Repository navigation
v0.0.2-alpha.77
Pre-releaseReleased 21 packages at 0.0.2-alpha.77 in lockstep. Every package below ships at this version.
What's changed
Patch Changes
-
#4261
ace6e4cThanks @nextly-agent! - A one-to-one relationship made in the Schema Builder keeps its column unique again, on PostgreSQL, MySQL and SQLite: the migration a collection's or a Single's Builder save writes now createsuq_<table>_<column>for it, as the Builder's own table creator always made the columnUNIQUE. Since collections, and then Singles, began saving through a written migration, such a column was given a plain index and took two entries naming the same target. A one-to-one saved that way gets its unique index on the next save of its collection or Single, or fromnextly migrate:create; where its column already holds duplicates, that migration is refused until they are resolved. -
#4074
6c032d3Thanks @aqib-rx! - A busy table no longer waits without end. Whennextly migrateornextly migrate:downruns a migration file in one transaction on PostgreSQL or SQLite and a statement waits for a table another database session is using, each try now ends after about 5 seconds of waiting for locks, the run pauses (1, 2, 4, 8, then 15 seconds) and tries again, and after about 90 seconds in all it stops withTABLE_BUSY(exit 1): "table busy, nothing changed", the table, the tries made, the session holding it (never its query text) and the next step. A try that ran a statement which may not be sent again (setval, an index builtCONCURRENTLY, a trigger or a function of the site's own) stops after one try.--lock-timeoutand--lock-timeout-total, orNEXTLY_LOCK_TIMEOUTandNEXTLY_LOCK_TIMEOUT_TOTAL, change the bounds (whole seconds: one try 1 to 3600, the total from one try to 86400). A one try set longer than the default total, with no total given, raises the total to match; an explicit total below the one try is refused. These bound the wait for a busy table, not the wait for the schema lock.Plugin migration modules,
migrate:fresh,prune,db:sync, the development push and start, and theupgradesteps now wait for a table another session holds in short tries, asnextly migratedoes, instead of waiting without a bound. Plugin modules run in tries throughnextly migrate, the boot step,plugins install,plugins uninstallandmigrate:down --plugin.migrate:fresh's drops,prune's statements, the development start's table creation and theupgradesteps run one statement per try on one connection. The schema pipeline's apply runs in tries fordb:sync, the development push and the Builder's apply, asking its questions once. A run that stops on a busy table names the table and what holds it, and changes nothing more.A Schema Builder save of a collection now waits for a table another session is using, as
nextly migratedoes: in short tries, within the boundsNEXTLY_LOCK_TIMEOUTandNEXTLY_LOCK_TIMEOUT_TOTALset, then it stops as "table busy, nothing changed", naming the table and what holds it. A save whose tries ran out on MySQL now withdraws its files and puts its definition back, as a save that changed nothing does. -
#4278
20df4edThanks @nextly-agent! - A command that opens its database no longer setsDB_DIALECTandDATABASE_URLin the process environment while it connects. The CLI now hands the dialect and URL straight to the adapter factory, so an open that outlives the command that started it (asnextly migrate:createallows when the database it counts deletions in is slow to answer) can no longer write its own target back over a later setting, and a variable the process never set stays unset. The adapter factory reads its pool and timeout settings (DB_POOL_MAX,DB_POOL_MIN,DB_POOL_IDLE_TIMEOUT,DB_QUERY_TIMEOUT) on their own, so a command that names its database opens it even where the environment names no database at all. -
#3950
0ed6383Thanks @faisal-rx! - On MySQL,db:syncand the development reload again narrow a code-first field group's legacy TEXT column to the varchar(255) its schema declares, when every value fits. A field group's text column is described by the rule of whoever owns it: unbounded for a Schema Builder field group, 255 wide for a code-first one. -
#3934
513486eThanks @faisal-rx! - A field-group or Single create no longer builds over a table that already stands under its name with nothing registered describing it. An empty one is replaced, so the table ends up with exactly the requested columns. One holding rows is refused, naming each table and how many rows it holds, and is left as it was. A create is also recorded as applied only once its table holds every requested column with the requested type; otherwise it is recorded as failed. A create of a field group whose name is already taken is refused naming that name. Collection creates keep their current behaviour for now. -
#4282
67c3f5eThanks @nextly-agent! - A plugin's deletion report made in its own transaction (ctx.principals.reportDeleted(person, { tx })) now holds the deleted-people lock from the moment it is called. An email sent alongside it in the same transaction waits for the report before recording its delivery, however long the report takes to begin. Before, a delivery record that arrived while the report was still checking its tables was written beside it on the same connection; on PostgreSQL, a failed insert there aborted the transaction under the report's next statement. The report's table check now runs on the transaction's own connection. -
#4200
aa1322fThanks @nextly-agent! - A field rewrite (rewriteField,@experimental) now reaches a Single as it reaches a collection entry, through the same request shape on both surfaces:{ single, field, expectedVersion, set, reason }beside{ collection, id, ... }, onctx.services.collections.rewriteFieldandnextly.rewriteField.- The live copy and every draft, without publishing. The Single's live copy and each pending draft that holds the field are written in one transaction; each draft stays a draft and the status doesn't change. A shared field is written to every language's draft, and a language draft whose author had edited it publishes its own rewritten value later. A translated field (
locale) is rewritten in that language only. - Only from the version the caller read. A Single saved since is refused with
STALE_VERSION, with nothing changed. - History and events. One
field-rewrittenevent names the field, each copy and the reason, and the plugin that made it. A change to the live copy recordssingle.updatedand revalidates the Single; a change to drafts alone records neither. No save hooks run. - Refusals and access. The same fields and values are refused as for an entry. Made for a person, it needs their permission to update the Single and, where it has a publish lifecycle, to publish it.
FieldRewriteRequestandRewriteFieldArgskeep their entry shape. A Single's request isSingleFieldRewriteRequest(RewriteSingleFieldArgson the Direct API), andDocumentFieldRewriteRequestnames either, which the plugin service'srewriteFieldtakes.For an entry as for a Single, a group or repeater that holds a password or a field group is now refused. A read gives a password-holding field without the password, so writing it back would drop the stored hash, and a draft would keep a given password as written. A field group held inline has its own write rules and type, which a value swap doesn't judge.
- The live copy and every draft, without publishing. The Single's live copy and each pending draft that holds the field are written in one transaction; each draft stays a draft and the status doesn't change. A shared field is written to every language's draft, and a language draft whose author had edited it publishes its own rewritten value later. A translated field (
-
#4176
688de00Thanks @nextly-agent! - A field rewrite (rewriteField,@experimental) of a live shared field now withdraws each language's workflow approval the change no longer matches, as an ordinary save of that field does. A field inside an unnamed layout group is now decoded and encoded by its stored type, so a JSON value reachesset.fromparsed and is written back encoded on every dialect. -
#4042
5584ce5Thanks @zeshan-rx! - A held draft is kept by language wherever a document keeps values per language.- Under per-language storage (
experimental.perLanguageStorage), a Single that is not translated but embeds a translated component holds drafts: each language's edit to the component is held in that language's draft and published to that language's row. Its publish of every language writes each translatable value to that language's row. - A version of a document that stores a component, at the top level or inside a group or repeater, is recorded under the language the component was read in, even when the component is empty, on every path that records one. A held save of such a document is now found as that language's draft (
?baseline=draftno longer answersNO_BASELINEfor it). - A second held edit of a translated field is kept: it is recorded as a version and is what the draft holds. It was matched against the draft it replaced, found unchanged and dropped.
Behaviour change: without per-language storage, a collection that is not translated but embeds a translated component no longer holds drafts, as a Single like it already did not. Its one draft had no language, so held edits in two languages shared it, and a publish of every language wrote one language's component values into another's row. A status-less save of such a collection now saves live. A draft such a collection already holds stays shown and can be discarded, but it is never published, since it names no language: while it is there, a save, a publish of one language and a publish of every language are refused (409). Copy what you need from it, then discard it.
- Under per-language storage (
-
#4168
5bb3231Thanks @nextly-agent! - A migration file, plugin module or down migration that commits part-way through is refused before anything runs, naming the statement. Its ownBEGIN,START TRANSACTION,COMMITorENDused to be dropped silently, so a file written to commit its first half on its own ran as one transaction instead. A file wrapped whole in oneBEGIN...COMMITstill runs as before: the runner's transaction already wraps it.To upgrade: a pending file with a
COMMITorBEGINpart-way through is refused until it is split there into two files, or marked with-- nextly:no-transactionas its first line (and itsBEGINandCOMMITremoved), which runs it statement by statement. -
#4292
030355eThanks @nextly-agent! - A new entry's own sections no longer ask to discard it. On a form not yet saved, opening one of its notifications (or any part of a new entry its editor addresses with?section=) asked whether to discard the unsaved changes, and discarding lost the whole new form. The create page is now one document to the unsaved-changes guard, as a saved entry's edit address is: moving between its parts keeps the unsaved work, while another collection's new entry, another language or a view address still asks. -
#4303
a76316fThanks @nextly-agent! - A form notification's page no longer says "Recent sends could not be read." before anything has been sent. The recent sends read asked for the submissions of no sends with an empty list, which the query layer refuses as a filter that can match nothing; it now answers an empty list without asking. -
#4221
8c20e83Thanks @nextly-agent! -ctx.publicAddresses.fingerprint()now also changes after a start that ran with another set of plugins taking part in keeping public addresses unique (@experimental). A plugin disabled, removed from the config, uninstalled, or switched off by a config reload in development reaches no save while it is not running, and when it came back its stored fingerprint still matched, so it skipped its catch-up and its record stayed blind to those saves. Nextly now keeps the set each start ran with, and a count that goes up when it changes, in onenextly_metarow; the fingerprint includes the count. A start reads the row with the meta rows it already reads, and writes it only when the set changed, so a start with the same plugins as the last costs nothing more. On upgrading, the first start of a site whose plugins take part records them, and their next catch-up runs in full once. -
#4259
60bb670Thanks @mobeenabdullah! - A plugin can now switch the fields itsextendadded off on some types, without a schema change (experimental). It contributes a service namedEXTENSION_SWITCH_SERVICE(from@nextlyhq/plugin-sdkandnextly) whoseswitchedOff(field)lists the types the field is off on now. On those types the admin draws the field nowhere in its editors (the entry and Single edit pages, their quick-edit forms and past versions), and offers none of the plugin'sextendedWithcontributions there. The column and every stored value stay, the API keeps reading and writing them, and switching the type on again brings everything back. The admin reads the switch each time it loads its workspace, so a change applies without a redeploy. A switch that fails switches nothing off and is logged. -
#4253
c4a6a98Thanks @mobeenabdullah! - A plugin that adds a field to every content type can now find out which types the field reached, and Page Builder's pages are among them. All three additions are experimental.typesExtendedBy(plugin, field), from@nextlyhq/plugin-sdkandnextly, lists the collections and Singles where a plugin'sextendadded a field at the latest start. Schema Builder types are included. A type a scope left alone because it already had a field by that name is not listed, and neither is one the scope excepts.- An admin contribution's
extendedWith: ["field"]offers it only on those types. A tab that edits the field follows it to Schema Builder types, and is never drawn on a type that kept a field of its own by that name. - A plugin's collection marked
siteContent: trueholds the site's own content, so "every content type" scopes reach it as they reach the site's own collections. Page Builder marks itspagescollection. Its other collections, and every other plugin's, stay out of scopes as before. Nothing changes until a plugin uses a scope: a site whose plugins name their collections, as every first-party plugin does today, sees no change.
