`file_change` is the biggest relation of the database: 683 GB in
production, of which almost all the TOAST is dead space (the legacy
data is already migrated to `file_data`; what is left is the xlog of
the revisions, 4.4 kB on average). It grows with every save of a file
and every version, and its rows are spread over the whole table, so
every query that filters by a file has to walk all of it.
The table is now partitioned by `HASH (file_id)` with 16 partitions, the
same bounds as `file_data`, so the queries that filter by a file can
prune to one partition and the join between the two tables is resolved
partition by partition. The primary key has to include the partition
key, so it becomes `(file_id, id)`, and every access path of the table
filters by `file_id` accordingly: the snapshot commands require it
(2.19), the frontend sends it, and the garbage collector locks the
pending rows by their `(file_id, id)` pair.
The migration creates the partitioned table with the final schema and
copies the rows, keeping the old one aside as `file_change_legacy` for
the rollback. The copy happens inside the transaction of the migration,
so it does not fit a big instance: there the migration is marked as
applied and the procedure of `operations/` does it by hand, in phases,
driven by the `updated_at` watermark.
Along the way:
- every mutation of the table moves `updated_at` to the moment of the
mutation, so the phased copy does not lose the rows changed while it
runs. It is never set to the `deleted_at` of the row, which for the
system snapshots is a date in the future.
- a new `kind` column tells the snapshots (1) and the revisions (2) of
the table apart, and the two paths that insert in it fill it. It has
no default on purpose, so an insert that forgets it fails instead of
storing the wrong kind silently. It is a reserve: the queries still
tell a snapshot from a revision by its label, and no constraint ties
the two together.
- the revision does not repeat the features of the file anymore: they
were 348 bytes written on every save and nobody read them from there.
- the legacy file data backend is gone, and with it the `data`,
`data_ref_id` and `data_backend` columns, which nobody wrote anymore.
- the migration copies the legacy blobs of the snapshots to `file_data`
before the column goes away, and stops when a snapshot with legacy
data is left without its blob: the backend is better down than losing
content.
- the foreign key of the legacy table on `file_id` goes away with the
copy: it is `NO ACTION`, so while the legacy table has rows, the
objects GC cannot delete the files they point to.
- the quota of the snapshots of a team resolves the files of the team
first and counts their versions with `= ANY`, instead of joining the
team with every version of the table.
- the pool sets `enable_partitionwise_join` on every connection, so the
join between the two partitioned tables is resolved partition by
partition.
- the snapshot lookup reports an unknown snapshot with its own error
instead of the generic "object not found" of the database helper, and
the decode script fails instead of printing an empty JSON.
- the procedure of `operations/` stops the backend and the workers
during the cutover: the pool keeps server-side prepared statements,
and a prepared statement holds the OID of the table it was prepared
with, so a connection would keep writing to the legacy table after
the rename.
- the procedure creates the partitions with their final name, so the
result is the schema a new installation gets, and its verification
checks the names and not only how many there are.
- the index of the system snapshots is gone: it filtered them by
`deleted_at IS NULL`, and they always carry a deletion date in the
future, so it never matched a row.
- the configuration guide, the memories and the docstring of the decode
script say that the `legacy-db` backend is gone, instead of offering
a value that takes the service down.
- a test fixes the schema the migration leaves: the 16 partitions with
their name, the primary key, the foreign keys, the check of the kind,
the six indexes, the absence of the legacy columns, and that an insert
without a kind or with an unknown one fails.
Closes#12416
AI-assisted-by: glm-5.3-flash
AI-assisted-by: minimax-m3.1-flash
Rewrite sql:restore-storage-objects so referenced storage ids are
collected as UUIDs first and then joined by primary key. The previous
form cast storage_object.id to text and OR-ed that with id IN, which
forced a scan of deleted storage rows and made restore-deleted-team-
files time out in production.
Closes#12727
* 🐛 Keep swapped and moved children in place when a component syncs
A sync resets an untouched shape of a copy to the main's geometry and the
layout then moves it back. Children the sync keeps, because they are
swapped or the user moved them, did not take the first move but did take
the second, so each sync pushed them away from their parent by the gap
between the copy and the main (e.g. a swapped icon in a copy that fills a
wider layout).
