Previously the function always probed contacts.simplex_name first and
fell through to groups for NTPublicGroup misses. But the discriminator
(`@`/`#`) is embedded in the stored bytes via strEncode, so an
`#group.simplex` lookup can never match a contact row. Reorder to
case on nameType up front, saving one DB query and one withFastStore
transaction acquire on the group path.
The lookup `getContactBySimplexName` (Store/Direct.hs:781) filters
`AND deleted = 0`, but the index predicate `WHERE simplex_name IS NOT
NULL` covered tombstoned rows too. Forward-compat trap: once writers
land a non-Nothing simplex_name, soft-deleting a contact would block
re-claiming its name (UNIQUE conflict) even though the lookup reports
the slot as free.
Tighten the partial-index predicate to also require deleted = 0 so the
constraint scope matches the live-lookup scope. Groups have no soft-
delete column, so their index stays as-is.
connectPlanName now distinguishes "name not found" from "connection
link is invalid". CEInvalidConnReq's message ("Connection link is
invalid, possibly it was created in a previous version") was
misleading when a user typed @alice.simplex against a database that
simply has no contact by that name.
The two "missing prepared link" cases stay on CEInvalidConnReq —
the lookup found a row but the stored link is unusable, which is
closer to the existing semantics. The two truly-missing cases
(no contact found / no group found) move to CESimplexNameNotFound,
which also surfaces the name back to the client for a precise UX.
Follow-up to f71c579c. The SchemaDump test runs against SQLite only;
the parallel PostgresSchemaDump suite gates on -fclient_postgres and
a running localhost PG instance, which this environment doesn't have.
Updated the Postgres schema dump by hand to mirror the migration
change (two lines: CREATE INDEX → CREATE UNIQUE INDEX).
A simplex name is a stable, per-user identity (one name → one contact
or group). Without a unique constraint, a later writer that populates
the column twice for the same name would silently produce two matching
rows, and getContactBySimplexName/getGroupInfoBySimplexName would
return whichever the planner picks first.
Promote the partial indexes added in M20260603 to UNIQUE before any
caller wires the writes. Predicate (WHERE simplex_name IS NOT NULL)
already scopes the constraint to rows that opted in.
When a contact or group has a simplex_name stored, the share-link
render path emits the canonical simplex:/name... URI (via strEncode)
instead of the underlying connection link. Falls back to the existing
link rendering when simplexName is Nothing.
Final commit of the ConnectTarget plumbing chain: end-to-end users
can now (a) connect via @alice.simplex / #group.simplex with the
agent layer carrying the name, (b) see the simplex name on the
contact/group records and in viewConnectionPlan, (c) share the
contact using the namespace-canonical form rather than the raw URI.
viewConnectionPlan now shows the simplex name beneath known
contacts/groups (ILPKnown, CAPKnown, CAPContactViaAddress, GLPKnown
active/prepared/deleted). The TH-derived Contact / GroupInfo JSON
instances automatically expose `simplexName` (omitted when Nothing
per defaultJSON's omitNothingFields), which unblocks client-side
display and search.
No JSON test added: there is no Contact-level JSON test module in
this codebase; coverage is already provided by defaultJSON's
omitNothingFields = True behaviour.
Server-side substring search has no existing pattern in this
codebase; client renderers index simplexName themselves once it
appears in the JSON shape.
External-link share (preferring simplex:/name… form over the raw
link when the contact has a simplexName) lands in the next commit.
APIConnectPlan/Connect flip from Maybe AConnectionLink to Maybe ConnectTarget.
connectPlan dispatches CTLink -> connectPlanLink (the prior body, renamed)
and CTName -> connectPlanName (new) which looks the name up against
contacts.simplex_name and groups.simplex_name via the new
getContactBySimplexName / getGroupInfoBySimplexName store helpers.
The hit path returns the contact's / group's stored conn link from
preparedContact / preparedGroup; missing prepared state or unknown
names return CEInvalidConnReq. RSLV on-chain resolution is out of
scope for this branch -- known-name lookup is enough for conversation
display, search, and external-link share.
connLinkP_ parser is unchanged: APIConnect's preparedLink_ stays
ACreatedConnLink-shaped, and the Connect / APIConnectPlan parsers
already use inline strP for ConnectTarget without going through the
helper.
Directory.Service call sites updated to wrap their AConnectionLink in
CTLink when invoking APIConnectPlan.
createPreparedContact/createPreparedGroup gain a Maybe SimplexNameInfo
parameter that they write to contacts.simplex_name / groups.simplex_name
directly. createConnection_ writes to connections.simplex_name as a
transient carrier for the connect-via-plan path. The XInfo handler
in Library/Subscriber.hs reads the connection's simplexName and passes
it to createDirectContact so the final contact row captures the name.
All current callers pass Nothing; the actual flow lights up when
APIConnectPlan accepts ConnectTarget and connectPlanName threads the
name through (later commits in this branch).
Uses the upstream ToField SimplexNameInfo (simplexmq 0b334b66) for
writes; reads continue to go via the soft-degradation helper.
Two helpers redundantly maintain the same g.* column list. A future
g.* addition must be applied to both sites; the cross-reference
comments flag this for maintainers. A proper refactor (reusing
groupInfoQueryFields from Connections.hs's inline SELECT) is out of
scope for this branch.
Adds `simplexName :: Maybe SimplexNameInfo` to the three records and
extends every SELECT path that reconstructs them to read the new
column. Decoded via eitherToMaybe . strDecode . encodeUtf8 (the
codebase's established pattern for Maybe Text -> typed-decode fields),
extracted as decodeSimplexName helper since the chain appears in
toContact / toContact' / toGroupInfo / toConnection. INSERT paths
still write Nothing - the write-side wiring lands in the next commit
(Task 7).
