PR #41736 - RocketChat/Rocket.Chat

Proposed changes (including videos or screenshots)

Continues the migration from the legacy i18next-sprintf-postprocessor (%s) to i18next named interpolation ({{param}}).

en contained 75 keys with %s. This PR resolves 50 of them: 29 deleted, 19 converted, 2 that were rendering a raw placeholder to users.

No changeset: everything here either has no user-visible effect or fixes a string that was already displaying a raw %s, and the affected keys are internal command responses.

Deleted — 29 keys with no reference anywhere

Verified by searching each literal key across apps/, packages/ and ee/, all file types, and by checking for dynamically constructed keys. Removed from all 58 locale files rather than en alone, because wipe-extra-keys fails CI on any non-en key that en lacks.

Created_at_s_by_s, Created_at_s_by_s_triggered_by_s, Duplicate_archived_private_group_name, Duplicate_channel_name, Duplicate_private_group_name, Invalid_Export_File, Invalid_notification_setting_s, Invalid_room_name, Invalid_setting_s, IssueLinks_LinkTemplate_Description, Login_with, N_new_messages, Post_to_s_as_s, Purchase_for_price, S_new_messages, S_new_messages_since_s, Showing_archived_results, Showing_results, Snippet_Added, Use_service_avatar, User_has_been_muted_in_s, WebRTC_direct_audio_call_from_%s, WebRTC_direct_video_call_from_%s, WebRTC_group_audio_call_from_%s, WebRTC_group_video_call_from_%s, WebRTC_monitor_call_from_%s, conversation_with_s, n_messages, since_creation

The five WebRTC_* keys were orphaned by 31a654d4be (chore!: Remove MeetPage and webRTC (#36671)), which removed the feature and its call sites but left the strings behind.

Also confirmed nothing outside this repo reads them. These keys are served over HTTP at GET /i18n/<lng>.json, but the only consumers of that route are the web client itself (TranslationProvider, loadPath: 'i18n/{{lng}}.json') and one e2e test. The mobile app bundles its own separate 1,075-key set and references neither the route nor any of these keys.

Modified — 21 keys converted to named interpolation

Keys	Params
`Channel_Archived`, `Channel_Unarchived`, `Channel_already_Unarchived`, `Channel_created`, `Channel_doesnt_exist`, `Duplicate_archived_channel_name`	`{{channelName}}`
`Edit_previous_message`, `Mark_all_as_read`, `Move_beginning_message`, `Move_end_message`, `New_line_message_compose_input`, `Open_channel_user_search`	`{{shortcut}}`
`SlackBridge_error`	`{{channelName}}`, `{{errorMessage}}`
`SlackBridge_finish`	`{{channelName}}`
`SlackBridge_start`	`{{username}}`, `{{channelName}}`
`User_doesnt_exist`, `Username_doesnt_exist`, `Username_is_already_in_here`, `Username_invalid`	`{{username}}`
`error-logged-user-not-in-room`	`{{roomName}}`
`Welcome`	`{{user}}`

Names follow the conventions already established by #41667. Welcome needed no code change at all, because both call sites were already passing named { user }.

The two SlackBridge_error call sites also needed interpolation: { escapeValue: false }. errorMessage is arbitrary text and the server i18n instance escapes by default (only the client sets escapeValue: false), so without the opt-out an error containing < or & would render as an HTML entity where sprintf never escaped anything.

Left as-is — 25 keys, with reasons

Count	Keys	Reason
19	`Channel_already_exist`, `Custom_oauth_helper`, `Do_you_want_to_change_to_s_question`, `Mail_Message_Invalid_emails`, `Mail_Messages_Subject`, `Max_length_is`, `Min_length_is`, `Prune_Warning_after`, `Prune_Warning_all`, `Prune_Warning_before`, `Prune_Warning_between`, `Showing_results_of`, `The_setting_s_is_configured_to_s_and_you_are_accessing_from_s`, `The_user_s_will_be_removed_from_role_s`, `The_user_will_be_removed_from_s`, `The_user_wont_be_able_to_type_in_s`, `User_has_been_removed_from_s`, `Users_Table_Generic_No_users`, `if_they_are_from`	Already converted in open PR #41667. Left alone to avoid a conflict.
4	`New_version_available_(s)`, `Sending_your_mail_to_s`, `The_server_will_restart_in_s_seconds`, `Your_push_was_sent_to_s_devices`	Blocked on a wire-format change, see below.
1	`Alias_Format_Description`	The `%s` is literal documentation text shown to the admin, not a placeholder. Must never be converted.
1	`MongoDB_version_s_is_deprecated_please_upgrade_your_installation`	Its only reference is commented out in `serverRunning.ts`. Held back from the delete batch pending a decision on that dead code.