-
#4215
13f61beThanks @nextly-agent! - A plugin whose schema prefix differs from the one its recorded tables were created under is refused at boot, bynextly migrateand bynextly plugins install, naming both prefixes and the tables, before any push or migration creates its tables again under the new names. -
#4167
17d503dThanks @nextly-agent! - A plugin, or server code through the Direct API, can now change one field of an entry on its live copy and on every pending draft that holds it, in one step, without publishing anything:ctx.services.collections.rewriteField(request, opts?)andnextly.rewriteField(args)(both@experimental).- Drafts stay drafts. The live copy and each draft holding the field are written in one transaction, and the entry's status doesn't change. A shared field is written to every language's draft; a translated one (
locale) only in that language. A draft whose author had edited the field publishes its own rewritten value later, rather than being refused for disagreeing with the live copy. - Only from the version the caller read.
expectedVersionis required, and an entry saved since is refused withSTALE_VERSION, with nothing changed. There is no overwrite. - A value, or a function of each copy's own.
set: { value }gives every copy the same value;set: { from(current, copy) }computes each copy's value from what that copy holds, for jobs such as turning a component's copies into plain blocks. - History says who and why. One
field-rewrittenevent names the field, each copy changed and the requiredreason(kept as the event's note). A rewrite a plugin makes on its own behalf names the plugin, from its own identity; history has a newpluginactor kind. It is never recorded as an overwrite. - Refusals. The id, status, slug and timestamps, a unique field, a field stored in tables of its own (a field group, a many-to-many relationship), a password and a virtual field are refused, as is a value the field's own checks refuse. Made for a person, it needs their permission to update the entry and, where the collection has a publish lifecycle, to publish it, and every field rule of their own update.
- No save hooks. A rewrite runs none of the collection's or fields' save hooks and no in-transaction handler. The live change records the
entry.updatedevent, announces the update to subscribers once it commits, and revalidates the entry's pages.
A collection entry only, for now: Singles are not rewritten yet.
- Drafts stay drafts. The live copy and each draft holding the field are written in one transaction, and the entry's status doesn't change. A shared field is written to every language's draft; a translated one (
-
#4308
9512680Thanks @nextly-agent! - A production start reads the stored schema version once again. It reads the version just before it loads its definitions, so that the first schema follow after the start loads nothing again; it then compared the database's version with the build's by reading the same row a second time. The comparison now uses that reading, or the one after boot migrations where they stored a new version, and reads the row itself only where this process marked a change since.A reload of the definitions keeps the fields the configuration declares, as the start does. Where an entity's registry row predates its configuration, the start registers the configuration's definition over the row's, but a reload (the Direct API's first schema-version follow, the reload after boot migrations) loaded the row's narrower definition alone, so a declared field the table holds was no longer served, and one the database was not migrated for was no longer reported as not migrated. Every reload now registers the configuration's entities after the rows, through the start's own step, from the configuration as it now stands: a config reload under
next devmakes the definitions it applied the ones later reloads register. Where it applied an edit only in part, because another entity's change was held back or its registry write failed, the entities it applied take their new definitions and the held-back ones keep their previous ones. -
#4305
a158d5aThanks @nextly-agent! -nextly.findSingles()refuses anoffsetwithout alimitbefore it does anything else. Since the Direct API began following the stored schema version, a call it was going to refuse first read that version, and could reload every definition, before the refusal. It now reads nothing. -
#4170
f9ae21dThanks @nextly-agent! -nextly plugins installof a plugin that is already installed no longer records a second install or runs itsonInstallagain. It still applies any migration module the plugin's package added since, still adds any permission or role the package newly declares (with every preset role's grants and the release record, in one transaction, as a first install writes them), and exits 0. A deploy script that runsplugins install --allon every release now leaves the ledger as the first install wrote it. An install after an uninstall still runsonInstalland records the install. -
#4214
5473185Thanks @nextly-agent! - A plugin that takes part in keeping public addresses unique is now handedsaveKey(@experimental) besidedbandtransaction: one frozen object for every call core makes about one save or onectx.publicAddresses.move, itsholderOfcalls and therecordafter them, and a new one for the next, even on the same transaction. A participant keeps by it work the question needs only once. It differs from an in-transaction handler'stransactionKeyon purpose: a transaction can roll back to a savepoint between two saves, which on PostgreSQL lets go of row locks taken after it.The SEO plugin uses it to take each address lock row once per save rather than once per call: a save of a post at
/blog/xtakes 4 lock rows rather than 8 on PostgreSQL and MySQL, and a slug change with its automatic redirect 8 rather than 16. Each lock row taken is a round trip made while the save holds its row locks, so an ordinary save makes four fewer. SQLite takes none and is unchanged. -
#4264
abd5323Thanks @mobeenabdullah! - A Single's read now honoursselect, with the meaning it has on a collection read: the fields namedtrueare answered, with the id and the system timestamps, and a selection naming none answers the whole document. Where no read hook or field read rule needs the other fields, the read takes only the selected fields' columns, so a field whose columnnextly migratehas not created yet fails only a read that selects it. -
#4063
31af891Thanks @faisal-rx! - A Single's table no longer carries a field's default as a column default: the default is applied when the Single's document is written, as a collection's is, and as the schema a migration is written from describes the column. A required field added to a Single that already holds its row fills that row and leaves its column with no default. A Single's columns that already carry such a default keep it until it is converted. -
#4291
90203c0Thanks @nextly-agent! - A start no longer runs its translation-table warm-up on its own time. The warm-up resolves each translated entity's translation table in the background; it now begins once the start in flight has handed its instance over, as the other background tasks after a start do. Before, registration began it part way through the start, so its registry reads ran among the start's own statements, and one could still be running whengetNextly()resolved.createTestNextlyalso settles the warm-up before it hands its instance over, so a test meets none of those reads among its own. -
#4238
72bb0feThanks @mobeenabdullah! - A title or description typed in a page's SEO tab is now also what shared links show (Open Graph and X cards), ahead of the social text the page's route sets in its own code. Many routes copy their heading, such as "Blog", intoopenGraph.title, which hid the title typed for the page on Facebook, LinkedIn and X. The social card's order is now: the social text typed for the page, the title or description typed for it, the route's social text, the Page Builder design's, and else the page's title or description. Where nothing is typed for the page, the route's own social text is still the one shared. -
#4310
2fd380cThanks @nextly-agent! - A variable that no page uses can now be removed on a site whose usage count cannot become whole. When some pages are too large to read in full, or keep their blocks in a field nested under a named group, the removal reads just those pages for the variable's stored reference. If none of them uses it, the removal goes ahead. If one does, the removal is refused and that page is named. A page that cannot be read within the removal's time, including a read that stalls, is also refused and named. If the list of those pages does not finish in that time, the removal is refused too. Before, every variable removal on such a site was refused, whether the variable was used or not. -
#4285
072beaaThanks @nextly-agent! - A Design variable's row keeps the keyboard's focus when its usage count arrives. The row was drawn again around its count's tooltip once the count was read, so focus fell back to the page and pressing Enter on a row reached by Tab opened nothing. The row now stays the same element, with its tooltip closed until the count is known. -
#4155
30aaf09Thanks @faisal-rx! - The Schema Builder can ask, before creating a collection, Single or field group, what the create would answer for an address: one route per kind (POST /api/collections/schema/address-check,/api/singles/schema/address-check,/api/field-groups/schema/address-check), gated by that kind's create permission, which runs the create's own checks and counts the entities created in the editor but not saved yet. A Single or field-group create sent without an address now answers with the address rule's "required" refusal (400) instead of an internal error. -
#4248
99e2042Thanks @nextly-agent! - The admin's history names a plugin acting on its own behalf with the plugin icon, and draws the admin for a session that does not say whether its account holds a role; only a session that says it holds none sees the no-access screen. -
#4197
2cae823Thanks @nextly-agent! - The admin downloads about 150 KB (gzipped) less before its first screen draws: 331 KB of shared code is now 179 KB.@nextlyhq/adminnow exports only what an application mounts and reads the admin with:RootLayout,QueryProvider,ThemeProvider,ThemePaintScript,useTheme,BrandingProvider,useBranding,ErrorBoundary,QueryErrorBoundary,CommandPalette,registerComponent,registerComponents,registerComponentMap,useCan,Can, the data hooks and UI primitives its README lists,Toasterandtoast, and the types all of these are written in (Media,MediaParams,MediaListResponse,DashboardStats,RecentActivityResponse,ErrorBoundaryPropsand the rest), so code using a hook can name what it returns. The parts plugin pages are built from (the media library, the field-UI kit, the document surfaces, the data table, the widget kit, the plugin-page parts) now load with the pages that use them.Plugins change nothing: they import these parts from
