Follow-ups from batch-v07c review (#848/#851/#859 residual minors) #867

Closed
opened 2026-10-08 22:33:33 +00:00 by Cordy · 5 comments
Owner

These are the residual findings parked by the batch-v07c reviews (PR #866). None of them blocked the merge. The rulings are in the PR's process notes.

Auth (#848)

  1. A future time on first sighting stops change detection. If Cairn's first sighting of a user returns a directory time in the future, pwdAt stores that future time (the cut itself is capped at now). Later real changes are then ignored until real time passes it. This needs a directory clock more than 2 minutes ahead, or an absurd pwdLastSet. The effect falls back to the pre-#848 behaviour: sessions end at their expiry, never an extra sign-out. Fix: store min(PwdChanged, now) as pwdAt.
  2. A "0" first sighting can cut twice. A first sighting of AD "0", then a lagging parsed T0, then "0" again can cut and audit once more. No fresh session is lost, since no bind succeeds while "0" is set.
  3. The rejection path is not scoped by source (pre-existing): RevokeUser(username, "") runs when the IdP reports a user deleted or disabled. In OIDC mode this also signs out a same-named local account. Scope it by source like RevokeUserIssuedBefore.
  4. FILETIME overflow. A pwdLastSet value above about 2.08e17 overflows to a garbage time. Guard it with MaxInt64/100.
  5. Keycloak test gaps. No test covers picking the newest credential from a descending array, or a missing/0 createdDate.

Notices / app passwords (#851)

  1. AddNotice does not roll back memory on a save failure (consistent with MarkAll/ClearAll/SetMuted). Consider rollback for all setters together.
  2. feed() whitelists notice kinds without a comment, and the sort of items with an equal At is not stable.
  3. revokeAppPasswords:true is silently ignored (204) when app passwords are off. Answer 409 or 400 instead?
  4. The reset 500 is matched by message text in the UI. An error code would be sturdier.
  5. The revoke toast count comes from the server, while the button count comes from the list. They differ only in a race.

UI (#859)

  1. The qa694 dash guard exempts the exact substring . _ - @ (narrow but special-cased).
  2. Badge order is name, username, You, CHECK NAME. Check that this reads well.

Housekeeping

  1. Eight Go files outside these changes are not gofmt-clean (e.g. internal/peering/client.go).
These are the residual findings parked by the batch-v07c reviews (PR #866). None of them blocked the merge. The rulings are in the PR's process notes. ## Auth (#848) 1. **A future time on first sighting stops change detection.** If Cairn's first sighting of a user returns a directory time in the future, `pwdAt` stores that future time (the cut itself is capped at now). Later real changes are then ignored until real time passes it. This needs a directory clock more than 2 minutes ahead, or an absurd `pwdLastSet`. The effect falls back to the pre-#848 behaviour: sessions end at their expiry, never an extra sign-out. Fix: store `min(PwdChanged, now)` as `pwdAt`. 2. **A "0" first sighting can cut twice.** A first sighting of AD "0", then a lagging parsed T0, then "0" again can cut and audit once more. No fresh session is lost, since no bind succeeds while "0" is set. 3. **The rejection path is not scoped by source** (pre-existing): `RevokeUser(username, "")` runs when the IdP reports a user deleted or disabled. In OIDC mode this also signs out a same-named local account. Scope it by source like `RevokeUserIssuedBefore`. 4. **FILETIME overflow.** A `pwdLastSet` value above about 2.08e17 overflows to a garbage time. Guard it with `MaxInt64/100`. 5. **Keycloak test gaps.** No test covers picking the newest credential from a descending array, or a missing/0 `createdDate`. ## Notices / app passwords (#851) 6. `AddNotice` does not roll back memory on a save failure (consistent with `MarkAll`/`ClearAll`/`SetMuted`). Consider rollback for all setters together. 7. `feed()` whitelists notice kinds without a comment, and the sort of items with an equal `At` is not stable. 8. `revokeAppPasswords:true` is silently ignored (204) when app passwords are off. Answer 409 or 400 instead? 9. The reset 500 is matched by message text in the UI. An error `code` would be sturdier. 10. The revoke toast count comes from the server, while the button count comes from the list. They differ only in a race. ## UI (#859) 11. The qa694 dash guard exempts the exact substring `. _ - @` (narrow but special-cased). 12. Badge order is name, username, You, CHECK NAME. Check that this reads well. ## Housekeeping 13. Eight Go files outside these changes are not gofmt-clean (e.g. `internal/peering/client.go`).
Author
Owner

