Skip to content

Sort handheld dev pkgvers above upstream omarchy-dev - #82

Merged
DanielJBurbridge merged 2 commits into
feat/ally-parity-integration-20261111from
fix/ally-pkgver-sort-20261117
Nov 17, 2026
Merged

DanielJBurbridge merged 2 commits into
feat/ally-parity-integration-20261111from
fix/ally-pkgver-sort-20261117

Conversation

@DanielJBurbridge

@DanielJBurbridge DanielJBurbridge commented Nov 17, 2026 •

Copy link
Copy Markdown
Owner

What changes

Handheld omarchy-dev and omarchy-settings-dev builds used dev.<yyyymmddHHMM>.<sha8>. pacman's vercmp ranks a numeric segment above an alphabetic one, so upstream 4.0.0.r7967.g3b36b9e sorted newer. omarchy update replaced the Ally packages with that upstream build and sddm went into a relogin loop.

Chose a variant of (a). Not (b) epoch.

omarchy dev pkg-test now sets:

<base>.<commit UTC yyyymmddHHMM>.r<count>.handheld.<sha8>[.dirty]

<base> is the checkout version file with a trailing .alpha removed, so 4.0.0.alpha publishes 4.0.0, the base upstream omarchy-dev already uses. The stamp is the committer time in UTC and sits in the segment where upstream puts the letter r. r<count> is still the total commit count, as the omarchy-dev PKGBUILD describes it.

Literal (a), <base>.r<count>.handheld.<date>.<sha>, loses as soon as upstream's commit count passes this fork. The fork is only about a hundred commits ahead of r7967. Epoch stays above every later upstream version, including 4.0.1 and 9.0.0, so coming back needs pacman -U or an epoch bump on each of them. This scheme stays above every 4.0.0.rN and yields when the base moves to 4.0.1 or later. Until that bump, going back is sudo pacman -U.

The next deploy upgrades an Ally that is still on dev.*.

This follows Arch's VCS pkgver and pacman's vercmp rule that a numeric segment is newer than an alphabetic one. It is the same idea as a CachyOS pkgrel of 1.1: newer than the package that is current, older than the next real upstream version. The Arch PKGBUILD guidelines treat epoch as a last resort because it blocks that next version.

Claims follow-ups L220–L229. L220 records the fix. L221 is open: when upstream's base moves to 4.0.1, omarchy update can bring the relogin loop back. The proposed hook is not in this PR. Re-merged onto e9fca470 (#81) and kept L230.

The handoff realign step is sudo pacman -Syu omarchy-settings-dev omarchy-dev, not a partial pacman -Sy.

Model: grok-4.7

Verification

vercmp from pacman 6.0.2. A positive result means the first version is newer.

vercmp 4.0.0.r7967.g3b36b9e dev.202611170222.e742281c → 1
vercmp 4.0.0.alpha.r8072.handheld.202611170222.e742281c 4.0.0.r7967.g3b36b9e → -1
vercmp 4.0.0.r8072.handheld.202611170222.e742281c 4.0.0.r7967.g3b36b9e → 1
vercmp 4.0.0.r8072.handheld.202611170222.e742281c 4.0.0.r8200.gabcdef0 → -1
vercmp 1:4.0.0 4.0.1.r1.g0000000 → 1
vercmp 1:4.0.0 9.0.0 → 1
vercmp 4.0.0.202611170222.r8072.handheld.e742281c 4.0.0.r7967.g3b36b9e → 1
vercmp 4.0.0.202611170222.r8072.handheld.e742281c 4.0.0.r12000.gabcdef0 → 1
vercmp 4.0.0.202611170222.r8072.handheld.e742281c 4.0.0.r999999.gabcdef0 → 1
vercmp 4.0.0.202611170222.r8072.handheld.e742281c dev.202611170222.e742281c → 1
vercmp 4.0.0.202611171200.r100.handheld.aaaaaaaa 4.0.0.202611170222.r8072.handheld.e742281c → 1
vercmp 4.0.1.r1.g0000000 4.0.0.202611170222.r8072.handheld.e742281c → 1
vercmp 5.0.0 4.0.0.202611170222.r8072.handheld.e742281c → 1

bash test/shell.d/dev-pkgver-test.sh passed those comparisons against versions the script built from a fixture checkout. With vercmp taken off PATH, that part of the file skips and the file exits 0.