The sync now moves those children by the offset it applies to their
parent. Flex-child attributes still sync as before, so swapped children
keep their absolute position.
Closes#12657
* 🐛 Place swapped children where the swap puts them on sync and switch
A swapped child that the user never moved now goes to the top-left of the
shape it replaced, as a swap does: on a sync, the place of the main's slot
occupant; on a variant switch, the place of the new variant's occupant.
A swapped child the user moved keeps its offset inside its parent.
This stops the variant switch from leaving the swapped child at its old
position while the new variant is created at the main's size, and puts
back children that drifted before the sync fix the next time their copy
syncs.
* 🐛 Make kept children follow their parent's constraints on component sync
When a sync resets the size of a copy's shape, for example a copy root
that fills a wider layout, the layout then grows it back and moves its
children by their constraints. Children the sync keeps (moved by the
user, or swapped) only followed the parent's top-left, so with a right,
center or scale constraint they drifted on every sync, and a swapped
child with a scale constraint grew on every sync.
The sync now passes the parent's geometry before and after down the
tree, and kept children follow that change through their constraints.
Untouched swapped children also undo the change through their
constraints before they go to the swap place, so they keep their size.
The default constraint modifier returns empty modifiers instead of an
empty vector, which add-modifiers could not take.
* 🐛 Make swapped children follow their parent's constraints on switch
A variant switch only moved a swapped child by the parent's position
change. When the new variant's root had a different size, a user-moved
child with a right constraint drifted, and a scale-constrained one kept
its grown size.
The switch now applies the parent change from the old parent to the new
one through the child's constraints. An untouched swapped child then
takes the place of the shape it replaces, as a swap does.
* 🐛 Sync nested swapped copies before the copies that hold them
A swapped child is a copy of its own component, so one sync handles it
twice: once on its own, and once when the copy that holds it places it.
Each step read the shape from the data as it was before the sync, so
the step that ran last won, and the order came from the order of shape
ids. When the swapped child came last, it wrote its old size and
position back over the placement, leaving its x and selrect apart, and
a swapped child with a scale constraint grew on every sync.
The sync now visits the most deeply nested shapes first, and each step
reads the shapes as the earlier steps left them, so the placement
always comes last and builds on the child's own sync.
* 🐛 Keep moved children with a scale constraint at their size on switch
A variant switch carries the size of a child the user moved as an
override. In a copy that fills a wider layout, a child with a scale
constraint already has the size the layout gave it, so the switch
carried that size onto the new variant, created at the main's size, and
the layout then scaled it again: the child grew on every switch.
Before comparing, the switch now resizes the previous child the way its
constraints follow the change from its old parent to its new one. A
child that only grew with the layout then matches the main and is not
carried as an override.
* 🐛 Keep the content of swapped children at its size on sync and switch
When a sync or a variant switch undid the parent change on a swapped
child through its constraints, it applied the child's resize to every
shape inside it. The layout then grows the child back through its own
constraint, but its content follows its own constraints (left and top
by default), so it stayed shrunk: the shape inside a swapped icon with
a scale constraint lost most of its width on every switch.
The change of the swapped child now goes down its subtree the way a
layout update does, each shape through its own constraints.
* 🐛 Accept mixed and empty values in the layout size constraints menu
With several shapes or a group selected, the sidebar merges the layout
item attributes and tokens of the shapes: nil when a shape lacks one,
:multiple when they differ. The props schema of layout-size-constraints*
only accepted the values of a single shape, so selecting such shapes
with fill sizing raised "invalid props".
The schema now accepts nil and :multiple, as the numeric inputs those
values go to do.
* ♻️ Share the swapped child follow-parent step between sync and switch
The component sync and the variant switch both make a swapped child and
its children follow the parent change through their constraints and then
move them to the swap place. Both now use generate-follow-parent, so the
two paths stay the same. The helper that writes geometry without marking
it touched is renamed to update-geometry-untouched.
Adds a test for a swapped child without absolute position in a slot
without layout, and fixes the test comments on why constraints are set
again after a swap.
`critical-info` is the first memory every agent reads, and its
`scripts/ci` entry still described the orchestrator as lint + tests +
format over eight modules. `AGENTS.md` repeated the same incomplete
description in its scripts section.