@nextlyhq/plugin-sdk/admin, as before, and every name there is unchanged. A component path in a site's or plugin's config that names one of them on@nextlyhq/admin(@nextlyhq/admin#MediaLibrary) keeps working:nextly generate:typesnow imports it from where it lives. Code that imported any of them from@nextlyhq/admindirectly, which plugins were never meant to do, must change its import: import from@nextlyhq/plugin-sdk/admin(or from@nextlyhq/admin/kit, only for application code that must). The documented ones areBreadcrumbs,Checklist,ChecklistItem,ChecklistMark,ChecklistProgress,ConditionRow,DataTableView,DocumentPrimaryAction,DocumentSaveIndicator,EDITABLE_VALIDATION_RULES,FieldDefaultValueInput,FieldOptionsEditor,FieldPicker,FieldSettingsPanel,FieldTypePicker,FileDropArea,FormLabelWithTooltip,FrameState,ListColumnsMenu,ListEmptyState,ListToolbar,ListView,PlaceTabs,SegmentedChoice,SettingsRowGroup,Stepper,TargetSlot,TopBarPlace,ValidationNumberField,ValidationRulesEditor,Widget,apiErrorMessage,describeFieldTypes,drawsAnyValidationRule,formatWidgetTime,humanizeFieldName,isApiError,operatorTakesValue,operatorsForType,readPluginRoute,registerKnownPlugin,toastMutationResult,useBeforeDocumentAction,useCurrentUserId,useDocumentActions,useDocumentField,useDocumentLocale,useDocumentValues,useFieldAddress,useFoldedPanel,useListColumns,usePluginClientConfig,usePluginFieldTypeEntries,usePluginRoute,usePluginRouteMutation,usePluginTranslate,useRefreshDashboard,useReportUnsavedWork,useSingleDocument,useSuppressAdminChrome,useTableColumns,useTrackedWrite,useUpdateSingleDocument,useUploadMedia,useWidgetClock,validationIssues,withOptionIds,writePluginRoute, and the typesActionGroup,ActionPlacement,AdminChromeLayer,ApiError,BeforeDocumentAction,BreadcrumbItem,BreadcrumbsProps,CardEmptyContent,CardState,ChallengeResolveResult,ChallengeViewProps,ChecklistDoneStyle,ChecklistItemProps,ChecklistMarkProps,ChecklistProgressProps,ChecklistProps,ChecklistStatus,ConditionOperatorName,ConditionRange,ConditionRowProps,ConditionSource,ConditionSourceOption,ConditionValue,CustomEditViewProps,DataTableGroup,DataTableGrouping,DataTableRowDetail,DataTableViewProps,DocumentAction,DocumentActionOutcome,DocumentActionReplacementProps,DocumentActions,DocumentChangeResult,DocumentField,DocumentPanelSectionProps,DocumentSaveState,DocumentTabProps,DocumentViewProps,FieldAddress,FieldDefaultOption,FieldDefaultValueInputProps,FieldGroupOption,FieldOption,FieldOptionsEditorProps,FieldPickerChoice,FieldPickerConfig,FieldPickerProps,FieldSettingsConditionsProps,FieldSettingsControlProps,FieldSettingsField,FieldSettingsPanelProps,FieldSettingsSurface,FieldSettingsSurfaceControlProps,FieldSettingsTab,FieldTypeDescription,FieldTypePickerProps,FieldTypeSettings,FileDropAreaProps,FixedFieldSetting,ListColumnsControl,ListEmpty,ListSearch,ListSlots,ListToolbarProps,ListViewProps,NextlyColumn,PlaceContributionProps,PlaceProps,PlaceTabBadge,PlaceTabProps,PlaceTabsProps,PluginRouteInvalidation,PluginRouteMethod,PluginRouteRead,PluginRouteRequest,PluginRouteWrite,PluginRouteWriter,SegmentedChoiceProps,SegmentedChoiceSize,SegmentedOption,SettingsRowGroupProps,SidebarSectionProps,SkeletonShape,StepState,StepperProps,StepperStep,SubmitRefusal,TargetSlotProps,TopBarPlaceName,TopBarPlaceProps,UseListColumnsOptions,UseTableColumnsOptions,UseTableColumnsResult,ValidationIssue,ValidationNumberFieldProps,ValidationRuleValues,ValidationRulesEditorProps,WidgetBarChartGroup,WidgetBarChartProps,WidgetChartUnit,WidgetFigureProps,WidgetListProps,WidgetListTotal,WidgetRenderSize,WidgetRowAction,WidgetSeries,WidgetSeriesChartProps,WidgetSeriesInterval,WidgetSeriesPoint,WidgetTableColumn,WidgetTableProps. The SDK takes them from@nextlyhq/admin/kit, which is internal and may change in any release: import the SDK, never the kit. -
#4186
5a5e28bThanks @nextly-agent! - An admin screen opened by its address starts downloading its code while the session is still being checked, rather than after the answer, so a direct visit to a screen no longer waits one extra round trip before it can draw. -
#4203
2a12b90Thanks @nextly-agent! - The admin's shared first load is about 22 KB (gzipped) lighter again: the form library, and the query hooks only entry screens use, now load with the screens that use them rather than before the first screen draws. -
#4210
ff45991Thanks @nextly-agent! - The admin's shared first load is about 9 KB (gzipped) lighter again: the two checks it runs on what it remembers in the browser and on the branding the page carries now use zod's smaller functional build, with the same rules. -
#4217
b19a0ceThanks @nextly-agent! - The admin downloads less on every screen and ships less in total. The licence banners its bundled packages carried in every chunk (one per icon, from the icon set) are replaced by a singleTHIRD-PARTY-NOTICESfile in the package, holding each bundled package's own licence text.tailwind-merge,clsx,class-variance-authorityand the Radix primitives are no longer bundled into the admin:@nextlyhq/uidepends on them at the same ranges, so an app's build now ships one copy of each instead of two. In a production build of an app with a plugin page, the admin's first load falls by about 10 KB (gzipped). -
#4245
83e6fb4Thanks @mobeenabdullah! - The SEO plugin's AI crawler names were checked again against each provider's own page on 13 November 2026, and are kept as data naming that page. robots.txt now also asksMistralAI-Trainingto stay away under "search but not training" and "block all", andMeta-WebIndexer,MistralAI-Index,Amzn-SearchBot,Amzn-UserandClaude-Agentunder "block all", since their providers list them. No choice asks OpenAI's ads checker (OAI-AdsBot) orGoogle-CloudVertexBotto stay away: they visit only what the site itself asked for. The settings say so beside the choices. -
#4337
e06e545Thanks @mobeenabdullah! - An admin session no longer ends when the browser's network address changes. Withsecurity.trustProxyon, as on Vercel, a refresh from another address than the last one signed the editor out and ended the account's sessions on every device, which happened within half an hour of signing in on networks that send each connection out through a different address. The refresh now goes through and logs arefresh-client-changedwarning instead. A page a browser opens without a usable session, such as Search Console's return from Google after the session expired, now goes to the sign-in page, which refreshes the session or asks you to sign in and then opens the page again, instead of showing a raw error. -
#4229
a6aefa3Thanks @nextly-agent! - A field rewrite of a collection entry (rewriteField,@experimental) now matches a Single's in two places:- Field rules judge the value being replaced. A rewrite made for a person applies the field write rules to each copy's new value against the value it replaces, as an update does. A replacement group or repeater that leaves out, or changes, a value the person may not change is refused with
FORBIDDEN, and nothing is written. Before, the rules saw only the new value, so leaving a protected child out of a group deleted it. The refusal names each protected field (publicData.errors[].path, such ascard.locked) and the copy, for an entry and a Single alike. - A translated rewrite moves only its language's version. Where a rewrite of a translated field changes only one language's live value, only that language's version moves. A read of another language stays current, so its rewrite or save isn't refused as out of date. The entry's own version still moves when a shared field or any draft changes.
- Field rules judge the value being replaced. A rewrite made for a person applies the field write rules to each copy's new value against the value it replaces, as an update does. A replacement group or repeater that leaves out, or changes, a value the person may not change is refused with
-
#3912
d9f3368Thanks @muzzamil-rx! - A schema apply that takes a value away from an enum column while a row still holds it is now refused before anything runs, naming the table, the column, the values the rows hold and how many rows hold them — instead of failing at the database after the old check's drop had already run, with an error naming none of them. A narrowing no row holds applies as before. The refusal also follows a column renamed in the same apply, and a value MySQL reports escaped, such as one holding a backslash.On MySQL, which commits each schema statement on its own, a check that is replaced now runs as one statement per table, drops and adds together, both when a development push applies it and in the files
migrate:createwrites. A row the new check refuses then fails the change with the old check still in place, rather than leaving the column unconstrained, and two checks that exchange names apply where they used to fail. A migration file keeps the old check through a rename or a type change of its column and replaces it afterwards, so a narrowing that fails leaves the renamed column still constrained. -
#4334
0a27fb8Thanks @mobeenabdullah! - An entry whose stored upload inside a group or a repeater is the media object rather than its id now opens in the admin editor, where it showed "This document kept changing while it loaded" and could not be edited. Saving such a document now stores every upload, at any depth, as its media id, as a top-level upload always was. -
#4194
eaf7a50Thanks @nextly-agent! - New API keys carry a public id:nx_live_<id>_<secret>, where the 24-character id holds a tag only this install can make. From a client whose address the server cannot resolve (the default withoutsecurity.trustProxy), a key's failed lookups are now bounded per verified id before any lookup, so a flood of invalid keys can no longer refuse a valid key that carries an id. Keys made before ids keep working under the shared bound as before; replace one with a new key to get the per-key bound. The key's display prefix now names its id rather than the first characters of its secret. -
#4205
5953cebThanks @nextly-agent! - The app's schema hook can change how a collection's, a Single's or a component's field column is stored, withschema.overrideColumn(table, column, col.varchar(10)). A string field's column may becometext,longText,shortTextorvarchar(n), a whole number abigint, and a decimal may gain digits. Every push and migration keeps the override, a SQLite rebuild included. A string narrowing is judged by the stored data. A change outside the column's value family, a plugin's override, and an override of anything but the storage are refused when the hook runs. -
#4311
d982df3Thanks @mobeenabdullah! - The SEO plugin's address-record catch-up no longer leaves the site waiting on a transaction that is not running. On a serverless platform such as Vercel, a function is frozen once its response is sent. The catch-up runs after a start without being awaited, so it could be frozen with a transaction open. That transaction kept its locks on the address record for minutes. Every other function's catch-up and every save of a page then waited for those locks until its statement timed out. The waiting requests used up the database's connections, pages slowed down, and some answered 503. On PostgreSQL, each transaction the catch-up opens is now bounded by the database. A lock it waits more than 2 seconds for fails its run, and the run is tried again later. A transaction of its own left idle for 5 seconds is ended by the database, which releases its locks. A failed catch-up is now tried again at most once every 10 seconds, rather than by every request that finds the record not ready. Moving the not-found log to the plugin's table, which also runs unawaited after a start, is bounded the same way.Plugins get the same protection through
ctx.db.withTransactionBounds({ lockWaitMs, idleMs }, work)(experimental). It bounds every transaction the work opens, including those of the core services it calls. Adapters get the matchingwithTransactionBounds(experimental). PostgreSQL applies the bounds; MySQL and SQLite run the work unchanged. -
#4297
608ed4fThanks @nextly-agent! - The blog template's category and tag pages list their posts again. They filtered a post's categories and tags withcontains, which a has-many relationship now refuses with a400(FILTER_OPERATOR_NOT_FOR_LIST), so those pages, the post counts beside each category and tag, and a post's related posts came back empty. They now filter within, which matches a post whose list holds the id on every database. -
#4127
3e7fad1Thanks @zeshan-rx! - A Schema Builder save of a field group or a Single now writes its migration, as a collection's always has. Creating, changing or deleting one in the Builder writes a migration and snapshot to the migrations folder. In development they are run and recorded applied; elsewhere the deploy'snextly migrateapplies them. Before, these saves changed their table with no migration recording it, so the next migration, a collection's included, was refused with "schema drift detected: table 'comp_' present in DB".A field group's or Single's table an earlier Builder save made with no migration is adopted the next time it is saved in the Builder: the save first writes a migration describing the table as it stands.