Live test on Bao, v0.6.324 (2026-10-09)

Passed:

  • Created 2 app passwords for nikola-test. The count showed "· 2 app passwords" and the menu showed "Revoke app passwords (2)" on the directory row.
  • The confirmation listed both by name and creation date.
  • Revoke behaved as expected:
    • toast "2 app passwords revoked.";
    • the count is gone;
    • the user's list is empty;
    • audit app-passwords-revoked-by-admin count=2 via=row;
    • bell item "Your app passwords were revoked", and its link opens Account › App passwords.
  • A second revoke returns {"revoked":0} with no audit line. Without user it answers 400.
  • The admin list carries no hash or sub.
  • Reset for sharer1 (0 app passwords): no checkbox, 204, toast "Password reset.".
  • Add account in DE with "Max Muster" shows the translated rule error.
  • The audit subtitle and the merged toggle text are as specified.
  • The session survived a directory re-check after the deploy, with no credentials warning in the log, so #848 does not sign people out falsely against the real Keycloak.

Not testable live:

  • reset with revoke: the local test accounts have no app passwords, and I can't sign in as them;
  • CHECK NAME: no non-conforming names exist;
  • the #848 cut itself: it needs a Keycloak password change.

New small findings:
14. The bell doesn't refresh when opened. It showed the feed from the last 60 s poll, so the new item appeared only after the next poll or a reload (pre-existing bell behaviour). Consider refetching on open.
15. The confirmation title breaks a hyphenated name across lines ("nikola-" / "test's"). Consider overflow-wrap:anywhere only for long names, or a non-breaking hyphen.
16. Mixed languages in the German UI. "Users & access", "Local accounts", "Directory accounts", the "Stored on this instance" note and the tab heading stay English (pre-existing, cf. #843).

## Live test on Bao, v0.6.324 (2026-10-09) **Passed:** - Created 2 app passwords for nikola-test. The count showed "· 2 app passwords" and the menu showed "Revoke app passwords (2)" on the directory row. - The confirmation listed both by name and creation date. - Revoke behaved as expected: - toast "2 app passwords revoked."; - the count is gone; - the user's list is empty; - audit `app-passwords-revoked-by-admin count=2 via=row`; - bell item "Your app passwords were revoked", and its link opens Account › App passwords. - A second revoke returns `{"revoked":0}` with no audit line. Without `user` it answers 400. - The admin list carries no hash or sub. - Reset for sharer1 (0 app passwords): no checkbox, 204, toast "Password reset.". - Add account in DE with "Max Muster" shows the translated rule error. - The audit subtitle and the merged toggle text are as specified. - The session survived a directory re-check after the deploy, with no credentials warning in the log, so #848 does not sign people out falsely against the real Keycloak. **Not testable live:** - reset with revoke: the local test accounts have no app passwords, and I can't sign in as them; - CHECK NAME: no non-conforming names exist; - the #848 cut itself: it needs a Keycloak password change. **New small findings:** 14. **The bell doesn't refresh when opened.** It showed the feed from the last 60 s poll, so the new item appeared only after the next poll or a reload (pre-existing bell behaviour). Consider refetching on open. 15. **The confirmation title breaks a hyphenated name across lines** ("nikola-" / "test's"). Consider `overflow-wrap:anywhere` only for long names, or a non-breaking hyphen. 16. **Mixed languages in the German UI.** "Users & access", "Local accounts", "Directory accounts", the "Stored on this instance" note and the tab heading stay English (pre-existing, cf. #843).
Author
Owner

Live test, part 2 (cairn-enc, v0.6.324): reset with revoke passed.

Setup: as test-local, I created 2 app passwords. As test-local, a non-admin, the admin endpoints returned 403.

As nikola-test I opened the Reset dialog:

  • The row showed "2 app passwords".
  • The checkbox read "Also revoke test-local's 2 app passwords", unticked.

After ticking the box and resetting:

  • The toast read "Password reset. 2 app passwords revoked."
  • The count and the list are empty.
  • The audit log has, in order: user-password-reset, auth-sessions-revoked count=1, app-passwords-revoked-by-admin count=2 via=reset.

The CHECK NAME badge stays covered by automated tests only: Add account now refuses non-conforming names, so it can only appear for older accounts.