Both now mention the opt-in production build and list `media-processor`
among the modules.
AI-assisted-by: deepseek-v4.1-flash
`./scripts/ci --build` compiles each module the way production does,
without going through `manage.sh build-bundle`, which wipes node_modules
and reinstalls dependencies. It runs the release commands already
defined per module and reuses existing caches, so a broken release build
(shadow-cljs optimizations, the wasm targets, the backend uberjar,
tsc/typedoc) shows up before a bundle is built.
The task is opt-in and runs last, after lint, test and fmt, so the
default flow keeps its duration. `common` has no build of its own and
reports as skipped. `media-processor` joins as the ninth module, with
its lint skipped and `tsc --noEmit` folded into its build.
All build outputs are gitignored. `scripts/test_ci.py` covers the flag
wiring and one command per module through `--dry-run`; it never runs a
real build.
AI-assisted-by: deepseek-v4.1-flash
pnpm 12.5.1 -> 12.16.1: the tarball pin of docker/devenv and of the
media-processor, exporter and mcp images, the packageManager field of
every workspace manifest, and the lockfile metadata of the 11
workspaces. The new pnpm also normalizes one engine range per lockfile
(>=18.* to >=18).
node 24.21.0 -> 24.22.2: devenv NODE_VERSION and both NODE_SHA256,
.nvmrc, and the runtime pin of the two plugins-deploy workflows. The
dhi.io image tags stay as they are until their digests are available.
AI-assisted-by: deepseek-v4.1-flash
The script walks the filesystem, so it stamped generated copies too, such
as media-processor/target/package.json: --check counted files git does
not track and failed whenever one of those copies was stale. Prune
target/ like the other generated trees, with a test for it.
Memory: the JDK and ImageMagick image pins have traps the code cannot
show, now in workflow/docker-images; memory-maintenance gains the rule
that a memory must not restate what the code already says; updating-pnpm
notes the lockfile-only refresh.
AI-assisted-by: deepseek-v4.1-flash
Sass adds @charset "UTF-8" to every module with non-ASCII characters,
and concatSass joined the modules as they came, so main.css carried it
in the middle of the file. Browsers only honor @charset as the first
rule, and the server sends text/css with no charset. When the browser
did not take the page encoding for the stylesheet, it read main.css as
Latin-1 and CSS content such as the bullet of the library element
counts showed as "•".
Drop every @charset while joining the modules and put one at the start.
Closes#12677
AI-assisted-by: claude-opus-5-5
A numeric input that can be scrubbed sets the ew-resize cursor on its
wrapper, and the token dropdown renders inside that wrapper, so the
whole list inherited the scrub cursor. Give the options dropdown the
default cursor, so it never takes the cursor of the control that opens
it.
AI-assisted-by: claude-opus-5-5
* 🐛 Stop thumbnail subscribers re-rendering on every store change
A thumbnail loaded from the server keeps a media id as :uri, and
workspace-thumbnail-by-id resolves it into a new map on each read. The
ref compared values with identical?, so every root frame, every
component in the assets panel and the interactions layer re-rendered
on every store change, mouse moves included. Compare by value instead.
This does not loop by itself, but it adds work to every commit and
brings other update loops closer to React's limit.
Relates to #12580
AI-assisted-by: claude-opus-5-5
* ✨ Report update loops throttled by the update scheduler
When the scheduler throttles an update loop, it only logs a console
warning, so a loop that no longer crashes leaves no trace. Send an
error report too, with the events in flight and the last events, at
most once per report window.
Relates to #12580
AI-assisted-by: claude-opus-5-5
* 🐛 Report throttled update loops as unhandled exceptions
The throttled update loop report used the handled-exception event, which
only lands in the audit log: error reports and alerts read
unhandled-exception and exception-page. A throttled loop is an
application defect (its cause has :type :internal), and the point of the
report is to find those loops, so it is now an unhandled exception.
AI-assisted-by: claude-opus-5-5
* 🐛 Report raw fetch failures as connectivity errors
A fetch that fails outside app.util.http (zip.js, plugins, schema
loads) rejects with a bare TypeError such as "Failed to fetch". It
reached the global handlers and was reported as an unhandled crash,
with no stack and no URL.