nextly migrate's drift refusal names such a table and says to save it once in the Builder. -
#3952
6115a78Thanks @faisal-rx! - The Schema Builder gains one rule for where a field may move, which its drag, menu and keyboard moves will all follow: into and out of groups and into repeating rows at any depth, refusing, with the reason, a move out of repeating rows, into a blocks field, into or out of a field group, of a read-only field, of a unique field (a one-to-one among them) into a group or repeating rows, or to a place where its name clashes or is reserved.nextly/confignow exportssiblingNameClashandbuilderColumnIsUnique(experimental), the server's own rules for whether two sibling names clash and whether a field's column is unique, so the Builder refuses exactly what a save refuses. -
#3951
047e16fThanks @faisal-rx! - The Schema Builder keeps every edit as a pending change until it is saved: the toolbar shows how many are unsaved, with Undo and Redo (also Ctrl or Cmd with Z, and Shift or Y to redo, outside text inputs), each saying what it did, and Discard all, which asks first and names how many changes go. Each changed row in the outline is marked new, changed, moved or deleted, an undone edit or one put back as it was saved counts as none, and a deleted field stays in the outline, marked with how many fields inside go with it, until it is saved or undone.Leaving the Schema Builder with unsaved changes asks first, the same way an entry's form does: an in-app link, a field group's editor or used-by link, and the browser's back each ask in one dialog, and a reload or closing the tab asks through the browser. Nothing is asked when there are no unsaved changes or the builder is read-only.
-
#4161
0ee9c04Thanks @nextly-agent! - The Schema Builder now keys a relationship made in the admin. A relationship whose target is named byrelationTo, as the admin writes it, gets its foreign key to the target collection's table when its collection is created or when the relationship is added, on PostgreSQL, MySQL and SQLite. A relationship to Users, a Single, several collections, or one holding several entries gets no key.Behaviour change, for relationships created or added from this release on (existing ones keep no key and nothing is planned for them on upgrade):
- Deleting an entry that a required relationship still points at is refused, with a message naming the collection, the field and how many entries point at it, instead of "Referenced record does not exist.". A bulk delete reports each such entry and deletes the others.
- Deleting an entry that an optional relationship points at empties that relationship in the database. The referring entries' hooks, versions and webhooks do not run for that change.
- Saving a relationship to an id that does not exist is refused by the database.
- A save that points a keyed relationship at another collection is refused before anything runs, naming the field; the key stays. Switching it to hold several entries still converts each value into a one-item list, the key going with the conversion.
A relationship whose
relationTonames a different collection from itsoptions.targetis no longer keyed to theoptions.targettable, since its values are ids of therelationTocollection. Deleting several collections together orders the drops by the keys the database actually holds. -
#4169
9134bb1Thanks @nextly-agent! - A relationship made in the Schema Builder that points at Users now gets its foreign key to the users table, on PostgreSQL, MySQL and SQLite, as a relationship to a collection does. Before, a relationship naming Users byoptions.targetwas keyed to adc_userstable that does not exist.Behaviour change, for relationships created or added from this release on: deleting a user that a required relationship still points at is refused before anything of the account is removed, with a message naming the collection, the field and how many entries point at it. An optional relationship to Users is emptied, and the account is deleted.
-
#4258
bdc174fThanks @nextly-agent! - A Schema Builder save of a field group that holds a group with no name no longer fails while it writes its migration. A field-group create under a name already taken is refused with the name, asThe field group "<name>" could not be created: its name is already taken., instead of the bare "Resource already exists.". On MySQL, a collection's Builder save refused because a column made required holds empty rows is refused with that reason, and its files are withdrawn; it was reported as a migration that failed part-way with its statements kept, and its files were left fornextly migrate:resolve, though none of them had run. -
#4301
c547d24Thanks @nextly-agent! - A collection save is read once. A save that publishes was read once for the value rules a publish applies and again for the write's own gate and the write, so a body whose values changed between reads could be judged as one save and stored as another. The value rules, the gate and the write now all judge one copy of what was sent. -
#4211
4ae9171Thanks @nextly-agent! - A key one owner adds to another owner's table is judged by the contributor's migration stream, so a plugin that keys a dependency's table to one of its own tables starts again instead of being refused. Before a run, a plugin module's key may reach a table the database already holds, such as an installed dependency's or an adopted table, and is refused, naming the module and the key, when its target neither exists nor comes from an earlier stream. A pending migration file or plugin module is judged against the server's limits on what it adds: the checks and expression indexes its snapshot adds over the one before it, its own contributions only on another owner's table, and the statements it would run, so a hand-written file with no snapshot is judged too, and a construct an earlier file added no longer refuses an unrelated later one. -
#4250
b52f3c3Thanks @nextly-agent! - A create refused over leftover tables now ends a command with its own error code. When a collection, field group or Single create finds tables nothing registered owns where it would build, and they hold rows, are referenced from another table, or stand in the way of a junction table or under a view, the refusal reads asCREATE_BLOCKED_BY_LEFTOVER_TABLE, naming the tables and what to do, instead of an unexpected error. The code has its own page under the CLI error docs. -
#4130
713bf1bThanks @nextly-agent! - A plugin's own document page (a custom edit view, such as a form submission or a form) leads back to its list under the list's name: "Submissions", not "Submission". The trail step above a custom edit or create view now says the collection's plural label, as the list and the default editor already do. -
#4266
b364a11Thanks @nextly-agent! - Dashboard cards keep their height from the first paint, so the rows under them no longer move when a card's data arrives:- Each card's drawn height is kept on the device, per card and band of width, as a number alone, forgotten with the rest of what the admin keeps (on sign-out and by the site's off switch). A fresh tab draws each card's loading state at the height the card last stood.
- The setup card's loading state draws its checklist, a progress line and four steps, rather than list rows, so a first visit stands as tall as the checklist that replaces it for most projects.
-
#4312
8bd2a57Thanks @mobeenabdullah! - The SEO plugin's structured data now dates a page'sdateModifiedby when its published content last changed, the same date its sitemap entry's<lastmod>uses. Before, it used the entry's own last save. An SEO-only save, such as importing old SEO values, therefore moveddateModifiedto the import time on every imported page while the sitemap kept the honest date. A field that says when a page changed, such as a correction or review date or the field a content type maps, still comes first. A page with no content date kept yet still uses its own last save, as its sitemap entry does. -
#4237
1d1efafThanks @nextly-agent! - An admin answer's dates are now rendered in the configured timezone once. The handlers that can be mounted on a route of their own (/singles, a Single,/general-settings) format their answer themselves, and the admin's catch-all route formatted it again, parsing, rewriting and serializing the body a second time to no effect. The catch-all now passes on an answer its handler already formatted as it is. The answers are unchanged. -
#4126
afd4d40Thanks @nextly-agent! -nextly db:synckeeps a field's values when the field stops being translated. Before, settinglocalized: falseon a field of a translated collection, Single or field group left every value in the translations table and gave the field an empty shared column, so every read of it came back empty. The sync now moves the values the waynext devdoes: the other languages go to the localization archive, the default language's value becomes the shared value, and then the translated column is removed. A Required or Unique field gets its setting once its values are copied. When they can't take it, the sync refuses before anything changes, naming the field.nextly db:sync --watchdoes the same on each re-sync. -
#4187
5e5f15fThanks @nextly-agent! - A development push no longer drops what a plugin removed from the configuration added to another owner's table. Its columns keep their values, and its indexes, foreign keys and checks stay, untilnextly plugins uninstallremoves them or the plugin comes back, as its own tables already did. A push now records who added each column, index, foreign key and check to a table another owner declared, asnextly migratedoes.A push that would still change one of those elements is refused before anything runs, naming the plugin and the element: another owner declaring an element of the same name with a different shape, or, on SQLite, a change that rebuilds the table holding a removed plugin's column, foreign key or check. A plugin still configured that stops declaring one of its elements still has it removed.
