fix(drivers): never lose a driver selection to a Core update by frahlg · Pull Request #1428 · srcfl/ftw

frahlg · 2026-09-25T07:26:04Z

Follow-up to #1422 and #1426, from a Codex review and an independent review of today's merged driver work.

Defects found

Retirement was destructive (independent review). RetireSupersededByBundled deactivated the owner's managed selection when a release bundled a version at least as new. An update trial that fell back, or ftw rollback, then ran the older release's older copy. The version installed early, for example to fix a bug, was lost.
A choice counted only against the bundled copy (independent review). Scenario: under a release with 1.3.0, the owner installs 1.3.3 early and goes back to 1.3.2. The choice was not kept, and the next release ran the rejected 1.3.3.
A failed retirement was never retried on that release (Codex). The release marker was saved anyway. The derived model below makes this moot.
used_by matched configured drivers by file name (Codex). An operator's own file elsewhere with the same name was credited to the catalog entry.

Change

Selections are never changed at start.
- driver-repository/active records what the owner selected.
- driver-repository/effective is derived from it and the release's drivers at every start and after every change. Paths resolve through it: config, catalog, API and driver inventory.
A selection runs when it is at least as new as the release's copy, or when the owner went back to it from a newer version that was running.
When a release brings a newer copy, that copy runs, and the selection is kept. An older release runs it again.
A rollback restores the choice it replaced, so recovering from a failed install leaves no mark.
The release marker is gone.
used_by compares the configured resolved path with the entry's file in its source directory.
Backups: the new symlinks live in the data directory and are archived like the existing ones.

Evidence

New or rewritten tests in go/internal/driverrepo/bundled_supersedes_test.go:
- a newer release driver runs, the selection is kept, and a rollback runs it again;
- the same version keeps the selection running;
- an older choice stays across releases;
- going back below the running version is a choice;
- a rollback restores the undone choice;
- "use bundled" forgets the choice;
- a newer or unbundled selection stays.
TestDriverCatalogCreditsOnlyTheFileThatRuns fails on the old basename matching.
go test passes for internal/driverrepo, internal/api, cmd/ftw, internal/driverinventory, internal/ftwcli and internal/backup.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7

chatgpt-codex-connector · 2026-09-25T07:26:20Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-25T07:28:15.041026Z	`88b6a6e`	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 88b6a6e6ee

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-09-25T07:28:12Z

+				continue
+			}
+			rel := filepath.FromSlash(strings.TrimPrefix(entries[i].Path, "drivers/"))
+			entries[i].UsedBy = usedBy[filepath.Clean(filepath.Join(root, rel))]


Canonicalize both sides before matching driver paths

When the catalog root is relative but the configured Lua path is absolute—e.g. FTW starts in /opt/ftw with -drivers=drivers while the config names /opt/ftw/drivers/demo.lua—both paths identify the same file, but filepath.Clean leaves one relative and one absolute, so this lookup produces no used_by entry. Absolute driver paths and relative -drivers values are both supported, and with this configuration ftw status can omit its Drivers line even though the cataloged driver is running; normalize both keys with filepath.Abs before comparing them.

Useful? React with 👍 / 👎.

Matches #1428: a newer release driver runs over an early selection without discarding it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7

…#1429) * fix(drivers): one id per driver, from the bundled copy to the channel Moves the pin to srcfl/device-drivers#132, where every driver's DRIVER table states the catalog id and version. FTW's identity rule for the old spellings (IdentifiesSameDriver) goes: a bundled copy and a channel artifact are the same driver when their ids are equal. - Catalog tests look drivers up by the catalog ids. - The recovery-snapshot provenance test for esphome_dsmr is gone: the bundled file is now the driver's source, and sync-bundled-drivers.sh --check already binds every bundled file to the pinned commit. - setup.js recognises the Zap driver by its id `zap`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7 * docs(release): promote the driver channel as the last stable step drivers-stable only serves installs that read the channel (1.x-3.x) and the stable rows of the Versions list; 0.x runs the drivers its release bundles. Promoting it is one existing command after a stable Core release, not new automation. The proposal to promote it with Core goes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7 * fix(drivers): pin the merged one-id device-drivers commit Moves the pin to bcf4683, srcfl/device-drivers#132 as merged on main, instead of the branch commit it was tested against before its rebase. The changeset no longer says esphome-dsmr is gone: #132 keeps that catalog entry for sites that installed it. The stable promotion step says what to do when main has moved past drivers-beta. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7 * docs(agents): a driver selection survives updates and rollbacks Matches #1428: a newer release driver runs over an early selection without discarding it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7 * fix(drivers): pin device-drivers 43710c8 (esphome_dsmr 1.0.5) srcfl/device-drivers#134 puts esphome_dsmr's host API fields on their own lines; FTW reads DRIVER fields line by line and its publication check refused the 1.0.4 file. The bundled set now publishes cleanly. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7 --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

…runs Codex review of #1422 found two defects: - RetireSupersededByBundled recorded the release as reconciled even when a Deactivate failed, so a retirement blocked once (for example by an unwritable active directory) was never tried again on that release. The release marker is now saved only when every retirement succeeded. - The catalog's used_by matched configured drivers by file name. An operator's own file elsewhere with the same name was credited to the catalog entry, and `ftw status` reported a version that was not running. It now compares the configured resolved path with the entry's file in its source directory. Both new tests fail on the previous code. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7

Independent review of #1422 and #1426: - RetireSupersededByBundled deactivated the managed selection when a release bundled a version at least as new. A failed update trial that fell back, or `ftw rollback`, then ran the older release's older copy and lost the version the owner had installed early. - A choice counted only when it was older than the bundled copy at that moment, so going back from an early 1.3.3 to 1.3.2 under a release with 1.3.0 was not kept, and the next release ran the rejected 1.3.3. The selection is no longer changed at start. driver-repository/active records what the owner selected; driver-repository/effective, which paths now resolve through (config, catalog, API, driver inventory), is derived from it and the release's drivers at every start and after every change. A selection runs while it is at least as new as the release's copy, or when the owner went back to it from a newer version that was running. A rollback of an activation restores the choice it replaced, so recovering from a failed install leaves no mark. The release marker is gone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7

chatgpt-codex-connector Bot reviewed Sep 25, 2026

View reviewed changes

frahlg changed the title ~~fix(drivers): retry failed retirements and credit only the file that runs~~ fix(drivers): never lose a driver selection to a Core update Sep 25, 2026

frahlg and others added 2 commits September 25, 2026 10:00

frahlg force-pushed the fix/driver-review-followups branch from c651ce6 to fd74e18 Compare September 25, 2026 08:01

frahlg merged commit 2487c90 into master Sep 25, 2026
16 checks passed

frahlg deleted the fix/driver-review-followups branch September 25, 2026 08:05

This was referenced Sep 25, 2026

Drivers follow the Core release: remaining steps #1423

Open

fix(cli): name a channel driver that matches the release's version #1430

Merged

fix(drivers): recover to the release's driver over a kept selection #1434

Merged

Navigation Menu

fix(drivers): never lose a driver selection to a Core update - #1428

fix(drivers): never lose a driver selection to a Core update#1428
frahlg merged 2 commits into
masterfrom
fix/driver-review-followups

frahlg commented Sep 25, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot commented Sep 25, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Sep 25, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

frahlg commented Sep 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Defects found

Change

Evidence

Uh oh!

chatgpt-codex-connector Bot commented Sep 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Sep 25, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

frahlg commented Sep 25, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Sep 25, 2026 •

edited

Loading