**Live test, part 2 (cairn-enc, v0.6.324): reset with revoke passed.** Setup: as test-local, I created 2 app passwords. As test-local, a non-admin, the admin endpoints returned 403. As nikola-test I opened the Reset dialog: - The row showed "2 app passwords". - The checkbox read "Also revoke test-local's 2 app passwords", unticked. After ticking the box and resetting: - The toast read "Password reset. 2 app passwords revoked." - The count and the list are empty. - The audit log has, in order: `user-password-reset`, `auth-sessions-revoked count=1`, `app-passwords-revoked-by-admin count=2 via=reset`. The CHECK NAME badge stays covered by automated tests only: Add account now refuses non-conforming names, so it can only appear for older accounts.
Author
Owner

Added (Nikola, 2026-10-10): "All files" in the search bar stays English

Root cause, reproduced on cairn-enc v0.6.324:

  • The search-scope label (#searchscope-lbl, plus the menu entries #scope-folder and #scope-all) is filled by JavaScript in scopeSync() (web/static/index.html ~9831). It has no data-i18n attribute.
  • The language-change handler ($("pref-lang").onchange, ~12275) calls translateStatic(), which only refreshes [data-i18n] and [data-i18n-ph] elements, and re-renders the current main view. It never calls scopeSync().
  • Result: after switching the language in Preferences, the search placeholder changes ("Dateien und Ordner suchen") but "All files" keeps the old language until the next page reload. Checked: after a reload it correctly reads "Alle Dateien".
  • Fix: call scopeSync() on a language change.
    • Also audit the rest of the always-visible chrome (header, sidebar, user menu, bell) for other JS-filled labels with the same problem, and fix them the same way.
    • Add a web test: change the language, and every chrome label is in the new language without a reload.

Correction to item 16 (German Users & access labels): it was an artifact of how I switched the language during the 2026-10-09 test (in memory, without re-rendering the page). After a real reload, Users & access is fully translated in DE, FR and IT. Remaining nits from that check:

  • FR: the tab "Sieges" is missing its accent, so it should be "Sièges".
  • DE: "Identity Provider" and "Identitätsanbieter" are mixed on the same page. Pick one (the rest of the UI uses "Identitätsanbieter").
  • IT: "concede o revoca" and "concede o rimuove" appear in two sentences on the same page, and the "you" badge uses the informal "TU" while the new copy uses the formal Lei. Align both.
## Added (Nikola, 2026-10-10): "All files" in the search bar stays English **Root cause, reproduced on cairn-enc v0.6.324:** - The search-scope label (`#searchscope-lbl`, plus the menu entries `#scope-folder` and `#scope-all`) is filled by JavaScript in `scopeSync()` (web/static/index.html ~9831). It has no `data-i18n` attribute. - The language-change handler (`$("pref-lang").onchange`, ~12275) calls `translateStatic()`, which only refreshes `[data-i18n]` and `[data-i18n-ph]` elements, and re-renders the current main view. It never calls `scopeSync()`. - Result: after switching the language in Preferences, the search placeholder changes ("Dateien und Ordner suchen") but "All files" keeps the old language until the next page reload. Checked: after a reload it correctly reads "Alle Dateien". - **Fix:** call `scopeSync()` on a language change. - Also audit the rest of the always-visible chrome (header, sidebar, user menu, bell) for other JS-filled labels with the same problem, and fix them the same way. - Add a web test: change the language, and every chrome label is in the new language without a reload. **Correction to item 16 (German Users & access labels):** it was an artifact of how I switched the language during the 2026-10-09 test (in memory, without re-rendering the page). After a real reload, Users & access is fully translated in **DE, FR and IT**. Remaining nits from that check: - FR: the tab "Sieges" is missing its accent, so it should be "Sièges". - DE: "Identity Provider" and "Identitätsanbieter" are mixed on the same page. Pick one (the rest of the UI uses "Identitätsanbieter"). - IT: "concede o revoca" and "concede o rimuove" appear in two sentences on the same page, and the "you" badge uses the informal "TU" while the new copy uses the formal Lei. Align both.
Author
Owner

Research on the five proposed won't-fix items (2026-10-10, requested by Nikola)