The four blocked keys travel as a translation key plus a positional array, and are translated client-side rather than server-side:

{ message, params } returned from Meteor methods (restartServer, sendSMTPTestEmail, push_test) and from REST /v1/push.test, consumed by ActionInputBase;
textArguments on banner documents persisted in users.banners, consumed by useUserBanners.

Named params need an object instead of an array, which means changing a public REST response shape and having a story for banner rows already in the database. That deserves its own ticket. It is also why i18next-sprintf-postprocessor cannot be dropped yet: even at zero %s keys, those readers still call it.

Bugs found and fixed along the way

Three slash commands crashed rather than rendering wrong. A malformed placeholder throws SyntaxError: [sprintf] unexpected placeholder inside i18n.t(). On develop, /archiveroom and /unarchiveroom throw on a Romanian-language server, /unarchiveroom throws on Serbian, and /slackbridge-import throws on Belarusian, instead of replying.
Welcome displayed a literal %s. Both call sites passed named { user } while en still held %s, so the welcome system message showed %s instead of the username.
Username_invalid displayed a literal %s. The call site passed no argument at all. Korean was worse: &s, an ampersand typo that no %s-based tooling could ever match.
/help showed no shortcut at all in 24 languages, which had dropped the placeholder from Mark_all_as_read entirely.
i18n.t(key, options, user.language) silently ignored the language. i18next's TFunction has no third-argument overload, so all four /slackbridge-import messages always rendered in the server default language regardless of the user's setting. The file is @ts-nocheck, so TypeScript never flagged it.
The i18n checker under-reported by 8 keys. find-sprintf-params declared /%s/g outside its per-key loop, so lastIndex persisted across .exec() calls on different strings and keys were skipped after any match. It reported 67 against a ground truth of 75, hiding Move_end_message, SlackBridge_finish and Username_is_already_in_here among others, all of which needed conversion. The regex is now gone in favour of translation.includes('%s'), which has no state to leak.
Unsubstituted placeholders across many commands. zh-HK used fullwidth ％s on 18 of these keys, with further mangled variants in ro, sr, ku, mn, be-BY and ko.

Issue(s)

CORE-2506

Steps to test or reproduce

Set Administration → Settings → General → Language to Romanian, then run /archiveroom in any channel. On develop this throws instead of replying. With this PR it posts Canal cu numele #<channel> a fost arhivată cu succes. Same for /unarchiveroom, and for /unarchiveroom in Serbian.
Set the language to Chinese (Hong Kong) and run /help. On develop every line shows a literal ％s; with this PR each shows its real shortcut.
Set the language to German (or any of the 24 affected) and run /help. On develop the "mark all as read" line has no shortcut at all; with this PR it shows Shift (or Ctrl) + ESC.
Run /invite @nonexistentuser, then /kick @nonexistentuser, /mute @nonexistentuser, /ban @nonexistentuser. Each names the user instead of printing %s.
Trigger the welcome system message in a room. It shows the username instead of %s.
Reach the username registration screen and submit an invalid username. The error names the username instead of printing %s.
Expected result: no screen displays a raw %s, and no command errors out.

Regression check: yarn workspace @rocket.chat/i18n lint passes. Its missing-placeholders task enforces that every locale's placeholders match en exactly, so any locale the migration missed fails CI rather than shipping silently.

Further comments

The diff spans 40 languages, so every conversion was verified mechanically rather than by review. A harness rendered each key's old value through the real i18next-sprintf-postprocessor and its new value through real i18next named interpolation, with identical arguments. A behaviour-preserving conversion must produce byte-identical output.

	Rows
Total	1175
Identical	1125
Different	45
Old value crashed sprintf	5
Unresolved `{{…}}` or `%s` left in output	0

The 50 non-identical rows were then asserted programmatically to be exactly the set of locales fixed by hand: no unexpected differences, and nothing claimed as fixed that was actually untouched. All 45 differing rows are strict improvements over a previously visible ％s, % с or &s.

Where a locale's placeholder was mangled, the rule was to relocate the placeholder and keep every existing word, never to author a translation. One deliberate exception: 24 locales had dropped the Mark_all_as_read shortcut prefix entirely, so it was restored using the exact structure of the four locales (en, hi-IN, nb, nn) that still had it correct.

One thing left for Lingohub: ku on SlackBridge_error now reads …çewtiyek çêbû general: connection refused and grammatically wants the preposition li before the channel name. Left alone rather than invent a Kurdish word.

ESLint on all 16 touched source files reports 5 warnings, every one byte-identical to the pre-change baseline. No new lint debt.