The global handlers now wrap the browser's network TypeErrors as
:network (or :offline when the browser says so), so they get the
connection toast and a compact handled report. Response body reads
in app.util.http also wrap a TypeError as :network with the request
uri, as fetch itself already did.
Relates to #12581
AI-assisted-by: claude-opus-5-5
* 🐛 Treat only network TypeErrors on body read as connectivity errors
process-response-type turned every TypeError from a body read into a
:network error, so a bug such as "body stream already read" showed a
connection toast. It now uses network-failure?, which moves to
app.util.http so that both the http client and the global error
handlers share it. Add Firefox's "Error in body stream" message.
Tests now check :offline and :network separately by setting
navigator.onLine, and read-error builds on async/->promise.
AI-assisted-by: claude-opus-5-5
* ♻️ Take the command limits before the root ones in the RPC limiter
The RPC concurrency limiter took the root limiters (the ones every request
passes through) before the ones a command declares, so a request waiting for
the limits of its own command held the root permits of its profile while it
waited: a profile with a pile-up of heavy requests saw its other requests
queue behind them and be rejected with 429.
That order was forced by two commands that entered the limiter from inside
their handler: a command that declares no limits of its own can only take
them deeper than the root ones, so two paths of one profile could take the
same limiters in opposite order and park for good.
The commands declare their limits in their metadata now and their handlers
call the work directly, so no command enters the limiter from inside its
handler and every path takes the limiters in the same order. The root
limiters are appended to the sequence first, so the limits a command
declares end up outermost and are taken first: a request waiting for them
does not hold the root permits of its profile. The photo of a team keeps its
limits, because it declares them now that the helper does not.
A call from inside a handler can only reenter the limits that thread
already holds, and it is refused when it asks for a new one, because that
would take it after the root ones and invert the order against the other
paths. The key of a limit is resolved the same way in every path, so both
paths point at the same limiter and the reentrancy check sees it.
Every entry of the shipped climit.edn declares a queue, so the number of
requests waiting for the same limit is bounded instead of piling up until
the timeout. The production file lives outside the repository and has to be
mirrored to pick them up.
The tests cover the acquisition order of a command with and without limits
of its own, the order of the limits as a whole (a ring of any size fails,
not only a pair in opposite order), the nested call that asks for a limit
the thread does not hold, and the queues and timeouts of the shipped
configuration. The import of a binfile declares its request as multipart in
its test, the way the commands that take an upload map do.
AI-assisted-by: minimax-m3.1-flash
AI-assisted-by: deepseek-v4.1-flash
* 🔧 Keep the commands' limits and the shipped config in sync
`profile-plugin-ops` guarded only an UPDATE on the caller's own profile row,
which the row lock already serializes, so its global cap could only reject a
burst of different profiles toggling plugins. Both commands declare the
by-profile limit alone now, and the example config drops the entry.
A declared id without an entry does not limit anything: `wrap` logs a warning
and runs the handler with no limiter at all, which is how a command ends up
unbounded in production. A test fails now when a command declares an id the
config does not list, and the header of the example says the production file
mirrors the shape and lists every declared id, not the values.
AI-assisted-by: deepseek-v4.1-flash
* 📚 Record that background work is not waited for with a sleep
A long command moved to the background reports its own completion, so
polling it with a sleep inside a foreground command only blocks the session
and the notification arrives anyway. The rule says to keep working on
something else, or to end the response and wait for it.
AI-assisted-by: minimax-m3.1-flash
Support currently fixes wrong profile emails and false bounce marks with
raw SQL. Add update-profile-email! (rejects collisions) and
clear-email-complaint-reports! (deletes global and profile reports and
unmutes the matching profile) so these ops run through the REPL.
Closes#12663
The Nextcloud link to the enterprise migration guide no longer works.
Point it to the new share.
The box also had a bold "Open the full guide (PDF)" line that looked
clickable but was not, with the real link below it. Keep a single
link and mark it as a PDF at the end of the text.
