: yes, nopython3to disable python3 autoreq (#48) - gen pypi2spec:
%global _unpackaged_files_terminate_build 1in generated spec (#48)
Changed¶
- tui: log viewer line wrapping is now character-based instead of word-based — better for logs with long strings without spaces (#54)
- build: default log filename format:
build-log.{builder}->build.{batch}.{builder}.log - build: removed separate
.errlog files for parallel builds — stderr is now written to the main log - gen pypi2spec: spec filename changed from
python3-module-<name>.specto<normalized-pypi-name>.spec(e.g..gear/hdbscan.spec) (#48) - gen pypi2spec:
Url:tag now always useshttps://pypi.org/project/<name>/instead of homepage from metadata (#48) - gen pypi2spec: removed
%docline from%files— don't package docs and license files (#48)
Fixed¶
- tui: last character lost when wrapping long lines in log viewer — byte offset for syntax highlighting was calculated incorrectly due to a missing space at the wrap boundary (#53)
- build: show SKIP (yellow) instead of FAIL (red) when architecture is excluded in spec file (
ExcludeArch) — detected from build log (error: Architecture is excluded: ...) (#49) - gen pypi2spec: add
%define pypi_nameto generated spec and use%{pyproject_distinfo %pypi_name}instead of%pyproject_distinfo %name— fixes incorrect dist-info path matching (#44)
0.32.0 - 2026-03-31¶
Added¶
- submit: automatically squash duplicate release commits — when re-running
zoryn submitafter changes, if the previous commit has the same version-release subject (e.g.6.151.10-alt1), amends it instead of creating a duplicate - PyPI version source: for Python packages with
URL: https://pypi.org/project/<name>/in spec,zoryn upandzoryn check versionnow use PyPI as the source of truth for version numbers instead of git tags — prevents updating to yanked (withdrawn) versions; PEP 440 normalization handles version format differences between PyPI and git tags; use--tagto bypass PyPI detection - task test-rebuild --top: full TUI rewrite using LTerm — no more screen flickering, syntax-highlighted build logs with search (
/,n/N), full-screen log viewer with follow-mode, build log list (l) showing all results, help screen (F1), hotkeys synced withtask manage - tui_logview: new shared library for log viewer logic (classifier, word wrap, caching, search) — used by both
task manageandtask test-rebuild zoryn task abortcommand — abort a running task by ID (with TAB completion for task IDs)zoryn task approve/zoryn task disapprovecommands — approve or disapprove task subtasks from CLI (by subtask number, package name, orall), with optional-mmessage and--revoke- Changelog CVE parser: support
CVE:YYYY-NNNNNformat (used by ISC Kea/BIND) — normalized to standardCVE-YYYY-NNNNN - Changelog version header: support
Product X.Y.Z (status) released on Dateformat (ISC ChangeLog) - Tests for
Io.read_package_list_file(path traversal, comments, blank lines, errors) - Tests for
Custom_packages.collect(combined--package/--packages-filelogic) - Tests for
List_utils.chunkwithsize <= 0 - Test for
Stats.create ~custom_packages:[]normalization toNone - Extract
Custom_packages.collectintotest_rebuildlibrary for testability - task test-rebuild:
--package PKG(repeatable) and--packages-file FILEflags to rebuild a custom set of packages instead of the full RDB dependency tree
Changed¶
- task manage: log viewer logic extracted to shared
tui_logviewlibrary (no user-visible change) - task test-rebuild --top: event loop migrated from
Unix.selectto Lwt for LTerm integration
Fixed¶
- gen version-up: tags with package name prefix and v-prefixed version (e.g.
--tag freeipmi-1-6-17 --to-version v1.6.17) generated wrong pattern with 4 groups instead of recognizingfreeipmi-as prefix — now stripsv/Vprefix before matching - task test-rebuild --top: pressing
[f]to view failures on the completion screen did nothing —renderalways showed the completion box ignoringview_mode, and the post-completion input loop never re-rendered - task test-rebuild --top: pressing Enter on failures without log files (e.g. "SRPM not found") did nothing — now shows the error message inline with a "press any key" prompt
- package_spec:
is_alt_releasenow accepts branch releases (alt1.M110P.1) and architecture-specific releases (alt1_e2k1) — previously only digits and dots were allowed afteralt, causing release to be merged into version and breaking version comparison - task test-rebuild:
--continue-no-refreshnow updatespackages_to_rebuildin stats when package list changes between runs, preventing stale diff comparisons on subsequent--continue - task manage: approve/disapprove with message and run-with-commit message failed in TUI —
interactive_quietmode appended `