Carry existing colour across a bake instead of reindexing it

Baking, painting colour, then baking again left colour debris on surfaces
nothing had painted and destroyed what the earlier bake had laid down.

TextureDisplacementBakeJob::finalize() deserialized the volume's pre-bake
mmu_segmentation_facets into a TriangleSelector built on the post-bake mesh,
on the assumption that the bake is topology-preserving. That holds for the
classic path, which moves vertices and leaves the topology alone, but not for
the one-run pipeline, which rebuilds the mesh. TriangleSelector::deserialize()
only checks that the stored triangle ids are in range, and a bake almost
always grows the triangle count, so every stale id passed that check and
landed on an unrelated face.

Measured on a 26 mm slab baked twice at 0.19 mm with a 750 k budget: 7580
triangles ended up carrying paint neither bake produced, as far as 31 mm from
any colour that was actually laid down, while the first pass's colour was lost
where the second bake left the surface alone.

TriangleSelector::remap_painting() already transfers paint between meshes by
position, and ModelVolume::restore_painting() uses it for the same reason, so
the bake job now uses it too. It runs in process(), on the worker thread: it
builds an AABB tree over both meshes and took around a second on the case
above, which would have frozen the UI in finalize(). The bake's own colour is
still applied last, so it wins where it painted and the rest is kept, which is
what makes successive bakes accumulate rather than overwrite. If the volume is
painted while the bake is running, a timestamp check falls back to the live
mask so that stroke is not dropped.

The out_triangle comment in TextureColorRequest claimed the bake is
topology-preserving, which is what justified the original indexing; it now
says which mesh the colours belong to.

The regression test bakes twice, checks both results are closed, and checks
the first bake's colour survives the second at the positions where it was
laid down and nowhere else.

.claude/skills/orca-wxwidgets/references/platforms.md

@@ -569,16 +569,12 @@ on the **first** call. So:
 #endif
 ```
 
-**Before GTK starts**, only the environment exists. `CLI::run` (`src/OrcaSlicer.cpp`) sets the NVIDIA
-DMABUF workaround and then branches on `GDK_BACKEND` (an `x11` prefix means the X11 opt-in). On both
-paths, when `/proc/driver/nvidia/version` exists and the linked WebKitGTK honours the switch (>= 2.44,
-`WebView::SupportsForceSharedMemory()`), it sets `WEBKIT_DMABUF_RENDERER_FORCE_SHM=1` and
-`WEBKIT_FORCE_DMABUF_RENDERER=1` (non-replacing): the proprietary driver cannot back WebKit's hardware
-DMABUF renderer. On the default path it forces `GDK_BACKEND=x11` when wx lacks EGL and `WAYLAND_DISPLAY`
-is set. It sets `WEBKIT_DISABLE_COMPOSITING_MODE=1` (non-replacing) only when both `DISPLAY` and
-`WAYLAND_DISPLAY` are set and the linked WebKitGTK is older than 2.42, and calls `XInitThreads()`
+**Before GTK starts**, only the environment exists. `CLI::run` (`src/OrcaSlicer.cpp`) branches on
+`GDK_BACKEND` (an `x11` prefix means the X11 opt-in). On the default path it forces `GDK_BACKEND=x11`
+when wx lacks EGL and `WAYLAND_DISPLAY` is set. It sets `WEBKIT_DISABLE_COMPOSITING_MODE=1`
+(non-replacing) only when both `DISPLAY` and `WAYLAND_DISPLAY` are set, and calls `XInitThreads()`
 only when `DISPLAY` is set. With neither variable set, the GUI refuses to start ("Neither DISPLAY nor
-WAYLAND_DISPLAY set"). The details and the WebKit rules (c12912e0df) are in
+WAYLAND_DISPLAY set"). The details and the WebKit rule (c12912e0df) are in
 `references/webview-gl-aui-media.md` §WebKitGTK on Linux sessions.
 
 **wx's own detectors**, for reference:

.claude/skills/orca-wxwidgets/references/webview-gl-aui-media.md

@@ -61,8 +61,7 @@ available; GTK2/WebKit1 is an opt-out configuration the GUI does not support.
 11. A web host that re-themes in place handles `EVT_WEBVIEW_RECREATED` without `Skip()`; one that
     needs a reload lets it through. §[Orca wrapper](#orca-webview-wrapper-widgetswebview)
 12. Keep `WebViewWebKit`'s destructor removing the `"wx"` handler. §[Orca wrapper](#orca-webview-wrapper-widgetswebview)
-13. Do not widen `WEBKIT_DISABLE_COMPOSITING_MODE` beyond Orca's default-path XWayland case for
-    WebKitGTK older than 2.42, and never
+13. Do not widen `WEBKIT_DISABLE_COMPOSITING_MODE` beyond Orca's default-path XWayland case, and never
     make a page's progress depend only on a C++→JS callback. §[WebKitGTK on Linux](#webkitgtk-on-linux-sessions)
 14. GL attribute lists: legacy `int[]` lists end with `0` and spell out `WX_GL_RGBA` and
     `WX_GL_DOUBLEBUFFER`; `wxGLAttributes`/`wxGLContextAttrs` end with `EndList()`; MSAA is requested
@@ -568,66 +567,33 @@ vue-resize workaround script and sets a persistent SQLite cookie store on the vi
 ## WebKitGTK on Linux sessions
 
 **Mechanism.** `CLI::run` (`src/OrcaSlicer.cpp`, under `#ifdef __WXGTK__`, before GTK initialises)
-sets the NVIDIA DMABUF workaround and then branches on `GDK_BACKEND`:
-- **Both paths**, before the split: when `/proc/driver/nvidia/version` exists and the linked WebKitGTK
-  honours the switch (>= 2.44, `WebView::SupportsForceSharedMemory()`), it sets
-  `WEBKIT_DMABUF_RENDERER_FORCE_SHM=1` and `WEBKIT_FORCE_DMABUF_RENDERER=1` non-replacing. The
-  proprietary driver cannot back WebKit's hardware DMABUF renderer, so the shared-memory transport keeps
-  accelerated compositing instead of dropping it; `WEBKIT_FORCE_DMABUF_RENDERER` clears the
-  Debian/Ubuntu/openSUSE `isNVIDIA()` early return, which otherwise returns before the FORCE_SHM check.
+branches on `GDK_BACKEND`:
 - **Default path** (`GDK_BACKEND` unset or not starting with `x11`): if wx lacks EGL (`!wxHAS_EGL`) and
   `WAYLAND_DISPLAY` is set, it forces `GDK_BACKEND=x11`; it sets `WEBKIT_DISABLE_COMPOSITING_MODE=1`
-  non-replacing **only** when the linked WebKitGTK predates 2.42 (`WebView::NeedsCompositingWorkaround()`)
-  **and** both `DISPLAY` and `WAYLAND_DISPLAY` are non-empty (a Wayland session with
+  non-replacing **only** when both `DISPLAY` and `WAYLAND_DISPLAY` are non-empty (a Wayland session with
   XWayland available — the session, not the backend GTK will pick); `XInitThreads()` runs only when
   `DISPLAY` is set.
 - **`GDK_BACKEND=x11…` opt-in path**: sets `DRI_PRIME=1` (and the NVIDIA PRIME variables when
   `/proc/driver/nvidia/version` exists), calls `XInitThreads()`, and deliberately does **not** disable
-  compositing, which keeps hardware acceleration for Device / Setup Wizard / login pages.
+  compositing (the comment: the WebKit2GTK XWayland bug appears fixed in WebKit2GTK ≥ 2.42, and leaving
+  compositing on keeps hardware acceleration for Device / Setup Wizard / login pages).
 
 Runtime backend detection elsewhere uses Orca's `is_running_on_wayland()` / `is_running_on_x11()`
 (see `references/platforms.md`), not these environment variables.
 
 **Pitfall.**
-- **Rule:** On NVIDIA reach for `WEBKIT_DMABUF_RENDERER_FORCE_SHM` before
-  `WEBKIT_DISABLE_COMPOSITING_MODE`. **Why:** the proprietary driver cannot back WebKit's hardware
-  DMABUF renderer, leaving web views blank or slow; FORCE_SHM keeps accelerated compositing and only
-  skips the hardware path, while disabling compositing drops it. Threaded scrolling is a separate,
-  release-dependent default (off in 2.44/2.45, on from 2.46). `WEBKIT_DISABLE_DMABUF_RENDERER` is not a
-  substitute — it is read first, leaves the transport set empty (no compositing from 2.44 on, where the
-  X11/Wayland fallback renderers were removed) and has a reported crash path. On Debian/Ubuntu and
-  openSUSE the `isNVIDIA()` patch returns before the FORCE_SHM check, so `WEBKIT_FORCE_DMABUF_RENDERER=1`
-  must clear it first; that variable exists only in those patches. WebKit reads all of these once per
-  process, before the first WebView.
-  The test is `/proc/driver/nvidia/version`, so a hybrid laptop that renders on its iGPU also takes the
-  shared-memory path, which is a per-frame GPU readback rather than zero-copy; exporting
-  `WEBKIT_DMABUF_RENDERER_FORCE_SHM=0` in the environment restores the hardware path.
-  ```cpp
-  // Wrong: kills compositing and threaded scrolling (and can crash on modern WebKit)
-  ::setenv("WEBKIT_DISABLE_DMABUF_RENDERER", "1", false);
-  // Right: keep compositing, skip only the hardware DMABUF path
-  if (::access("/proc/driver/nvidia/version", F_OK) == 0 && WebView::SupportsForceSharedMemory()) {
-      ::setenv("WEBKIT_DMABUF_RENDERER_FORCE_SHM", "1", /*replace*/ false);
-      ::setenv("WEBKIT_FORCE_DMABUF_RENDERER", "1", /*replace*/ false);
-  }
-  ```
-  Cite: WebKit 2.52.6 `Source/WebKit/UIProcess/gtk/AcceleratedBackingStore.cpp`
-  (`rendererBufferTransportMode()`), Debian `disable-nvidia-dmabuf.patch`, WebKit bug 262607.
 - **Rule:** Set `WEBKIT_DISABLE_COMPOSITING_MODE` only on the default path in Wayland sessions,
   non-replacing (`setenv(..., false)`, so a value the user already exported wins), before GTK
-  initialises; never unconditionally, and only for the WebKitGTK releases that need it (older than
-  2.42, `WebView::NeedsCompositingWorkaround()`). Keep JS-side fallbacks so a page never
+  initialises; never unconditionally. Keep JS-side fallbacks so a page never
   waits forever for a C++→JS call.
-  **Why:** WebKit2GTK accelerated compositing could fail silently under XWayland before 2.42, leaving
-  web views blank (Setup Wizard stuck on "Loading..."); the variable also turns off threaded scrolling,
-  so applying it on a newer WebKit needlessly drops both GPU compositing and async scrolling (the
-  Device tab loses its smoothness); without the JS timeout the user is stuck when the signal is lost.
+  **Why:** WebKit2GTK accelerated compositing could fail silently under XWayland, leaving web views
+  blank (Setup Wizard stuck on "Loading..."); disabling it everywhere needlessly drops hardware
+  acceleration; without the JS timeout the user is stuck when the signal is lost.
   ```cpp
   // Wrong: unconditional, kills HW acceleration everywhere
   ::setenv("WEBKIT_DISABLE_COMPOSITING_MODE", "1", false);
-  // Right (default path only): non-replacing, before GTK init, WebKitGTK < 2.42,
-  // Wayland session with XWayland
-  if (WebView::NeedsCompositingWorkaround() && display && *display && wayland && *wayland)
+  // Right (default path only): non-replacing, before GTK init, Wayland session with XWayland
+  if (display && *display && wayland && *wayland)
       ::setenv("WEBKIT_DISABLE_COMPOSITING_MODE", "1", /*replace*/ false);
   ```
   Cite: c12912e0df (`src/OrcaSlicer.cpp`; `resources/web/guide/0/load.js` `OnInit`

localization/i18n/zh_CN/OrcaSlicer_zh_CN.po

@@ -7478,7 +7478,7 @@ msgid "Update"
 msgstr "更新"
 
 msgid "Assistant(HMS)"
-msgstr "助理（HMS）"
+msgstr "助手(HMS)"
 
 #, c-format, boost-format
 msgid "Network plug-in v%s"

scripts/flatpak/entrypoint

@@ -1,7 +1,7 @@
 #!/usr/bin/env sh
 
-# WebKit's NVIDIA DMABUF workaround (gnome-build-meta#754) is applied by the
-# binary itself, gated on the runtime WebKitGTK version (src/OrcaSlicer.cpp).