-
#4249
7cbe0a9Thanks @nextly-agent! - The Direct API now follows the stored schema version as the admin API does: before a collection or Single operation, and before each form and field group call, a server copy reads the stored version (at most once every 2 seconds, sharing the read) and loads its definitions again when another copy changed the schema. Before, a process that served no admin request, such as an instance that had only rendered site pages since it started, or a script, kept the definitions it started with for as long as it ran, so a field added elsewhere was never read and a removed one failed its reads. Nothing is read inside an open transaction. The start now records the stored version it loaded the definitions at, so the first admin request or Direct API call after a start no longer loads every definition a second time to be sure.Deleting a collection's or Single's registry row now always moves the stored schema version, including the orphaned collections and Singles
nextly db:sync --remove-orphaneddeletes, so a running dev server stops serving a deleted collection or Single within 2 seconds rather than at the next unrelated schema change. -
#4164
e7a86b6Thanks @nextly-agent! - Hold the Direct API'snextly.login()to what every other sign-in does.- Its token is valid for 24 hours instead of 30 days. A token signed before this release keeps its 30 days, and still ends when its account's sessions are ended. A server that signed in monthly now signs in daily; a machine caller that wants a longer-lived credential uses an API key.
- Its token carries the account's roles in force, so
getSessionreads it with them, and it ends no later than the first of those roles expires. It used to carry no roles. - Each call is recorded on the audit trail: a
login-succeededrow naming the account, or alogin-failedrow naming none, both withmetadata.strategy: "password"andmetadata.via: "direct-api". - A deactivation, or an administrator setting the password, that lands while a login is running, after its password check, still ends the token the login issues.
-
#4179
dbdef0fThanks @nextly-agent! - A signed direct upload (clientUploads) is now checked before it is signed. Its name, extension, type and size go through the same checks and settings as any upload, an SVG is never signed (it goes through the standard upload, where it is sanitized), and the extension must name the claimed type. The S3 signature now binds the type and the exact size, so S3 refuses any other file, and a requestedexpiresInis honoured up to the configured lifetime. Callers ofPOST /api/nextly/storage/upload-urlmust now send the file'ssize, and Direct API callers pass{ size }tonextly.storage.getClientUploadUrl. Vercel Blob no longer reports client-upload support it never had. -
#4273
3539167Thanks @nextly-agent! - An open document's bar folds History, the preview and the panel toggle into its More menu by the admin bar's own width, never by how long the page's trail is: from a bar wide enough for every control, the controls keep 32rem and the trail truncates first. The width is measured before the first paint, so the bar is no longer drawn unfolded for a frame and then folded, and the editor's outline folds by the same rule, so More stands where the outline drew it.The editor's outline draws its side column only where the editor docks its panel (1000 px of editor, by the editor's frame rather than the window), and not where the reader folded the panel away; the editor reads that choice before its first paint too.
-
#4246
e3d0eebThanks @mobeenabdullah! - A collection whose entries are not all served under one pattern can name a text field holding an
entry's own public address:publicAddressField(experimental), besidepublicAddress, which an
entry with an empty field still follows.publicAddressOfanswers the entry's own address, so
canonical links, sitemap entries and redirects follow it. The value is judged as a template's
result is, and a save with an address no visitor could be served at is refused on the field; one
another page holds is refused as any clash, and changing it moves the entry's address. -
#4280
ed4b5c8Thanks @nextly-agent! - The environment's validation refuses an invalid value from one place, so the full check and the
pool settings' own read log the same problems and stop with the same message. -
#4157
0b8196dThanks @nextly-agent! -@nextlyhq/eslint-plugingainsno-experimental-sdk, which names each@experimentalsurface of@nextlyhq/plugin-sdka plugin uses, as a warning inconfigs.recommended. A scaffolded plugin picks it up with no change. What counts as experimental is generated from the SDK's stability ledger and TSDoc tags, so a surface that graduates stops being flagged in the next release. By default the rule checks imports;context: truealso checks plugin-context members such asctx.db, andallowlists surfaces adopted on purpose. -
#4243
b5e8248Thanks @mobeenabdullah! - FAQ markup is now added only where someone chose it. The SEO plugin emitsFAQPagestructured data only on a page whose structured data type is "Questions and answers", chosen for its content type, for the page or by its route'sstructuredDataHints.type. Before, any page of another type whose route passedstructuredDataHints.questions, or whosefaq,faqsorquestionsfield held questions, also got anFAQPagebeside its own markup. Sites that got FAQ markup this way lose it on those pages after upgrading. Google stopped showing FAQ rich results in May 2026, so search listings do not change; a site that still wants the markup on a page sets that page's type, or its content type's, to "Questions and answers". The type choosers now list it as "Questions and answers · no longer shown by Google", and the page's chooser no longer claims that only types Google still shows are offered. -
#4332
31f41acThanks @nextly-agent! - A field group created where no Schema Builder save writer is set up, and a code-first field group's table created bynextly db:sync, now get the same column types as one the Schema Builder creates: the field group's creator reads every column type from the column descriptor. Before, it spelled some of them its own way: on PostgreSQL an email, password, select or radio wasVARCHAR(255)and a single referenceUUIDwhere the descriptor saystext; on MySQL a date wasDATETIMEwhere it saystimestamp; a multi-select was JSON and ahasManynumber a scalar column. Existing tables keep their columns, nothing is migrated, and a field group whose columns carry that earlier spelling still reconciles as healthy.nextly repair-columnsnames such a column as created before 0.0.2-alpha.77. -
#4335
fe5e20cThanks @nextly-agent! - Reconcile no longer reports a repair, or refuses, on a healthy field group with a single reference or a checkbox default. It now asks the table built from the definition which index and default each field's column carries: a single reference is indexed whether or not it declaresindex, and a Builder save writes no column default for a checkbox. The default the field group's creator writes (nextly db:sync) is still accepted. A missing reference index, or a default neither writer gives, is still reported. -
#4196
e4b33a2Thanks @nextly-agent! - The general settings row (site name, address, logo, time zone, sidebar groups) is now read at most once every 2 seconds by each server copy, with one read shared by every request that arrives meanwhile, instead of once for every API answer: an entry's edit screen read it 22 times, and now reads it once. A change saved through the copy that serves the save is seen there at once; one saved through another copy is seen within 2 seconds. Two kinds of read still go to the database every time. Preview links: their revocation generation and the site address they send their holder to are read on every mint, verification and redirect, so a revoke or a new address saved anywhere applies at once. AndpublicSiteSettings()(readPublicSiteSettings), whose answer pages cache under the tag a settings save expires: a refill after another server copy's save holds that save, never an older copy kept here. -
#4216
50b660dThanks @nextly-agent! - Two files Nextly writes now go below the root its services were registered with (basePath), as its generated migrations and schemas already did, rather than always below the working directory: the migration log (.nextly/logs/migrations.log), and an upload to local disk for a collection with no storage plugin (public/uploads, with the.gitignoreentry kept beside it). An app that passes nobasePathtogetNextlyhas the working directory as its root, so nothing moves; one that passes abasePathnow finds both below it, beside the migrations it already generates there. The test harness gives each instance a folder of its own, so a plugin's integration tests no longer leavepublic/,.gitignoreand.nextly/in the package they test. -
#4233
29a4610Thanks @nextly-agent! - A plugin whose schema prefix no longer matches the tables it created now ends a command asPLUGIN_SCHEMA_PREFIX_CHANGED, with what changed, why and the declaration that keeps its tables, instead of an unexpected error.getCachedNextly()called from work a plugin starts during a boot thatgetNextlydid not begin (a request-path boot, orcreateTestNextly) now waits for that boot instead of failing, and the instancegetNextly()returns carriesrewriteField, as the Direct API does.nextly migrateno longer loads the service registry when the command is imported. -
#4134
2de82edThanks @nextly-agent! - Awherefilter on a has-many relationship or upload now matches by membership, on every database, through REST and the Direct API.inandequalsfind entries related to any of the ids.not_inandnot_equalsfind entries related to none of them, including entries with no related entries. Before,inandequalscompared the whole stored list and found nothing, andcontainsmatched the list's text, so one id could match part of another.contains,searchandlikeare now refused on such a field with a400(FILTER_OPERATOR_NOT_FOR_LIST) that says to usein. -
#4172
16e4516Thanks @nextly-agent! - Pointing a Schema Builder relationship whose column holds a foreign key at another collection now moves the key there, on PostgreSQL, MySQL and SQLite. Every value is checked against the new collection before anything runs; if an entry holds a value the new collection does not have, the save is refused naming the field and how many entries are in the way, and the old key stays. The new key is added under a name of its own before the old one is dropped, so a failure part-way on MySQL leaves the relationship keyed. A target with no table a key can check (a Single, or a slug that names nothing) is refused. Any key added over a column that already holds values is judged against its rows the same way. -
#3907
c4dc7e9Thanks @muzzamil-rx! - A foreign key may only target a table an earlier migration stream creates, a plugin keying to another plugin's table must declare the dependency, a plugin module's keys are judged against its own modules' order at generation and before a run, and a declared check or expression index is refused on a server that cannot hold it, naming its version. A key whose target the app's stream or a later plugin creates was left to the database to refuse at a fresh install; now the declaration is refused, naming the stream. On MySQL before 8.0.16 a declared CHECK parsed and held nothing, and before 8.0.13 an expression index failed to build; both are now refused at the compile the push and the migration share, and a shipped plugin module's snapshot is judged again before any module runs. -
#4208
53aaeb5Thanks @aqib-rx! -nextly migrate:downnow undoes the rolled-back migrations' own changes to the permissions at every point, not only at the registry baseline's. A permission a rolled-back migration declared, which the point it returns to does not, is marked orphaned with its grants kept: rolling back a plugin release's permission-only migration withdraws the permission it added, and rolling back a migration that added a collection orphans the collection's permissions. A permission a rolled-back migration orphaned has its mark cleared where the point declares it; a plugin's, only while the plugin is installed and its configured release still declares it. A plugin installed after that point keeps what its install wrote, and the rollback adds and removes no grant.Plugin upgrades that change permissions: a plugin release that only adds or removes a permission travels as a recorded migration written by
migrate:create.nextly migrateapplies it like any other, so the admin role holds a new permission before any start, andmigrate:downof it orphans the permission with its grants kept. -
#4141
b572621Thanks @nextly-agent! - A save with no status, made while another editor's draft is pending, now keeps that draft's value of each group field the saver may not write. Before, the save put the live value back there, which undid the other editor's pending edit. The rest of the group is saved as sent. If the saver sends a value for such a field that is neither the live value nor the pending one, the save is refused withFORBIDDEN, naming the field, and nothing is written. -
#4178
58cc0c5Thanks @nextly-agent! - An app'spasswordHasheris now the one scheme for every account password. Creating an account, an administrator setting a password, a person changing theirs, a reset link, an invite, the first password and every sign-in all hash or check with it; before, only the users service'sauthenticateandchangePassworddid, so a password one path wrote could fail to sign in through another. It can now be set in the config asauth.passwordHasher, which every way Nextly starts reads (a request to the route handler starts Nextly withoutgetNextly's options). A hasher may declareisHash, so a stored value that is neither its own nor bcrypt is never compared. Accounts that already hold a bcrypt password keep signing in, and the first sign-in stores the password again with the app's hasher, keeping its sessions and when it was set. Without a hasher nothing changes: core's bcrypt, as before, andauthenticateandchangePasswordno longer fail with an internal error. -
#4218
dd785d0Thanks @nextly-agent! - A user's permissions are now resolved once when several of their requests arrive together, instead of once per request. The first admin screen after a server starts, or after the minute a user's resolved permissions are kept for, used to resolve them for every request in flight (up to 18 times over, about 4 statements each); the requests arriving meanwhile now wait for the one resolution under way. A role or permission change made on the same server copy still applies at once to every request after it, since a request never joins a resolution begun before the change, a resolution that fails refuses every request that waited on it and is not kept, and one that outlived a role assignment it read (an assignment that expired while it ran) is never served: the request resolves the user again on its own. -
#4151
6756a00Thanks @nextly-agent! - Fixes to plugin one-purpose secrets (ctx.security.secrets) and plugin names:- A secret's expiry is held to the whole second on every database. Before, SQLite and MySQL stored whole seconds while the mint returned milliseconds, so a secret could expire up to a second before the
expiresAtit was given. - A subject's 255-character bound counts characters, so a subject of emoji fits up to 255.