Follow-up to the simplexmq pin bump (ee0a45e9). The new
namesConfig :: Maybe NamesConfig field on ServerConfig (introduced
in simplexmq's namespace branch) needs to appear in the test
fixture's record literal, otherwise the test suite fails to compile
under -Werror. Disabled by default (Nothing).
Nullable TEXT column on all three tables, with partial indexes on
contacts(user_id, simplex_name) and groups(user_id, simplex_name)
for the upcoming connectPlanName lookup. connections.simplex_name
is the transient carrier from APIConnect -> XInfo handler, where
the value is copied to contacts.simplex_name at delayed create.
No reads or writes yet - column threading lands in subsequent commits.
* plan: web previews for channels
* types for recipient side to support channel web previews and domain names
* fix
* migrations
* update schema and api types
* update schema
* rename migrations
* core: check member role
---------
Co-authored-by: Evgeny @ SimpleX Chat <259188159+evgeny-simplex@users.noreply.github.com>
* desktop: don't copy non-message items when selecting message text
Selecting text across messages also copied the text of event/info
items (e.g. "connected") that fell inside the selection, even though
those items are never highlighted as selected.
getSelectedCopiedText emitted text for every merged item between the
selection bounds. Event/info items have no msgContent but a non-empty
text, so as interior items their text was copied. Skip items whose
content has no msgContent - only real messages are copyable.
* plans: add 2026-05-20-fix-copy-non-msg-items.md
* core: fix /start remote host parser when iface name contains a space
The iface= field used jsonP (which calls takeByteString and strict-decodes
the entire remaining input as JSON). When port= followed iface=, the strict
decode failed on the trailing data and the text1P fallback stopped at the
first space inside the JSON-quoted interface name (e.g. "Ethernet 2"),
leaving unparseable junk and producing "Failed reading: empty".
Replace jsonP with a bounded quotedP that consumes only up to the closing
quote, leaving port=… for the next parser.
* plan: document fix for /start remote host iface-with-space parser bug
Move SectionDividerSpaced() out of the SectionView { ... } block so it acts
as inter-section spacing instead of being rendered as the section's last
child. Matches the pattern used by every other section in ChatInfoLayout.
Plan: plans/2026-05-25-fix-e2e-encryption-section-divider.md
* desktop: fix video playback hang caused by stuck preview snapshot
Problem: clicking play on a video did nothing when an earlier video's
preview generation was stuck — every subsequent VideoPlayer.play() was
queued behind it on the shared playerThread.
Cause: helper player reuse across previews exhausted the libavcodec h264
frame-buffer pool with --avcodec-hw=none (PR #6924), and the synchronous
libvlc snapshots().get() call then hung waiting for a frame that was
never decoded.
Fix: drop the helper-player pool (release each helper after use) and run
preview generation on a dedicated previewThread so a stuck preview can
no longer block playback.
* plans: add 2026-05-15-fix-video-preview-snapshot-hang.md
* desktop: capture preview via callback surface, keep helper pool
Follows up on the previous commit (4a964c66). The actual hang was in
libvlc's synchronous snapshots().get() on a reused helper, not in the
pooling itself. Replace the polling loop with a CallbackVideoSurface
(the existing SkiaBitmapVideoSurface) wrapped in withTimeoutOrNull —
the wait is bounded, so a non-decoding helper can't block previewThread.
Restore the helper-player pool that the previous commit dropped.
* plans: update 2026-05-15-fix-video-preview-snapshot-hang.md for final fix
* desktop: fix in-app updater silently failing on Windows
chooseGitHubReleaseAssets ran `which dpkg` unconditionally to probe for
Debian-derivative systems. On Windows there is no which.exe, so
Runtime.exec threw IOException, which the outer catch in checkForUpdate
logged and swallowed -- the update dialog never appeared. Gate the
probe on desktopPlatform.isLinux().
* desktop: fix in-app updater install step on AppImage
xdg-open on the downloaded .AppImage opened it in whatever the desktop
environment's default handler for the AppImage MIME type is -- usually
an archive viewer, which reports 'Archive format not recognized'. The
running AppImage was never replaced.
Detect $APPIMAGE (set by the AppImage runtime to the path of the
running .AppImage file). Copy the downloaded file to a staging file in
the target's own directory, mark it executable, then atomic-move it
onto $APPIMAGE. Staging in the target directory keeps the final move a
same-filesystem rename(2), so an interrupted copy never leaves the
running AppImage partially overwritten. On failure (permission denied,
target read-only, etc.) fall back to opening the parent directory so
the user can install manually -- the same fallback the existing
xdg-open path already used.
* desktop: fix in-app updater silently failing on aarch64 AppImage
The LINUX_AARCH64 githubAssetName had a literal leading space
(" simplex-desktop-aarch64.AppImage"), so the exact-name filter in
chooseGitHubReleaseAssets never matched the real release asset name
"simplex-desktop-aarch64.AppImage". The asset list came back empty
and checkForUpdate's early-return at "No assets to download for
current system" suppressed the dialog. Same silent-failure pattern as
the Windows bug.
* plans: justify desktop in-app updater fixes
* core: update simplexmq
* update agent api
* update simplexmq
* core: add flag to User to use client services
* update simplexmq
* cli command to toggle service for a user
* test, fix
* query plans, core/bot api types
* remove local package reference
* increase server queue size in tests
* show client service status in users list
* update query plans
* cli: fix redraw slowness (#6735)
* cli: add pland to fix redraw slowness
* updtae doc
* cli: decouple key reading from processing via TQueue
* schema and bot types
---------
Co-authored-by: sh <37271604+shumvgolove@users.noreply.github.com>
shum