Screenshots

No visible change: package version strings only.

Open in Web Open in Cursor 

Model: grok-4.7
Co-authored-by: Daniel Burbridge <DanielJBurbridge@users.noreply.github.com>

@DanielJBurbridge DanielJBurbridge left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

dev-red review: FAIL (mechanical only) at a1ced8d

B1: this head conflicts with integ e9fca47 (#81 merged) in docs/handheld/alpha-followups.md, where L220 and L230 sit next to each other. Re-merge. Tests also has to go green on the new head; it was still running on this one. Once both are done I'll pass it on a narrow delta check, and the code needs no other change.

(1) vercmp: all claims are verified with a port of alpm's rpmvercmp. Taking 4.0.0.202611170358.r8123.handheld.a1ced8d6 as our build:

  • It is newer than 4.0.0.r7967.g3b36b9e, 4.0.0.r999999.gffffffff and dev.202611170222.e742281c. The 12-digit stamp is a numeric segment against the letter r, and a numeric segment wins. Against dev, the numeric 4 beats the letters.
  • It is older than 4.0.1.r1.g0000000.
  • Date beats count: …202611180000.r1… is newer than …202611170358.r8123….
  • .dirty sorts above the clean build of the same commit. That's harmless.

(3) Upgrade from the dev.* builds: a plain pacman -U is an upgrade, because the numeric segment beats dev. A deliberate move back to upstream while it's still 4.0.0.r* is an explicit downgrade (pacman -U of the repo or saved package, or pacman -S), and the HANDOFF says so. Without that, omarchy update keeps ours.

(2) The 4.0.1 yield: medium follow-up (L220 range), not blocking this PR. It's acceptable for the alpha only if it fails safe. As things stand, the day upstream moves to 4.0.1, a routine omarchy update silently swaps our builds out again, and the Ally falls back into the same sddm relogin loop, a device with no way in. Documenting it isn't enough, because the person running omarchy update won't read HANDOFF first. I wouldn't add an epoch or IgnorePkg: an epoch sticks forever, and IgnorePkg is hidden pacman.conf state that also blocks intended updates. The safer fix is a pacman hook in omarchy-gaming-session on omarchy-dev and omarchy-settings-dev upgrades. When the installed pkgver lacks .handheld., it turns off handheld autologin and leaves the normal desktop greeter, with a message saying how to reinstall the handheld build. That stops the loop whatever upstream does.

Low: HANDOFF:34 uses pacman -Sy <pkgs>, which is a partial upgrade on Arch. Use pacman -Syu first, or pacman -S without -y.

There's a Model trailer (grok-4.7), and no follow-up lines are removed.

Keep L220 and L230. L221 records the open relogin risk when upstream
moves to 4.0.1. The handoff realign step is a full pacman -Syu.

Model: grok-4.7
Co-authored-by: Daniel Burbridge <DanielJBurbridge@users.noreply.github.com>

@DanielJBurbridge DanielJBurbridge left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

dev-red delta review: PASS at ea471ee

  • B1 is fixed. e9fca47 is merged in, and git merge-tree against current integ 10a24c6 is clean. Nothing is removed from the follow-ups.
  • bin/ is unchanged since a1ced8d. The only test/ change is #81's own test coming in with the merge, so the vercmp verification from review 5932603045 still holds.
  • L220 points to #84's hook, and HANDOFF:34 now uses -Syu.
  • Model trailer: grok-4.7.
  • Actions test and i18n-check are green on this head.

Cleared to merge at this sha.

@DanielJBurbridge
DanielJBurbridge merged commit bbaac86 into feat/ally-parity-integration-20261111 Nov 17, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants