The frontend build now emits the sha256 hashes of the inline scripts of every page it writes into resources/public, the image moves them out of the document root, and the entrypoint splices them into the default script-src. This removes one of the two reasons why enforcing mode was not usable.
The hashes are computed on the rendered output rather than on the mustache templates, since the digest covers the exact bytes served between the script tags. All four served pages contribute, not just index.html: challenge.html handles the redirect, render.html is loaded by the exporter in a headless browser, and rasterizer.html is initialised by the frontend itself, so leaving any of them out would have broken those paths under enforcing mode. The storybook previews are excluded because that container does not serve them.
A bundle predating this change yields no hashes and the policy stays as it was, so older bundles keep building.
The three external locations were also passing through the security headers of their upstreams. raw.githubusercontent.com returns its own Content-Security-Policy and both it and fonts.googleapis.com return Strict-Transport-Security. Browsers enforce the intersection of every policy they receive, so the upstream one takes precedence on those responses, and the HSTS one lands on our own host, meaning a deployment that deliberately disables HSTS would get it set anyway by a third party. Hide all three at the proxy.
Signed-off-by: David Barragán Merino <david.barragan@kaleidos.net>
The bundle and docker-image build/dedup checks used different cache
keys: the bundle was cached by ref name (`penpot-<gh_ref>.zip`) while
the docker image marker was cached by commit sha
(`markers/images-sha-<sha>`). A tag built from a commit already
promoted under another ref (e.g. `develop`) would rebuild the bundle
unnecessarily, while `build-docker`'s `promote` job silently inherited
the skip from `build` and never created that ref's branch tags
(`backend:<gh_ref>`, `frontend:<gh_ref>`, ...), even though the
underlying sha-tagged images already existed.
- Key the bundle S3 object by commit sha (`penpot-sha-<sha>.zip`)
instead of by ref name, matching the docker marker's semantics.
- Drop the S3 metadata round-trip for `bundle_version` in
build-docker.yml; compute it locally with `git describe`, same as
build-bundle.yml (requires fetch-depth: 0 on that checkout).
- Split `promote` into two mutually-exclusive jobs, `promote` (needs
`build` to succeed) and `retag` (needs only `prepare`, runs when
`prepare.outputs.exists == 'true'`), each moving the `:<gh_ref>`
branch tags to the current sha. This replaces relying on `build`'s
skip/success state with two explicit conditions, so the tags always
get moved regardless of which path built the images.
Signed-off-by: David Barragán Merino <david.barragan@kaleidos.net>
The bundle and docker-image build/dedup checks used different cache
keys: the bundle was cached by ref name (`penpot-<gh_ref>.zip`) while
the docker image marker was cached by commit sha
(`markers/images-sha-<sha>`). A tag built from a commit already
promoted under another ref (e.g. `develop`) would rebuild the bundle
unnecessarily, while `build-docker`'s `promote` job silently inherited
the skip from `build` and never created that ref's branch tags
(`backend:<gh_ref>`, `frontend:<gh_ref>`, ...), even though the
underlying sha-tagged images already existed.
- Key the bundle S3 object by commit sha (`penpot-sha-<sha>.zip`)
instead of by ref name, matching the docker marker's semantics.
- Drop the S3 metadata round-trip for `bundle_version` in
build-docker.yml; compute it locally with `git describe`, same as
build-bundle.yml (requires fetch-depth: 0 on that checkout).
- Split `promote` into two mutually-exclusive jobs, `promote` (needs
`build` to succeed) and `retag` (needs only `prepare`, runs when
`prepare.outputs.exists == 'true'`), each moving the `:<gh_ref>`
branch tags to the current sha. This replaces relying on `build`'s
skip/success state with two explicit conditions, so the tags always
get moved regardless of which path built the images.
Signed-off-by: David Barragán Merino <david.barragan@kaleidos.net>
Two entries in THANKYOU.md had the first part of the Weblate username
pasted in front of the URL (pablo.https://... and swapnil.https://...),
which makes GitHub render them as broken relative links. Point them at
the real profiles, https://hosted.weblate.org/user/pablo.alba and
https://hosted.weblate.org/user/swapnil.cx, which both exist.
Signed-off-by: Alma Faris <almazaf19@gmail.com>
* 🐛 Fix shadows on a masked group in the WASM renderer
A masked group renders in two passes: its content, then the mask shape
composited with DstIn so everything outside the mask silhouette is
erased. Both happen inside one save_layer, and the group's drop shadow
was composited into that same layer before the mask pass — so the mask
erased it. A drop shadow lives mostly outside the silhouette, so it
disappeared entirely.