AI-assisted-by: claude-opus-5-5
The full `git describe` output is not stable across checkouts: the
length of its -g<sha> suffix depends on the local object count, so
build-bundle and build-docker could compute different keys for the same
commit on different runner workspaces. The sha already pins the commit,
so the closest tag is enough to tell versions apart.
Signed-off-by: David Barragán Merino <david.barragan@kaleidos.net>
The bundle stamps the version into version.txt, so a commit that gets a
new tag (2.18.1-RC1 -> 2.18.1) needs a new bundle and image set. Keying
them by commit sha alone reused the RC artifacts for the final release.
Signed-off-by: David Barragán Merino <david.barragan@kaleidos.net>
* 🐛 Remap library refs inside deleted components on absorb
Deleted components keep their shapes in :objects, to be restored and
to resolve their remaining copies. When a library was absorbed into a
file (export with libraries merged, unsharing or deleting the
library), absorb-assets only looked for usages in pages and in live
components. The shapes inside deleted components kept pointing to the
library, so they no longer matched their copies in the pages: the file
failed validation with :component-id-mismatch and could not be saved.
Also look for usages in deleted components, so their library
components, colors and typographies are remapped to the local file.
Relates to #10806
AI-assisted-by: claude-opus-5-5
* 🐛 Persist pointer-map fragments when absorbing a library
absorb-library-by-file saved the file without its :features. Without
"fdata/pointer-map" there, encode-file does not persist the changed
fragments, so in files with pointer maps the pages and components
that absorb-assets remapped were never written. After unpublishing or
deleting a library, its consumers failed to load with "fragment not
found".
Pass :features to update-file!. Exporting with shared libraries does
not go through this path, so it was not affected.
AI-assisted-by: claude-opus-5-5
Setting a library as tokens source did not update shapes that use tokens,
because set-tokens-source never ran the token propagation that every other
token operation runs. Run it after the commit.
Going back to the local tokens, or unlinking the library that was the
source, left no local set active: the file has a single tokens status, and
switching to a library replaced it with the library's one. Keep the status
of the file's own tokens lib aside in a new optional :own-tokens-status
attribute while an external library is the source, and restore it when the
file is its own source again. Exporting without libraries and absorbing a
deleted source library restore it as well.
Closes#12430
AI-assisted-by: claude-opus-5-5
The RPC layer emitted the event as soon as the handler returned the
SSE response, before the import ran, so the event could not say which
files the import created. Emit it from inside the SSE handler once the
import ends, with the ids and count of the created files and whether
the import completed or failed.
Closes#12485
AI-assisted-by: claude-opus-5-5
LibraryComponent.restore() did not check that the component belongs to
the current file's library. Called on a deleted component of a connected
shared library, it restored that component and saved the change to the
library file, while remove(), annotation and duplicate() reject such
handles with a validation error.
restore() now rejects them the same way: "Can only restore components
from the current file's library".
Closes#12607
* ✨ Add new slides content for 2.19 release
* 🎉 Update user settings instructions in v2_19.cljs
Signed-off-by: Juan de la Cruz <delacruzgarciajuan@gmail.com>
---------
Signed-off-by: Juan de la Cruz <delacruzgarciajuan@gmail.com>
Fixes the problems found after #12326 and listed in #12327.
Mousetrap now tries several names for each key press, in order, and
fires the first one with a binding: letters and digits by keyCode when
a modifier is held, punctuation by position (layout map, or a US table
when the browser has none), the typed glyph, and the typed glyph
without the Shift it needed. With AltGr the typed glyph comes before
the position. The shortcut recorder asks Mousetrap for the key it
stores, so a recorded shortcut always fires.
This makes "?" and "+" work on every layout, Shift and Ctrl+Shift with
punctuation work in Firefox and Safari, Option+letter work on macOS,
and AltGr+º hide the UI on Linux with a Spanish layout.
Chrome on Linux returns the map of the first configured layout, not
the active one. Mousetrap now drops the map when an unmodified
punctuation key types something else.
The layout twins are no longer needed and are removed: "shift+'" made
"?" zoom to the selection on Spanish layouts. The Add node shortcut is
removed too: on US layouts its key is the "+" zoom key, and on Firefox
with a Spanish layout it cannot fire. The Firefox and Safari warning
in the shortcuts settings is gone, since those shortcuts now work
there. The user guide and translations follow these changes.