- With no
NEXTLY_SECRETset but retired secrets still listed inNEXTLY_SECRET_PREVIOUS(allowed outside production), new secrets are written with the unkeyed generation, not the first retired key. Clearing the retired setting later no longer strands them. definePluginrefuses a plugin name that isn't well-formed Unicode. Two names differing only in lone surrogates would otherwise key the plugin's stored values alike.
- A secret's expiry is held to the whole second on every database. Before, SQLite and MySQL stored whole seconds while the mint returned milliseconds, so a secret could expire up to a second before the
-
#4241
764f76fThanks @nextly-agent! - The admin's loading outlines now match the screens they stand for, so nothing moves when a screen's data arrives:- A collection's document screens can start with the rail's panel folded away by declaring
views[slug].foldedPanel: true(experimental). The admin's first frame reads the declaration, so the outline is drawn folded and the frame no longer slides 256 px when the editor arrives. Only the addresses the router opens in the editor are outlined so: a document's version history keeps the panel, and has an outline of its own, its header, rail and comparison where the page draws them. The history's header keeps a line for the document's name while the name is read, and its loading line gives way to its answer without moving. The Forms builder declares it, and keeps its submissions count's place in the bar while the count loads. The reader's choice to open the panel is now remembered per collection, so a choice made on the Forms builder before this release is asked again once. - A list's outline draws as many rows as the list drew the last time this person opened it, kept on the device as a number alone and forgotten with the rest of what the admin keeps. Before 0.4 s an outline now holds its place, laid out but not drawn, so a plugin's content under a list is not drawn up into an empty space and pushed back down.
- The users list lays its columns out by fixed widths, shared with its outline, so the columns no longer jump when the rows land, and its Columns control is drawn at once.
- The release tabs hold their counts' place while the counts load, and the releases list waits for their first answer before it decides whether to show its search. A later recount, after a search or the filter, keeps the search box, its focus and its text, and empties only the counts.
- A list's toolbar keeps its controls side by side on a narrow column, the search giving way first.
- A list's outline lays its table out as the table does, as wide as its columns' content, so a column's head no longer wraps in the outline alone, and draws its pager piece by piece as the pager does, so neither is taller than what replaces it. Each list's pager is named for the outline's check.
- The releases list's outline keeps the toolbar's place it drew until the list replaces it whole, even when the counts answer first.
- The users list's outline draws the columns plugins add to it, from the same registrations the table is drawn with.
- A collection's folded panel follows the signed-in metadata once it has answered, even when it lists no plugins at all; the prebuilt page answers only while that is on its way or could not be had.
- A route whose code is on its way still says it is loading to assistive technology while it draws the screen's outline.
- A signed-in screen whose code is still on its way draws its outline, the one the frame drew, instead of a loading state of its own.
- A collection's document screens can start with the rail's panel folded away by declaring
-
#4195
712ca3bThanks @nextly-agent! - The Page Builder now moves a published page on its live copy too, where before
only its draft moved. Removing a Layout with "Move N pages, then delete" (or
unpublish) clears the Layout from each page's live version and from its pending
draft in one field rewrite, and deleting a component that pages use turns its
copies into plain blocks on the live copy and the draft together. Each draft
stays a draft, so nothing an author hasn't published goes live, and the page's
History records the change as made through the Page Builder, with the reason.Changing a live copy needs the right to publish the page. For a person who may
edit a page but not publish it, the draft moves as before and the live copy is
left: a Layout removal still finishes, and its dialog names those published
pages so someone who can publish them knows where to look; a component delete
keeps the component until the page is published, as before.A page saved while it is being moved is read again and tried a few times, then
reported if it keeps changing.A field rewrite also reads the collection table on its own transaction now, so
on a PostgreSQL or MySQL pool of one connection a rewrite of a collection this
process had not read yet no longer waits on itself. -
#4275
2805999Thanks @mobeenabdullah! - A Section or a Box can be marked Never quote this in the Page Builder's Content tab, after Show when. Its element is rendered withdata-nosnippet, so search engines that read it, Google among them, show none of its text under the page's link and Google won't use it as a direct source in AI Overviews or AI Mode; visitors still see it. It is the Page Builder's own setting and works with or without the SEO plugin. Google reads the mark onsectionanddivelements only, so a block whose HTML tag is another one says so. The canvas tags a marked block "not quoted in search" on hover, and its Layers row says so to screen readers.The block document format gains an optional node field,
nosnippet, written only astrue. Documents written before it read and render exactly as they did.containerRootTag(experimental) in@nextlyhq/blocks-reactsays which element a Section or Box renders as its root. -
#4331
917aa5bThanks @nextly-agent! - Page builder: a refused variable or shared style removal no longer waits on naming the page it is refused for. Looking up the title of the first page the caller may open is held to the removal's 5 second time box; if it does not answer in time, the refusal still says why and counts the pages instead of naming one. -
#4242
ca3f81cThanks @nextly-agent! - The admin's password forms check the password rule the server publishes on/auth/setup-status: the configured minimum length, and the byte limit of the app'sauth.passwordHasher(none where the hasher reads a password whole). Before, they always applied 8 characters and bcrypt's 72 bytes, so an app with an unbounded hasher could not set a longer password from the admin, even though the server accepts one.nextly users:clear-plaintext-passwordsandnextly migratenow load the config, and never count a stored value that the configuredauth.passwordHasherrecognises.--confirm-default-hashernow states that every stored password was written by core's bcrypt or by that configured hasher. -
#4149
34286d7Thanks @nextly-agent! - The plugin password helper's breached-password check now reads a range answer only up to 256 KiB. A larger answer leaves the password unchecked rather than being read whole into memory. Asecurity.breachedPasswords.sourcethat names both aurland arangeis now refused by the configuration schema. At runtime it asks neither, and startup warns about it. -
#4263
6633f83Thanks @nextly-agent! - A part a plugin adds to an editor's places (collection.edit,single.editand the form inside them,entry.form) can now name the reads it makes as it opens, withstartupReads(experimental), the same declaration a plugin page takes. Its paths may name the place's ids in braces (/templates/entry-designs?collection={collection}&entry={entryId}atcollection.edit), which the page fills in, URI-component encoded, from its address. The page starts these reads as it first renders, beside its own reads, under the keyusePluginRoutereads, so the part's first read takes the started one and sends no request of its own. Boot refuses more than four reads, a path the admin could not ask for, an id the place does not give, orstartupReadsat a place whose page does not start them, naming the plugin, the contribution and the path. A started read is waited for on a revisit too, where an earlier answer is kept, and is handed only to a part that mounts as the page opens: a part first mounted later, such as a panel opened afterwards, reads afresh.An entry's and a Single's editor now wait for those reads, for 250 ms at most, and for their claim on the document, for a second at most, before they are drawn: a colleague's "being edited" banner and a plugin's side panel section are drawn with the editor instead of pushing the fields and the panel's sections down as they land. The Page Builder's Design section declares its read, so it no longer moves Status, Publishing and Details 202 px down after the editor appears.
While it checks a document's releases, the Publishing section also holds the place of the line saying why scheduling is off (a colleague holds the document), so the sections under it stay put when the answer lands.
-
#4213
e854617Thanks @nextly-agent! - A site whose plugins have admin pages loads less before the admin's first screen draws: a plugin's admin module, which the app's component map loads up front, now brings only the admin parts it uses, not the whole plugin kit. In a production build of an app whose plugin draws one plugin-page part, the admin's first load falls from 741 KB to 570 KB (gzipped). Plugins change nothing: they import from@nextlyhq/plugin-sdk/adminas before, which now takes each part from its own internal area (@nextlyhq/admin/kit/<area>). -
#4171
49ffee3Thanks @nextly-agent! - A plugin can add foreign keys, checks and relations to another plugin's table.extendTabletakesforeignKeys,checksandrelationsin the shapedefineTabletakes them, on a plugin's or the app's own table: the app on any plugin's table, and a plugin on the tables of a plugin it depends on. Each key and check ships in the contributor's migrations and is recorded as its element, so the table's owner never reads it as drift and uninstalling the contributor removes exactly what it added.A plugin migration that rebuilds another plugin's SQLite table to add a key or check, and a DOWN that rebuilds one to take it off, are no longer refused by the schema guard.