+# Work-around https://gitlab.gnome.org/GNOME/gnome-build-meta/-/issues/754
+grep -q org.freedesktop.Platform.GL.nvidia /.flatpak-info && export WEBKIT_DISABLE_DMABUF_RENDERER=1
 
 # Work-around https://github.com/bambulab/BambuStudio/issues/3440
 # Use LC_NUMERIC instead of LC_ALL to prevent decimal separator issues

src/OrcaSlicer.cpp

@@ -161,7 +161,6 @@ using json = nlohmann::json;
 #include <X11/Xlib.h>
 #endif
 #include <unistd.h>
-#include "slic3r/GUI/Widgets/WebView.hpp"
 #endif
 
 #ifdef SLIC3R_GUI
@@ -1154,9 +1153,8 @@ int CLI::run(int argc, char **argv)
     // opt-in path. The ~2020-era WebKit2GTK XWayland compositor bug it
     // worked around appears fixed in WebKit2GTK >= 2.42; leaving it
     // unset preserves WebKit hardware acceleration on Device / Setup
-    // Wizard / login / store. The default path applies it only on
-    // XWayland sessions and only when the linked WebKitGTK predates the
-    // fix (< 2.42), as a conservative fallback for older WebKit.
+    // Wizard / login / store. The default path still applies it on
+    // XWayland sessions as a conservative fallback for older WebKit.
     // ------------------------------------------------------------------
     {
         const char* gdk_backend = ::getenv("GDK_BACKEND");
@@ -1164,14 +1162,6 @@ int CLI::run(int argc, char **argv)
         // honours the first backend in the comma-separated list).
         const bool x11_opt_in = gdk_backend && boost::starts_with(gdk_backend, "x11");
 
-        // NVIDIA's proprietary driver cannot back WebKitGTK's DMABUF renderer.
-        // Force the shared-memory transport, which keeps compositing and threaded
-        // scrolling; the second variable clears Debian/Ubuntu/openSUSE's isNVIDIA().
-        if (::access("/proc/driver/nvidia/version", F_OK) == 0 && WebView::SupportsForceSharedMemory()) {
-            ::setenv("WEBKIT_DMABUF_RENDERER_FORCE_SHM", "1", /* replace */ false);
-            ::setenv("WEBKIT_FORCE_DMABUF_RENDERER", "1", /* replace */ false);
-        }
-
         if (x11_opt_in) {
             // ===== X11 / XWayland (opted in via GDK_BACKEND=x11) =====
             BOOST_LOG_TRIVIAL(info) << "GDK_BACKEND=x11 detected; applying v2.3.2 X11/XWayland workarounds.";
@@ -1215,15 +1205,15 @@ int CLI::run(int argc, char **argv)
             }
             #endif
 
-            // WebKit2GTK accelerated compositing could fail under XWayland
-            // on WebKit releases before 2.42. Disable it only then, and only
-            // when both DISPLAY and WAYLAND_DISPLAY are set (an XWayland
-            // session). Disabling it also turns off threaded scrolling, so on
-            // newer WebKit it only costs performance.
+            // WebKit2GTK compositing can fail under XWayland on older
+            // WebKit releases. Disable it only when both DISPLAY and
+            // WAYLAND_DISPLAY are set (i.e. an XWayland session is in
+            // use). On pure X11 or native Wayland, compositing is left
+            // enabled so WebKit retains HW acceleration.
             {
                 const char* display_env_wk = ::getenv("DISPLAY");
                 const char* wayland_env_wk = ::getenv("WAYLAND_DISPLAY");
-                if (WebView::NeedsCompositingWorkaround() && display_env_wk && *display_env_wk && wayland_env_wk && *wayland_env_wk) {
+                if (display_env_wk && *display_env_wk && wayland_env_wk && *wayland_env_wk) {
                     ::setenv("WEBKIT_DISABLE_COMPOSITING_MODE", "1", /* replace */ false);
                 }
             }

src/dev-utils/platform/unix/build_linux_image.sh.in

@@ -429,8 +429,7 @@ if [ "\$XDG_SESSION_TYPE" = "wayland" ] && [ "\$ZINK_DISABLE_OVERRIDE" != "1" ];
                 export __EGL_VENDOR_LIBRARY_FILENAMES=/usr/share/glvnd/egl_vendor.d/50_mesa.json
                 export MESA_LOADER_DRIVER_OVERRIDE=zink
                 export GALLIUM_DRIVER=zink
-                # WebKit's NVIDIA DMABUF workaround is applied by the binary itself,
-                # gated on the runtime WebKitGTK version (src/OrcaSlicer.cpp).