Inner shadows never drew at all: render_fill_inner_shadows needs fill
geometry to paint into, and a group has none.
Both now ride on an image filter set on the masked-group layer, which
Skia evaluates after the mask is composited, so the shadow comes from
the real masked pixels rather than the group's own, empty geometry.
The effects compose in the order the SVG renderer uses for a group:
drop shadows, then the source, then inner shadows, with the layer blur
over all of it.
That layer is opened on a canvas carrying no transform, so the filter
is built in device units. Shadow::scale_to_device does that rather
than scale_content, because radius_to_sigma is affine: scaling the
radius applies its constant term once at device scale, while a filter
built in document space has the term scaled by the canvas matrix. The
two would blur differently by 0.5 · (scale - 1) sigma, visible as a
masked group's own shadow being narrower than the same shadow on its
parent. The masked-group layer blur had the same flaw.
Three paths are suppressed for masked groups so nothing is drawn twice:
the silhouette composite, the nested_shadows inheritance that would
reach text descendants, and the fill inner-shadow pass.
Every save and restore around that layer is keyed on the shape alone.
Enter and exit run on different walker passes, and a pan or zoom in
between changes fast mode, so deriving them from render state could
leave the canvas clip stack unbalanced.
Refs #11697
AI-assisted-by: claude-opus-5
* 🐛 Fix a container's drop shadow over a masked group in WASM
A container builds its drop shadow by drawing each descendant as a
black silhouette and blurring the result. The walk descends only
through children that can be flattened, and a masked group never can,
so it stopped there and asked the group to draw its own geometry. A
group has none, so nothing was drawn and the shadow layer stayed
blank: no shadow at all for a group, board or frame holding a masked
group. This is what the file attached to the issue reproduces.
render_drop_black_shadow now draws the masked silhouette for such a
group — content children flat black, DstIn the mask, and only then the
offset, blur and spread. Masking after the blur would trim the shadow
along the wrong edge.
The walk recurses, so it narrows the clip the way the main walker
does: content a clipping container hides must not widen the shadow.
The clip rule now lives in one place, shared with the walker, and a
test pins the two against each other. The shadow layer is sized to the
silhouette plus the shadow's reach rather than falling back to the
clip, so a wide blur is not cut at the tile edge.
Spread keeps the renderer's existing behaviour: the silhouette goes
through the same get_drop_shadow_filter every other shadow uses, so a
masked group gains no ordering of its own.
Closes#11697
AI-assisted-by: claude-opus-5
* ⚡ Capture canvas snapshot only when the renderer is idle
* ⚡ Skip descendant silhouettes for frames with nested drop shadows
* 🔧 Add a profiling build mode for render-wasm
Review assumed int? was 32-bit; it covers Long/Integer/Short/Byte.
Note it in mem:clojure/idioms so the mistake is not repeated.
AI-assisted-by: muse-spark-1.3-contributor
Migrate zoom-widget-workspace to zoom-widget-workspace* following the
modern rumext component syntax: drop ::mf/wrap-props false and switch
the callsite from [:& ...] to [:> ...]. No behavior change.
Part of #9260
AI-assisted-by: muse-spark-1.3-contributor
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
* 🐛 Reflow auto grid cells on flow direction change
Remap only single-span auto cells to the new
:layout-grid-dir traversal order, keeping source
order, manual and area placements untouched.
Update both grid direction controls to use the
new change-grid-direction event and refresh the
stale active button on persisted direction.
Add a RED-to-GREEN model regression covering a
2x2 row-to-column transition and source-order.
AI-assisted-by: muse-spark
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
* 🐛 Keep source order on grid flow change with areas
Skip the generic grid cell pass for the
direction event, since reflowing already
places every eligible auto item and a blind
reorder rewrites shapes around pinned areas.
Pin area/span grids with a regression test
covering direction change and source order.
AI-assisted-by: muse-spark
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
* 🐛 Scope grid skip to direction changes only
Replace the translation flag with a narrow
skip-grid-reassignment option so component
sync and reflow metadata stay intact while
the generic grid cell pass is skipped.
AI-assisted-by: muse-spark
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
* 🐛 Clear leftover auto cells on grid flow change
Write remapped shapes to every target auto cell and
empty leftover cells so sparse grids cannot duplicate
a child across target cells. Manual, area and
spanned cells stay untouched; source order is kept.
AI-assisted-by: muse-spark
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
* 🐛 Pin grid flow invariants with span and manual regressions
Keep the direction-change design unchanged and lock the
claimed invariants with tests: a real 2x1 manual span
cell and an occupied manual cell stay byte-identical,
row->column->row round-trips to the original cells,
and a mixed auto/manual/span/area grid shows no shape
loss or duplication. Also drop the unused page-objects
binding from the direction-change watcher.
AI-assisted-by: muse-spark
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
* 🐛 Unoverlap mixed grid fixture and assert movement
Move auto C to (1,3) so it no longer overlaps the 2x1
manual span at (1,2). Row auto order A,C,B,E becomes
column order A,B,E,C; assert the exact placement
while keeping pinned, source-order and no-loss
checks. Test-only change.
AI-assisted-by: muse-spark
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
* 🐛 Unify plugin dir setter and normalize missing direction
Route GridLayoutProxy.dir through change-grid-direction so the
plugin API shares the UI direction-change path with its reflow
and source-order guarantees. Normalize a missing
:layout-grid-dir to :row at the change-grid-direction entry
point and cover it with a missing-direction regression plus a
plugin setter routing regression.
AI-assisted-by: muse-spark
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
* 🐛 Fix grid plugin dir setter syntax
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
AI-assisted-by: opencode-go/muse-spark-1.3-contributor
* 🐛 Fix comments and tests
---------
Signed-off-by: makesomethingshit <junsoo1172@gmail.com>
Co-authored-by: alonso.torres <alonso.torres@kaleidos.net>
Give every plan a write-restricted Status (draft, reviewed,
done) and an append-only Review Log with UTC ISO 8601 lines.
Make-a-plan creates plans as draft and is the only flow
writing reviewed, on explicit user apply. Review-plan stays
read-only. Implement-plan closes the plan to done with the
issue URL when one exists, in the same commit as the code.
Document the lifecycle in the agents README.
AI-assisted-by: muse-spark-1.3-contributor
Define derived plan naming for .agents/plans/.
Parent basename stays intact and derivatives append
--review-NN for review followups and --task-NN for
roadmap sub-plans, with no new date so ls groups them.
Document the rule in the planner skill, the in-place
vs new-file policy in make-a-plan, and examples in
the agents README.
AI-assisted-by: muse-spark-1.3-contributor
Packed atlas compose inset Linear samples after blitting the full
drawable into the slot, which dropped edge texels and opened
multi-pixel gaps on text that crossed tile seams (#11696), especially
under HiDPI packing. Write the drawable into the inset content rect
and clamp-pad the 1px frame so Linear compose keeps coverage without
bleeding into the next cell.
Closes#11696
Clipped frames (show-content=false) were applying clipPath to the drop
silhouette. Spread/offset fills were outset correctly but then truncated
to the true selrect, so spread rings vanished on export. Match the GPU:
paint drop silhouettes outside the content clip; keep clipping on the
real content pass only.
Closes#11653
SkSVGDevice drops save_layer composition for text strokes, so emit
center strokes (with <g opacity> when alpha), inner strokes via glyph
clipPath, and outer strokes via inverse-glyph luminance mask.
Closes#11386
* 🐛 Fix stroke to path dropping caps and markers
Caps and markers (round ends, arrows, square/diamond/circle markers) are
painted separately by `handle_stroke_caps` after the stroke itself, so
`stroke_to_path` — which only outlines the shape path — dropped them.
Converting a stroke to a path lost every end decoration except the
Round/Round and Square/Square pairs Skia draws natively, and even those
were missing because the outline paint kept the default butt cap.
Cap geometry now lives in `shapes/stroke_paths.rs` as plain path
builders, shared by the canvas renderer and by `stroke_to_path`, which
unions the caps into the outline for open paths and honors
`to_skia_linecap`. The SVG export no longer overlays the caps a second
time on top of the expanded outline, which would double the alpha of
translucent strokes; the clip silhouette for image strokes gets them
from the outline too.
Co-authored-by: Shreyash Agare <agareshreyash26@gmail.com>
Add the verified REST procedure for linking an issue as a
sub-issue of an umbrella/EPIC: get the REST id, POST to the
parent's sub_issues endpoint with a typed -F field, and verify
both directions. Route it from the create-issue skill.
AI-assisted-by: deepseek-v4.1-flash