-
#4207
3204535Thanks @nextly-agent! - A plugin table's relation,ref()column or foreign key to another plugin's table (or to a collection, Single or field group another plugin declared) now needs that plugin independsOnoroptionalDependsOn, and is refused at startup otherwise, as one added to another owner's table already was. A table's own single-column foreign key now carries a relation edge, sowithcan follow it as it follows aref()column. A full plugin uninstall is refused while another owner's relation leads into one of its tables, a disabled plugin's included. -
#4166
309fc7bThanks @nextly-agent! - A plugin that builds on another plugin's schema now has to depend on it. Adding a hidden column or an index to a collection, Single or field group another plugin declared needs that plugin independsOn(oroptionalDependsOn), and so does an index over a column a third plugin contributed. Either one missing stops the app at startup, naming both plugins. The app is never asked: its schema runs after every plugin's.To upgrade, add the owning plugin to the dependent plugin's
dependsOn. -
#4160
822f54dThanks @nextly-agent! - A role that holds only a plugin's own settings permission now finds that plugin's settings in the admin: the Plugins panel lists " settings" for it, expanded and collapsed, and the rail shows the Plugins entry for it. The page already let that role in; the only built-in link to it sat behindmanage-settings. A settings manager's navigation is unchanged, and a role holding neither permission sees nothing new. -
#4175
c8d3da3Thanks @nextly-agent! - An installation on PostgreSQL can keep its tables in a schema other thanpublic: setdb.postgres.schemaand every core, collection, Single, component and plugin table, and the migration history, is created and migrated there. Several installations can now share one database, each in a schema of its own. Until this release any schema butpublicwas refused withPOSTGRES_SCHEMA_UNSUPPORTED; that code is retired.A start whose configured schema holds no installation, while another schema of the same database holds one, is refused with
POSTGRES_SCHEMA_MOVED, naming both schemas, because it would otherwise create and serve a second, empty installation. Pointdb.postgres.schemaback at the schema the installation is in, or setNEXTLY_POSTGRES_SCHEMA_NEW_INSTALL=1for the first start of a separate installation started on purpose. Existing installations inpublicsee no change. -
#4034
b886e0eThanks @zeshan-rx! - Publish or unpublish the languages of an entry you choose, and its shared part, in one request:nextly.publish()andnextly.unpublish(), andPOST .../entries/{id}/publishand.../unpublishover REST (experimental). Everything chosen moves together or nothing does, and a refused request lists every part refused, each issue naming its language. A language can go live before the default one, taking the shared part with it on the entry's first publish; the shared part cannot be taken down while another language stays live; a publish of languages alone leaves the shared part's pending edits pending; and a unique translated field needs its own value in each language that goes live.describe: trueanswers what a request would do without doing it. A collection can setrequireEveryLanguage: true(experimental) so each language must hold its own value for a required translated field. A validation issue now keeps the language it is in through the error envelope. -
#4143
0af6203Thanks @nextly-agent! - A publish that carries a pending draft's fields no longer refuses a plain value only because the draft's own siblings denied it. Such a value is judged again on the document the publish will produce, and published if the rules allow it there. A rule that reads another field now reads the value the publish writes, not the draft's, so a publish that sets that field back, and makes the result allowed, succeeds. A group, a list, or a value the draft deleted is still judged on the draft alone, and a publish that would discard a pending edit to a field the publisher may not write is still refused. -
#4294
22433a8Thanks @mobeenabdullah! - Publishing every language of a Single at once now runs the hooks a publish of one language runs before it writes:beforeOperation, the Single'sbeforeValidate(beforeUpdate), everybeforeUpdateheard on all types, itsbeforeChange, and the fields' hooks for what the publish carries. A hook that throws refuses the publish before any language changes, and one that takes the published status away refuses it as a conflict. Before this, a site's rule that refused a publish could be passed by publishing every language at once. -
#4290
f8d9fe9Thanks @mobeenabdullah! - Publishing every language of a Single at once now runs the update hooks a save of it runs, once the publish is durable: the Single's ownafterChange, everyafterUpdatehook heard on all types, and its fields'afterChange. Plugins that follow saves (a search index, content dates, a problem check) now see a publish of every language.publishAllLocaleson a Single also takescontextandrequestfor those hooks. -
#4307
f5a7885Thanks @nextly-agent! - Reconciling a field group no longer refuses one the Builder has just created. A Builder create builds its table through the migration it writes, and reconcile compared its columns with the spelling the earlier creator gave (VARCHAR(255)for an email on PostgreSQL,DATETIMEfor a date on MySQL), so a healthy field group was reported as changed. Reconcile now accepts today's spelling and the earlier one, since field groups of both kinds exist, and still refuses any other. No stored data changes. -
#4227
890215aThanks @nextly-agent! - The admin's collection and Single lists (/collectionsand/singleswith a page, a size, a search or a sort) are now answered from the registry listing each server copy already holds, instead of a count and a page query on every request: a dashboard, list or edit screen sends 4 to 6 fewer statements, and none for an unchanged registry. The reader's permissions still decide what is counted, so a total never includes a collection or Single the reader cannot see. A change is seen at once on the server copy that made it, and within 2 seconds on any other, as every registry read is.The collection list's search now behaves the same on every database: a case-insensitive match on the slug and description (PostgreSQL matched case-sensitively before, so searching "Post" now also finds "posts"), with
%and_matched as the characters they are rather than as wildcards. Sorting by name compares slugs character by character on every database, where the database's collation decided before. -
#4230
3f056d6Thanks @nextly-agent! - A registry row or listing that is not kept yet is now read once for every request that asks for it while the read runs, instead of once per request. The first admin screen after a server starts asks for the same collections', Singles' and field groups' rows from several requests at once, and read each of them as many times over (16 to 20 row reads on its first sign-in); they now share one read each. A read begun before a registry change is never shared with a request that arrives after it, and a read that fails fails every request that waited on it and is not kept. -
#4316
9027ad3Thanks @mobeenabdullah! -nextly migrate:statusand the end ofnextly migratenow name every registry row that differs from the newest applied migration's record of it. That includes a row lacking a field, a row holding a field the record does not, a row whose settings differ, and a row that is missing though the record holds it. Each line says to runnextly migrate, which brings the row back.migrate:status --jsonlists them underregistry.behindSnapshot. A production start also logs each code-first collection or Single whose row lacks a field the build declares, or has no row, with the same advice. It sends no statement for this, since it compares the rows it already reads.Such rows come from upgrading a site on a serverless host from 0.0.2-alpha.75 or earlier. The previous deployment keeps serving while the new one builds, and a start of that release wrote every registry row from its own configuration. A start of the previous deployment after the new build's
nextly migratecould therefore leave rows without fields the new build added, such as a plugin's field group. The admin then showed those types without the field, and nothing said why. Runningnextly migrateonce more after the new deployment is live writes the rows again. -
#4302
93be55cThanks @nextly-agent! - Removing a Design variable or a shared style no longer waits for someone to press Count now on a site no scheduler drains. Where the page builder's usage index has not counted every page, the removal counts them first, for up to 5 seconds, and then judges: an unused variable is removed, a used one is refused as in use, a site too large to count in that time is told the pages are still being counted, to try again shortly or use Count now, and a site whose index cannot become whole (a page too large to read in full, or blocks in a field nested under a named group) is told so, without counting in vain. Two removals at once in one process share one count.Plugins get an
@experimentalonMigrate(ctx, { applied })hook, whichnextly migrateandnextly migrate:freshcall for each enabled, installed plugin once a run leaves nothing pending, after the schema and access steps. It runs after migrate's lock is released, never at boot, in a dry run, or when--stepleaves migrations pending; a hook that throws makes the command exit non-zero while the applied schema stays applied. The page builder uses it, and itsonInstall, to count the site's pages into its usage index (for up to 60 seconds, stopping early where no count can complete the index), so a deployed site starts counted whether it migrates before or after installing the plugin. The bootnextly plugins installanduninstallmake foronInstallandonUninstallis now shared with it. -
#4306
2708c5cThanks @nextly-agent! - The ESLint plugin'sno-experimental-sdkrule now reportsPageRobotsandPageImagePreviewas experimental, like the rest of the page-metadata contract they belong to. -
#4327
f95ce8fThanks @nextly-agent! - The ESLint plugin'sno-experimental-sdkrule now reportsPluginTransactionBoundsas experimental, as the SDK tags it alongsidectx.db.withTransactionBounds. -
#4153
6946e46Thanks @nextly-agent! - The one-purpose secrets sweep now removes a code together with the older codes of the same subject and purpose that it replaced. Before, sweeping a short-lived newer code alone could leave an older spent code answering for it, changing a refused code's reason fromnonetospent. -
#4188
d3263f9Thanks @nextly-agent! - The SEO plugin's address record no longer goes stale when servers of the earlier release keep running during a deploy. 0.0.2-alpha.76 moves the record from the collectionnx_seo_addressesto its own tables, once, at start; a server still running 0.0.2-alpha.75 went on writing the collection afterwards, so a page it published, moved or deleted was missing from the record, and the redirects it saved were not answered. The move now keeps the collection's size and newest write, and each start (and each long-running server once more, half an hour after it starts) compares them. When they changed, the plugin rebuilds the record from the pages, makes the redirects owed for the moves those servers noted, and takes over the redirects they saved. A site that already upgraded to 0.0.2-alpha.76 while old servers ran is repaired at its first start on this release. A start where nothing changed costs one more small read and writes nothing.A page left without its address in that repair, because another page took it meanwhile, now also marks the SEO Problems list's last full check out of date. Both moves (the address record and the not-found log) now record in
seo__noteswhen they have been taken in, for the later release that stops declaring the old collections. -
#4206
35bdb54Thanks @nextly-agent! - The SEO plugin brings its address record up to date after it is reinstalled on a site where it was uninstalled with--keep-data. While it was uninstalled, pages were published, moved and deleted with no record of them, and the first start after the reinstall found the address templates unchanged and skipped the catch-up: a deleted page's address could never be used again, and a page published meanwhile could later have its address taken by another. The plugin's install hook, and its uninstall hook when data is kept, now forget the catch-up's fingerprint, so the first start after the reinstall runs the whole catch-up.Installing the SEO plugin with
nextly plugins installnow boots the app once to run that hook, as it does for any plugin with an install hook. -
#4279
0aa6104Thanks @mobeenabdullah! - The SEO settings' Content types tab now carries each type's SEO switch.- Collections and Singles are listed in two groups, each type with where it was made (in code or in the Schema Builder), its entry count and an example address. Rows start closed and open one at a time to the title pattern, Hide from search and structured data. Singles with SEO are listed for the first time.