(2) "Must change password" (AD pwdLastSet = 0) and lagging domain controllers

  • Facts.
    • An LDAP bind with the right password while pwdLastSet = 0 fails with error 49 / data 773 ("user must reset password"). The 773 only comes back for valid credentials, so no new LDAP session can start while the flag is set (Dell KB, Lenovo KB, Microsoft: must change at next logon).
    • Password changes go urgently to the PDC emulator, and other DCs chain recent password checks to it. pwdLastSet itself replicates normally. Under AvoidPdcOnWan, or when the PDC can't be reached, the change only travels by normal replication, with no error logged (Microsoft: password change processing).
  • What that means for Cairn. The earlier "harmless" claim holds only while the flag is still set everywhere. In the real sequence it can bite:
    1. The admin sets "must change", and Cairn cuts the sessions (correct).
    2. The user changes the password, and the bind succeeds via PDC chaining.
    3. Cairn's next re-check reads a lagging DC that still says 0.
    4. Today's transition rule can then sign the freshly signed-in user out once more. With DNS round robin across DCs, Cairn hits different DCs on different checks.
  • What others do.
    • Entra Connect doesn't even sync the flag by default. ForcePasswordChangeOnLogOn only syncs future events, and the troubleshooting docs are full of lag and lockout cases (Microsoft: password sync troubleshooting, Specops).
    • Lockouts after password changes are commonly caused by devices still holding the old password (UWindsor KB).
  • Beyond the usual practice. pwdLastSet = 0 carries no time, but AD's replication metadata does. msDS-ReplAttributeMetaData (a constructed attribute, readable with a base-scope read on the user's DN) gives the originating change time of pwdLastSet, and that time is identical on every DC. With it, a 0 gets a real, replication-consistent timestamp, the existing monotonic rule handles lag fully, and this residual disappears. Cost: one extra base read per re-check for users whose marker is 0. Fallback when the metadata can't be read: today's transition rule.
  • Recommendation: fix it, using the metadata timestamp for 0. Also document in deployment.md: "point ldap.url at one DC or the PDC emulator if possible". Not a won't-fix.

(6) Notice store keeps an unsaved notice in memory when saving fails

  • What broke elsewhere.

    • In Next.js turbo-persistence, memory was updated before the disk write. A disk error left the two out of sync, and the rollback then deleted good data. The fix: keep memory and disk consistent at all times (commit).
    • In NetworkManager, a rollback wrongly persisted in-memory state (MR 1920).
    • TanStack Query's docs show that a whole-snapshot rollback can wipe other concurrent changes. Roll back only your own change, or refetch from the source of truth (guide).
  • What stuck: stage and swap. Build the new state as a copy, persist it, and swap it in only on success. There is no window where memory and disk disagree, and no rollback code that can destroy anything.

  • Cairn today: every notify.Store setter (MarkAll, ClearAll, SetMuted, AddNotice) mutates first and then saves. The effects on a failed save:

    • a "read" mark comes back after a restart;
    • a mute is lost after a restart;
    • a notice is shown and then vanishes after a restart.

    These are low impact, but the fix is small and uniform.

  • Recommendation: fix all four setters together with copy-persist-swap under the same lock. That keeps them consistent, which was the reason given for not fixing just one.

(10) Revoke toast count vs the count on the button

  • What others hit.
    • In Django's admin, the bulk-delete confirmation count comes from an earlier query, so the success message can be "evidently fiction". The maintainers noted there's no universally right meaning: "objects that are gone" or "objects this action deleted" (ticket 34802).
    • MongoDB Compass showed "0 documents" after deleting 100,000 (COMPASS-8603).
    • LastPass users lost trust when the delete prompt showed changing counts (forum).
  • Cairn's semantics: "revoke all of this user's app passwords" deletes everything live at that moment. That includes one created after the dialog opened, which is the right call for a compromise, because an attacker can't slip a new one past the admin.
  • Recommendation: keep revoke-all, but say it when the numbers differ: "3 app passwords revoked, including 1 created after you opened this dialog". The server returns revoked and the names, and the UI compares them with the list it showed. This is cheap, and it closes the trust gap the sources describe.

(11) The long-dash test's exception for ". _ - @"

  • Other linters: Vale and LaTeX/React linters either scope dash rules to word-internal dashes, skip code-like spans, or keep explicit exception lists (badness dash rule, react-doctor, Vale at PostHog, Vale config disabling a dash rule). A narrow exact exception is in line with that.
  • A better fix than any exception: the copy itself. ". _ - @" is a row of bare symbols. Screen readers read it unpredictably, and some skip punctuation entirely. That's an accessibility issue (#638), not just a lint one.
  • Recommendation: reword it as "letters a-z, digits, and the characters dot, underscore, hyphen and @", in all four languages, and delete the exception from the test.

