src/types/schema.ts had drifted from the official schema contract: a `clientOnly` field had been added to `FilterEnum` to support client-side log filtering. Restored it to match upstream exactly and moved the deviation-only type into a new intersection type in src/lib/schemaDeviationTypes.ts instead. Audited the codebase for every place the UI does something the official schema doesn't support and tagged each with `// SCHEMA-DEVIATION: <id>`, documented in the new SCHEMA_DEVIATIONS.md registry (what, why, and the ideal server-side fix): - log-client-filters: Level/Event filters on Log Entries, applied client-side because x:Log/query rejects them as JMAP filters. - account-quota-usage-column: synthetic quotaUsage column on the Accounts list. - mailbox-client-hierarchy-sort: full-fetch + client-side sort to reconstruct mailbox parent/child hierarchy. - webapp-enabled-column-fallback: synthetic "Enabled" column label for x:Application when the schema list doesn't define one. Other viewName/objectName special cases (x:OtpAuth, x:Expression, x:Rate, x:Action, x:Trace, CustomComponent/*) were checked against upstream and are pre-existing architecture, not fork deviations.
6.2 KiB
Schema deviations
Stalwart WebUI is schema-driven: the server's JSON schema is the single source of truth for what forms, fields, filters, and columns exist. This fork tries to stay aligned with that philosophy (see stalwartlabs/webui#discussion and the maintainer's note on why exceptions belong server-side, not in the UI).
Everything in this file is a deliberate exception: a place where the UI
does something the schema doesn't (yet) describe, because the equivalent
server-side capability doesn't exist in stalwartlabs/stalwart
or stalwartlabs/webui. Each entry
is tagged in code with // SCHEMA-DEVIATION: <id> so they're greppable
(grep -rn "SCHEMA-DEVIATION" src/).
The goal is not to remove these — they're real functionality this fork wants to keep — but to track them separately from schema-driven code, so it's always clear which is which, and so each one can be dropped the day the server (or upstream webui) grows the equivalent native capability.
src/types/schema.ts is never touched for a deviation
src/types/schema.ts mirrors the server's schema contract exactly and
must stay aligned with the official webui/server types — it is never
edited to accommodate a deviation, even to add an extra optional field.
If a deviation needs to carry extra data on an otherwise-official schema
shape (e.g. a flag consumed only by the deviation's own code), the
augmented type lives in the deviation's own module or in
src/lib/schemaDeviationTypes.ts, as
an intersection with the official type (OfficialType & { extra?: ... }),
and is imported only where the deviation is actually used. schema.ts
itself stays byte-for-byte alignable with upstream's version of the file.
Status legend
- 🟡 Workaround — client-only, would be removed if the server supported it natively.
- 🔵 Upstream tracked — an issue has been filed upstream; link included.
Deviations
log-client-filters 🟡
- Where:
src/lib/logFilters.ts, type augmentation insrc/lib/schemaDeviationTypes.ts - What: injects
levelandeventas filterable columns on thex:Loglist, applied entirely client-side (clientOnlyflag consumed byDynamicList). TheclientOnlyflag is declared asClientOnlyFilterEnum(an intersection type), not on the officialFilterEnuminschema.ts. - Why: Stalwart's JMAP
x:Log/queryreturnsunsupportedFilterfor both properties today, even though they're returned per row. - Ideal fix:
stalwartlabs/stalwartacceptslevel/eventas real query filters; the schema then advertises them normally andlogFilters.ts+ theClientOnlyFilterEnumaugmentation are deleted.
account-quota-usage-column 🟡
- Where:
src/lib/accountColumns.ts - What: adds a synthetic
quotaUsagecolumn to thex:Account/Userlist that isn't a real schema property —DynamicListresolves it from theusedDiskQuota+quotas.maxDiskQuotapair and formats it specially. Also re-addsroles(a real property, just not in the list's default columns). - Why: the Accounts list schema doesn't expose usage/role as list columns, only as detail-view fields.
- Ideal fix: the server's
x:Account/Userlist schema includesrolesand a computed usage/quota column natively; this file is deleted.
mailbox-client-hierarchy-sort 🟡
- Where:
src/components/lists/DynamicList.tsx—sortMailboxesByHierarchyand theisMailboxListbranch - What: for the
Mailboxlist, fetches the entire result set (bypassing normal server pagination) and sorts it client-side so each parent mailbox is immediately followed by its children, with indentation depth tracked in React state. - Why: the server returns mailboxes in whatever order the query produces, not grouped by parent/child, and a mailbox's parent can land on a different page than the mailbox itself, so hierarchy can't be reconstructed one page at a time.
- Ideal fix: the server offers a native tree/hierarchical ordering (or a
sortthat groups by ancestry) forMailbox/query; the client-side full-fetch-and-sort is deleted in favor of normal paginated queries.
webapp-enabled-column-fallback 🟡
- Where:
src/components/lists/DynamicList.tsx—displayColumnsin theisWebApplicationsbranch - What: reordering the
x:Applicationlist's real schema columns (Description first, Enabled second) is not a deviation, but if the schema'slist.columnsdoesn't include anenabledcolumn at all, a fallback column definition with a hardcoded label is fabricated client-side so the toggle still renders. - Why: the
x:Applicationlist schema is not guaranteed to exposeenabledas a list column, even though it's a real object property (fetched separately viaproperties.push('enabled')). - Ideal fix: the server's
x:Applicationlist schema always includesenabledas a real column; the fallback branch is deleted (only the reordering logic remains, which is not a deviation).
Not a deviation (for reference)
A few other viewName === '...' / objectName === '...' checks exist in
DynamicList.tsx, MainContent.tsx, Sidebar.tsx, layout.ts, and
FieldWidget.tsx (e.g. x:OtpAuth, x:Expression, x:Rate, x:Action,
x:Trace, CustomComponent/Dashboard and other CustomComponent/*
pages, the x:Application column reordering itself, the active-WebApp
info card). These are not tracked here: they render real schema data
with a custom widget or extra display, the same pattern already used
upstream for special object types — they don't fabricate data or bypass
the server's filtering/pagination. Verified against upstream/main for
each: the object/view names above already drive special-cased rendering
there too, except x:Application/Mailbox/x:Log/x:Account/User
which are fork-only and covered by the entries above (or explicitly
noted as presentation-only, e.g. the Web Applications column reorder).