- Each type has a switch, on by default. Switching one off says before saving that its pages lose their SEO and that nothing is deleted; switching a saved-off type back on says its values are used again. The bar under the tab names each switch waiting to be saved. Without permission to change the settings the switches show their state and cannot be moved.
- A site's own type without SEO sits greyed at the end of its group with the reason: left out in the site's code, or holding its own field named
seo, with a link to Import. - Past ten types, a search box and an All, SEO on, SEO off filter appear.
- Pages marks a page of a type with SEO off as "SEO off", says its title is set by its own code, and offers no Edit SEO. The Overview's Pages card counts those pages on a "Pages with SEO off" line that opens Pages filtered to them (
?status=seo-off). - The content types route answers each type's
kindandsource, and a type without SEO itsreason; the settings document route answerscanChange.
-
#4277
0800759Thanks @mobeenabdullah! - A page-metadata answer whose date to leave search results is not an ISO 8601 date and time with its zone, or names a day or time that does not exist, now has that date left out of the page's robots, with one warning in the log, and the rest of the answer served. In the SEO tab, typing something that is not a limit under "Shorter" keeps the limit already stored until a valid one replaces it. -
#4239
003373fThanks @mobeenabdullah! - The SEO plugin's Import no longer says a value was "never served". Where a page's code now sets
its own title or description, a different stored value is still skipped unless included, and the
screen and the report say what the plugin knows: the page's code sets its own value now, so
importing this one would change what the page shows; include it if the page showed it before and
should again.A site can show its own page for an address made Gone:
seoPlugin({ gonePage: "/gone" })
(experimental). The Proxy reads that page and answers with it and status 410; where it can't be
read or doesn't answer 200, the short plain page is sent, still with 410. The SEO docs also show
how to log addresses not found from Next.js'sglobal-not-foundpage withlogNotFound. -
#4315
79469a2Thanks @mobeenabdullah! - With Cache Components on, a page built in code that answers not found whilenext buildrenders
it is left out of the sitemap andllms.txt, as it already was with them off: Next.js keeps that
answer only in the record beside the page, which the plugin now reads. The guide's
global-not-foundsample reads the request inside<Suspense>, so a site with Cache Components
on builds with it. -
#4276
0c27883Thanks @mobeenabdullah! - SEO can now be switched off for one content type from the SEO settings, without a schema change and without losing anything.- The SEO settings hold a new list,
typesWithSeoOff, of the content types whose SEO is off (a collection's slug, orsingle:<slug>). Every type is on until it is listed. Only someone who may change the SEO settings can change it. - While a type is off, its entries have no SEO tab and the
seogroup is drawn nowhere on their pages. The plugin answers nothing for its pages, so each route serves only what its own code sets: no title pattern, no structured data, no social card, and nonoindexof the plugin's. Off is not "hidden from search"; use the type's Hide from search for that. Its pages leave the sitemap andllms.txt, and the problem list shows none of their problems. - Nothing is deleted. The column and every saved value stay, and the pages keep their addresses: a clash is still refused and a moved page still gets its redirect. Switching the type on again brings everything back as it was, without a redeploy. Values are only ever dropped by
nextly prune, after the configuration leaves the type out. - Pages lists the type's pages marked as SEO off (
standing: "seo-off"), and the Overview's page coverage counts them as a group of their own (seoOff). - The site's verification codes stay on the home page whatever its type's switch says.
- Upgrading adds one column to the SEO settings: run
nextly migrate:createandnextly migrate. Until then every type reads as on.
- The SEO settings hold a new list,
-
#4313
410c252Thanks @mobeenabdullah! - SEO: plain-text copies start with front matter (title, description, url, last_updated), keep the emphasis and code a reader sees, show picture fields with their alt text and make every link whole. The home page's copy is at/index.md(and/index.html.md). A copy is kept out of search withX-Robots-Tag: noindex, cacheable for five minutes, and points to llms.txt; each listed page's head links its copy. A request for a page's own address that asks for Markdown in itsAcceptheader is answered with the copy, never kept by a shared cache. Core: a page-metadata answer can name other forms of the page by media type (formats), placed as alternate links in the head. -
#4255
e9894d6Thanks @mobeenabdullah! - The SEO plugin can now add itsseogroup to every content type, with no list to keep up to date.seoPlugin()withoutcollectionsorsinglesadds SEO to every collection and Single the site defines, in code or in the Schema Builder, including types made later (each gets the group when the site next starts), and to Page Builder's pages. Users, media and other plugins' collections never get it. A type that already has a field of its own namedseois left as it is, with a warning at start, and gets no SEO tab.- Sites that name
collectionsorsingleskeep exactly that list. Nothing changes for them on upgrade. To move to every type, remove the list, then runnextly migrate:createandnextly migrateso each type gains itsseocolumn. exclude(experimental) leaves named types without SEO, such as an internal log. A type left out later keeps its column and values untilnextly prune.include(experimental) adds a plugin's collection whose pages are public. Both apply only where no list is named, and are refused beside one.- Where SEO is on every type,
pagePermissionsis refused, because each type's permission is declared before the site starts; name the types to use it. The dashboard's SEO issues card and its Needs attention line are shown to whoever holdsmanage-seo. - The SEO tab, the problem list, Inspect, bulk edit, import, the setup checklist, IndexNow and the settings' Content types tab all read the types that have SEO when they run, so a Schema Builder type is covered as soon as it has the group.
-
#4265
691fcf3Thanks @mobeenabdullah! -@nextlyhq/plugin-seono longer asks to confirm, or lists as problems, robots.txt blocks on a site that serves its own robots.txt file, since the plugin's rules are not the ones crawlers read there. AnAllowexception to a rule closing the whole site is judged by every page it can open, so a partial policy is no longer refused because a few sample pages happened to be closed. -
#4257
436c853Thanks @mobeenabdullah! -@nextlyhq/plugin-seolets a site add its own lines to robots.txt on SEO > Settings > Search engines and AI, served after the plugin's own and never in their place. A rule keeping every crawler, Googlebot or Bingbot away from the home page or the whole site is refused, pointing to Hide from search; a rule blocking pages the sitemap lists names them and saves only once confirmed, and each such page is an error on the problem list while the rule stands. Misspelt rules are warned about, lines that are not standard rules (such asContent-Signal) are kept with a note, a group naming an AI crawler is explained as joined with the AI tools choice for it, and the screen says when the site serves its own robots.txt. The tab also says that robots.txt can't take a site out of Google's AI Overviews or AI Mode. Upgrading: the rules are a new field of the SEO settings, so a site that applies its schema with migrations runsnextly migrate:createand thennextly migrateafter upgrading. -
#4271
85af770Thanks @mobeenabdullah! - Pages can say how search results may show them. Every listed page now asks for large image previews (max-image-preview:large), which Google needs to show a big picture in results and in Discover; switch it off on Settings › Search engines and AI. Under each page's Advanced section in the SEO tab, a page can choose its own image preview size, a shorter text under its link or none (max-snippet,nosnippet), keep its pictures out of image search (noimageindex), set a date after which it leaves search (unavailable_after; after it the page is servednoindexand leaves the sitemap andllms.txt, and the problem list names it while it is published), and switch onnoarchiveornocache, each off by default and set by nothing else.A page-metadata provider's answer may now carry these display controls in its
robots(PageRobots, experimental), and the route helpers serve them; a route's own stricter limits are kept.The new values are optional fields of the
seogroup and of the SEO settings. Nothing changes for a page that sets none, beyond the large image preview default. The SEO settings Single gains alargeImagePreviewscolumn: runnextly migrate:createandnextly migrateafter upgrading; until then, pages keep asking for large previews.
Notes truncated to stay within GitHub's release body limit. The complete changelog for every package is in its
CHANGELOG.mdat v0.0.2-alpha.77.
Packages
@nextlyhq/adapter-drizzle@nextlyhq/adapter-mysql@nextlyhq/adapter-postgres@nextlyhq/adapter-sqlite@nextlyhq/admin@nextlyhq/admin-css@nextlyhq/blocks-engine@nextlyhq/blocks-react@nextlyhq/builder@nextlyhq/eslint-plugin@nextlyhq/plugin-form-builder@nextlyhq/plugin-mcp@nextlyhq/plugin-page-builder@nextlyhq/plugin-sdk@nextlyhq/plugin-seo@nextlyhq/storage-s3@nextlyhq/storage-uploadthing@nextlyhq/storage-vercel-blob@nextlyhq/uicreate-nextly-appnextly
Upgrading
- Run
nextly migrateright after upgrading (the scaffolded build script already runsnextly migrate && next build). A production start only serves whatnextly migratewrote, so a new field, setting or table from this release answers with a message namingnextly migrateuntil it has run.nextly migrate:statusnow names every content type whose stored record differs from the newest applied migration's, each with "Runnextly migrate". - From alpha.75 or earlier on a host that keeps the old deployment serving during the build (such as Vercel): run
nextly migrateonce more after the new deployment is live (a redeploy does it). An alpha.75 start can rewrite the content types' records after the new build's migrate, dropping fields a newly added plugin contributes; alpha.76 and later starts no longer write them. - Serverless hosts on PostgreSQL: the SEO plugin's background work now runs in bounded transactions, so a function frozen mid-transaction no longer holds locks. Setting
idle_in_transaction_session_timeouton the database (for example30s) is still a good guard for any app on a serverless host. - Sites with
security.trustProxyon: an admin session now survives a change of the visitor's network address (the refresh is logged asrefresh-client-changedinstead of signing the account out everywhere). - SEO plugin: the structured data's
dateModifiednow follows the page's content, as the sitemap does, so an SEO-only save no longer moves it. FAQ markup is drawn only for a page whose structured data type is "Questions and answers"; a page that relied on automatic FAQ markup sets that type on its SEO tab. Sites that list code-built pages useseoStaticParamsfor their routes. - An upload stored inside a group or repeater as the whole media object (older saves) now opens in the editor, and is stored as its id the next time the document is saved.