Recover updates from retired package conflicts

[yashranaway]

Keep upstream package splits from wedging unattended updates.

<<< AI wording below >>>

Problem

pacman -Syu --noconfirm answers No to package-conflict removal prompts. With the current QEMU transition, qemu-common conflicts with the retired qemu-block-gluster, so an otherwise valid update stops every time.

Fix

• Retry with Pacman's narrowly scoped conflict answer only when every removal target was installed as a dependency and is absent from the configured sync repositories
• Leave explicitly installed, still-published, and mixed safe/unsafe conflicts for a human
• Keep the existing one-retry boundary so a second failure cannot loop

Fixes #6818.

Verification

• bash test/shell.d/update-file-conflict-test.sh
• bash -n bin/omarchy-update-system-pkgs bin/omarchy-update-system-pkgs-when-conflicted test/shell.d/update-file-conflict-test.sh
• git diff --check

[Copilot]

Pull request overview

Adds safe recovery for unattended Pacman updates blocked by retired dependency conflicts.

Changes:

• Detects dependency-installed packages absent from sync repositories.
• Retries once with Pacman’s conflict-removal answer enabled.
• Tests safe, unsafe, mixed, and direct-invocation scenarios.

Tip

If you aren't ready for review, convert to a draft PR.
Click "Convert to draft" or run gh pr ready --undo.
Click "Ready for review" or run gh pr ready to reengage.

Reviewed changes

Copilot reviewed 1 out of 3 changed files in this pull request and generated no comments.

File	Description
bin/omarchy-update-system-pkgs	Conditionally enables conflict removal on validated retries.
bin/omarchy-update-system-pkgs-when-conflicted	Validates retired dependencies and initiates one retry.
test/shell.d/update-file-conflict-test.sh	Covers package-conflict recovery and safety boundaries.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[dhh]

Closing in favour of #6830, which takes the other route out of #6818: hand the conflict question back to the person running the update rather than prove an answer safe.

Two things this branch runs into that are worth recording, both reproduced against pacman 7.1 with real packages in a fakeroot:

1. Under --noconfirm, pacman prints its questions to stdout, not stderr. omarchy-update-system-pkgs captures only 2>"$errors", so the :: … Remove …? [y/N] line never reaches the report the handler parses — conflict_removal_count is 0 and the new block never fires. Only the two error: lines are on stderr.

2. The real prompt ends [y/N] with a trailing space, so the $-anchored lookahead (?=\? \[y/N\]$) cannot match it even when the line is present.

The tests pass because write_raw_report hand-writes the fixture, so neither the stream nor the trailing space is modelled.

The vetting itself was sound, for what it's worth — -Qd plus -Qm, all-or-nothing, and the three-variable guard — and --ask 4 does what it claims: it inverts the default for only the masked question type, leaving Replace X with Y? [Y/n] answered Yes as before. Thank you for digging into this one.

— 🤖 Claude, posting on behalf of @dhh