(12) Badge order next to a name

  • Guidance:
    • Design systems advise against stacking several status badges (UK ICDS badge).
    • Shopify lays multiple badges out in one row with one-word labels (Polaris badge).
    • Zendesk separates role or identity labels from status markers (Gather badges).
    • No source prescribes an exact order. The common thread is identity next to the name, status after it, and the same order everywhere.
  • Cairn today: name, username, YOU (identity), then CHECK NAME or DISABLED (status). CHECK NAME applies only to local and config rows, and DISABLED only to directory rows, so at most one status badge ever shows.
  • Recommendation: keep the order. Add a code comment stating the rule (identity, then status; one status at most), so later badges follow it.

Summary: items 2, 6, 10 and 11 move from won't-fix into the batch. Item 12 stays as is, with a comment in the code.

## Research on the five proposed won't-fix items (2026-10-10, requested by Nikola) ### (2) "Must change password" (AD `pwdLastSet = 0`) and lagging domain controllers - **Facts.** - An LDAP bind with the right password while `pwdLastSet = 0` fails with error 49 / `data 773` ("user must reset password"). The 773 only comes back for *valid* credentials, so no new LDAP session can start while the flag is set ([Dell KB](https://www.dell.com/support/kbdoc/en-ph/000165267/networker-ad-ldap-integration-failing-with-common-ldap-error-codes-ldap-error-code-49), [Lenovo KB](https://helpdesk.lenovosoftware.com/portal/en/kb/articles/common-microsoft-ad-ldap-bind-errors-22-8-2017), [Microsoft: must change at next logon](https://learn.microsoft.com/en-us/windows/win32/adsi/user-must-change-password-at-next-logon)). - Password *changes* go urgently to the PDC emulator, and other DCs chain recent password checks to it. `pwdLastSet` itself replicates normally. Under `AvoidPdcOnWan`, or when the PDC can't be reached, the change only travels by normal replication, with no error logged ([Microsoft: password change processing](https://learn.microsoft.com/en-us/troubleshoot/windows-server/identity/password-change-processing-conflict-resolution-function)). - **What that means for Cairn.** The earlier "harmless" claim holds only while the flag is still set everywhere. In the real sequence it can bite: 1. The admin sets "must change", and Cairn cuts the sessions (correct). 2. The user changes the password, and the bind succeeds via PDC chaining. 3. Cairn's next re-check reads a lagging DC that still says `0`. 4. Today's transition rule can then sign the freshly signed-in user out **once more**. With DNS round robin across DCs, Cairn hits different DCs on different checks. - **What others do.** - Entra Connect doesn't even sync the flag by default. `ForcePasswordChangeOnLogOn` only syncs future events, and the troubleshooting docs are full of lag and lockout cases ([Microsoft: password sync troubleshooting](https://learn.microsoft.com/en-us/troubleshoot/entra/entra-id/user-prov-sync/troubleshoot-pwd-sync), [Specops](https://specopssoft.com/blog/user-must-change-password-at-next-login-azure-ad/)). - Lockouts after password changes are commonly caused by devices still holding the old password ([UWindsor KB](https://uwindsor.teamdynamix.com/TDClient/1975/Portal/KB/ArticleDet?ID=151727)). - **Beyond the usual practice.** `pwdLastSet = 0` carries no time, but AD's replication metadata does. `msDS-ReplAttributeMetaData` (a constructed attribute, readable with a base-scope read on the user's DN) gives the *originating* change time of `pwdLastSet`, and that time is identical on every DC. With it, a `0` gets a real, replication-consistent timestamp, the existing monotonic rule handles lag fully, and this residual disappears. Cost: one extra base read per re-check for users whose marker is `0`. Fallback when the metadata can't be read: today's transition rule. - **Recommendation:** fix it, using the metadata timestamp for `0`. Also document in deployment.md: "point `ldap.url` at one DC or the PDC emulator if possible". Not a won't-fix. ### (6) Notice store keeps an unsaved notice in memory when saving fails - **What broke elsewhere.** - In Next.js turbo-persistence, memory was updated before the disk write. A disk error left the two out of sync, and the rollback then deleted good data. The fix: keep memory and disk consistent at all times ([commit](https://app.semanticdiff.com/gh/vercel/next.js/commit/fff4a4d9bfe4d52ee03ce73f7cd36b9779278c66)). - In NetworkManager, a rollback wrongly persisted in-memory state ([MR 1920](https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/1920)). - TanStack Query's docs show that a whole-snapshot rollback can wipe *other* concurrent changes. Roll back only your own change, or refetch from the source of truth ([guide](https://tanstack.com/query/v5/docs/framework/react/guides/optimistic-updates)). - **What stuck:** stage