AI-assisted-by: claude-opus-5-5
The backport of docker/ from develop (2df5f417a) also brought the subpath
support from #12358 and its fix (95bc48a08). It needs the Admin Console
from 2.20, so with the current one every Admin Console URL fails. Restore
the previous nginx configuration and docker-compose.yaml.
AI-assisted-by: Claude
Signed-off-by: David Barragán Merino <david.barragan@kaleidos.net>
* 🚑 Drop invoke_* glue checks from staging CI
The .github backport from develop added check-no-invoke-v to WASM and
Exporter CI, but staging still ships glue that uses invoke_* wrappers.
Remove those steps (and the WASM production build only used by the
check) so staging CI matches the binary it builds.
* 🔧 Disable credential persistence in commit-checker checkout
zizmor fails CI: Workflows when actions/checkout omits
persist-credentials: false. Match the rest of the workflows.
* 🐛 Fix restrict per-side stroke width tokens to rectangles in plugins API
* ♻️ Reuse per-side stroke restriction check across plugin API
Move the per-side stroke width restriction into the public
`per-side-stroke-attrs-restricted?` helper, now taking `page-id`, and reuse
it from both `shape.applyToken` and the token application path. Keep the
error message in a single `per-side-stroke-widths-message` def used by all
three rejection sites.
Wait for token resolution in the uniform stroke width regression test
instead of relying on a bare `settle`, which could assert before the
applied token lands.
AI-assisted-by: deepseek-v4.1-flash
---------
Co-authored-by: Luis de Dios <luis.dedios@kaieldos.net>
Text image shaders are baked into Skia paragraphs at layout time.
touch_shapes_with_image only scanned shape.fills/strokes, so text
spans were never invalidated after store_image. Detect text content
image fills and force a layout rebuild before touching tiles.
Closes#12563
* 🐛 Skip unknown entries when importing shortcuts
The shortcuts import rejected the whole file when it held a single entry
it did not know, so a profile exported from an older Penpot version, or
edited by hand, could not be imported at all. The spec in #9924 asks to
skip unknown keys, tell the user which ones, and apply the rest.
Importing now keeps the entries that map to a known shortcut with a valid
key combination and skips the rest: unknown contexts, unknown shortcut
keys, and commands that are not a key combination. The import modal
lists the skipped entries. The file is only rejected when it is not a
map or holds nothing that can be applied.
A context whose entries are all skipped is left out: an empty context
clears the user's shortcuts on import, so keeping it would wipe them.
valid-command? reads a command the way mousetrap does, so it accepts
every default shortcut and rejects strings mousetrap could never match.
AI-assisted-by: claude-opus-5-5
* 📎 Sort requires in shortcuts settings test
cljfmt expects the namespace requires in alphabetical order.
AI-assisted-by: claude-opus-5-5
* ✨ Reserve cancel, copy and paste shortcuts
Cancel, copy and paste are no longer customizable, in the workspace and
in the path editor, which shares their shortcut keys. Their key
combinations (Escape, Ctrl/Cmd+C, Ctrl/Cmd+V) are now reserved: the
shortcut recorder warns and blocks saving one of them for another
action, and the import skips entries that use them.
Reserved combinations come from the non-customizable workspace
shortcuts, so marking another one as non-customizable reserves its key
too. The path editor's cancel also binds V, the move tool key, so path
shortcuts are left out. Combinations compare the way mousetrap matches
them: aliases and modifier order do not matter.
apply-custom-overrides now ignores overrides of non-customizable
shortcuts. An override saved before this change would otherwise still
apply, with no way to reset it from the settings page.
AI-assisted-by: claude-opus-5-5
* 🐛 Pass version tag to worker so it loads the current render wasm
* 🐛 Keep wasm thumbnail queue alive when one render fails
* 🐛 Fail wasm thumbnail request when canvas context cannot init
* ✨ Support running several exporter replicas
* 📚 Document one exporter per machine instead of compose replicas
* 📚 Trim exporter replica docs to the essentials
* ✨ Cancel an exporter's jobs when it shuts down
* 📚 Replace exporter replica docs with a short beta note