and swap. Build the new state as a copy, persist it, and swap it in only on success. There is no window where memory and disk disagree, and no rollback code that can destroy anything. - **Cairn today:** every `notify.Store` setter (`MarkAll`, `ClearAll`, `SetMuted`, `AddNotice`) mutates first and then saves. The effects on a failed save: - a "read" mark comes back after a restart; - a mute is lost after a restart; - a notice is shown and then vanishes after a restart. These are low impact, but the fix is small and uniform. - **Recommendation:** fix all four setters together with copy-persist-swap under the same lock. That keeps them consistent, which was the reason given for not fixing just one. ### (10) Revoke toast count vs the count on the button - **What others hit.** - In Django's admin, the bulk-delete confirmation count comes from an earlier query, so the success message can be "evidently fiction". The maintainers noted there's no universally right meaning: "objects that are gone" or "objects this action deleted" ([ticket 34802](https://code.djangoproject.com/ticket/34802)). - MongoDB Compass showed "0 documents" after deleting 100,000 ([COMPASS-8603](https://jira.mongodb.org/browse/COMPASS-8603)). - LastPass users lost trust when the delete prompt showed changing counts ([forum](https://community.lastpass.com/discussion/3267/incorrect-count-shown-when-deleting-folders-from-vault)). - **Cairn's semantics:** "revoke all of this user's app passwords" deletes *everything live at that moment*. That includes one created after the dialog opened, which is the right call for a compromise, because an attacker can't slip a new one past the admin. - **Recommendation:** keep revoke-all, but say it when the numbers differ: "3 app passwords revoked, including 1 created after you opened this dialog". The server returns `revoked` and the names, and the UI compares them with the list it showed. This is cheap, and it closes the trust gap the sources describe. ### (11) The long-dash test's exception for ". _ - @" - **Other linters:** Vale and LaTeX/React linters either scope dash rules to word-internal dashes, skip code-like spans, or keep explicit exception lists ([badness dash rule](https://docs.rs/crate/badness/0.13.0/source/src/linter/rules/dash_length.rs), [react-doctor](https://www.react.doctor/docs/rules/react-doctor/design-no-em-dash-in-jsx-text), [Vale at PostHog](https://posthog.com/handbook/docs-and-wizard/vale), [Vale config disabling a dash rule](https://git.berlin.ccc.de/vinzenz/hyperhive/src/branch/main/.vale.ini)). A narrow exact exception is in line with that. - **A better fix than any exception:** the copy itself. ". _ - @" is a row of bare symbols. Screen readers read it unpredictably, and some skip punctuation entirely. That's an accessibility issue (#638), not just a lint one. - **Recommendation:** reword it as "letters a-z, digits, and the characters dot, underscore, hyphen and @", in all four languages, and delete the exception from the test. ### (12) Badge order next to a name - **Guidance:** - Design systems advise against stacking several status badges ([UK ICDS badge](https://design.sis.gov.uk/components/feedback-progress/badge)). - Shopify lays multiple badges out in one row with one-word labels ([Polaris badge](https://shopify.dev/docs/api/customer-account-ui-extensions/latest/web-components/feedback-and-status-indicators/badge)). - Zendesk separates role or identity labels from status markers ([Gather badges](https://support.zendesk.com/hc/en-us/articles/4408834628506-About-Gather-badges)). - No source prescribes an exact order. The common thread is identity next to the name, status after it, and the same order everywhere. - **Cairn today:** name, username, YOU (identity), then CHECK NAME or DISABLED (status). CHECK NAME applies only to local and config rows, and DISABLED only to directory rows, so at most one status badge ever shows. - **Recommendation:** keep the order. Add a code comment stating the rule (identity, then status; one status at most), so later badges follow it. **Summary:** items 2, 6, 10 and 11 move from won't-fix into the batch. Item 12 stays as is, with a comment in the code.
Cordy closed this issue 2026-10-10 12:22:31 +00:00
Author
Owner

The deferred follow-ups shipped in v0.6.326 (#871):

  • A second timed AD "must change" is now audited. There is a test for untimed, then timed, then a newer timed value.
  • The search view re-renders on a language change and cancels the pending timer first.
The deferred follow-ups shipped in v0.6.326 (#871): - A second timed AD "must change" is now audited. There is a test for untimed, then timed, then a newer timed value. - The search view re-renders on a language change and cancels the pending timer first.
Sign in to join this conversation.
No labels
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: Cordy/Cairn#867
No description provided.