+                export WEBKIT_DISABLE_DMABUF_RENDERER=1
             fi
         fi
     fi

src/libslic3r/TextureDisplacement.hpp

@@ -771,9 +771,12 @@ struct TextureColorRequest
     // single facets; an image detail a few facets wide still leaves thousands of pinhead islands
     // that the slicer's multi-material segmentation cannot digest. 0 turns it off.
     float           min_color_region_mm2 = 0.5f;
-    // Filled per *base mesh* triangle (the bake is topology-preserving, so this indexes the returned
-    // mesh too): the quantize callback's index plus one, or 0 for "this triangle takes no colour from
-    // the texture". The +1 lines up with EnforcerBlockerType, where 0 is NONE ("use the volume's own
+    // Filled per triangle of the *returned* mesh: the quantize callback's index plus one, or 0 for
+    // "this triangle takes no colour from the texture". That is the same indexing as the base mesh
+    // only on the classic path, which moves vertices and leaves the topology alone; the one-run
+    // pipeline rebuilds the mesh, so nothing recorded against the base mesh - an existing paint mask,
+    // say - may be carried over by triangle id (see TriangleSelector::remap_painting()).
+    // The +1 lines up with EnforcerBlockerType, where 0 is NONE ("use the volume's own
     // filament"): where every palette entry is a filament, these go straight to a TriangleSelector. A
     // palette with mixes maps each index to the mix's filament slot first (see
     // GLGizmoTextureDisplacement::palette_filaments()).

src/slic3r/GUI/Jobs/TextureDisplacementBakeJob.cpp

@@ -13,6 +13,7 @@
 
 #include "libslic3r/Model.hpp"
 #include "libslic3r/format.hpp"
+#include "libslic3r/Point.hpp"
 #include "libslic3r/TriangleSelector.hpp"
 
 #include "slic3r/GUI/GLCanvas3D.hpp"
@@ -76,6 +77,23 @@ void TextureDisplacementBakeJob::process(Ctl &ctl)
         },
         color, m_input.volume_to_world, nullptr, &m_stats));
 
+    // Bring the colour that was already painted onto the mesh this bake just built, while still on
+    // the worker thread. Triangle ids only mean the same thing on both meshes when the bake preserved
+    // the topology - the classic path moves vertices and nothing else - and the one-run pipeline
+    // rebuilds the mesh from scratch. Deserializing the old mask against the new one only checks that
+    // the ids are in range, and a bake almost always *grows* the triangle count, so every stale id
+    // passes that check and lands on an unrelated face: colour debris where nothing was painted, and
+    // the earlier pass's colour destroyed rather than kept. remap_painting() transfers it by position
+    // instead, which is what lets successive bakes accumulate. It is spatial work over both meshes,
+    // which is why it belongs here and not in finalize().
+    if (!m_input.mmu_segmentation.bitstream.empty() && !m_result.empty() && !ctl.was_canceled()) {
+        if (m_result.its.indices.size() == m_input.base_mesh.indices.size())
+            m_carried_mmu = m_input.mmu_segmentation;
+        else
+            m_carried_mmu = TriangleSelector::remap_painting(m_input.base_mesh, m_input.mmu_segmentation,
+                                                             m_result.its, Transform3d::Identity(), {});
+    }
+
     // Always finish at 100: this is what closes the notification. Reported even on cancel, where
     // build_texture_displacement() returns an empty mesh and finalize() commits nothing.
     ctl.update_status(100, status);
@@ -126,6 +144,18 @@ void TextureDisplacementBakeJob::finalize(bool canceled, std::exception_ptr &ept
                                                      { mix.a_percent(), 100 - mix.a_percent() });
             });
 
+        // Read while the volume still holds the mesh the paint was recorded against, and after the
+        // filament slots above - adding one can clamp the paint, and that clamp has to survive.
+        // Normally process() has already carried the colour onto the baked topology; it only has to
+        // be redone here when the paint changed after the job was queued, because the snapshot
+        // process() worked from is then not what the user is looking at.
+        const bool mmu_changed = volume->mmu_segmentation_facets.timestamp() != m_input.mmu_segmentation_timestamp;
+        const indexed_triangle_set                    pre_bake_mesh = mmu_changed ? volume->mesh().its
+                                                                                  : indexed_triangle_set{};
+        const TriangleSelector::TriangleSplittingData live_mmu      = mmu_changed
+                                                                          ? volume->mmu_segmentation_facets.get_data()
+                                                                          : TriangleSelector::TriangleSplittingData{};
+
         volume->set_mesh(std::move(m_result));
         volume->set_new_unique_id();
         volume->calculate_convex_hull();
@@ -134,13 +164,19 @@ void TextureDisplacementBakeJob::finalize(bool canceled, std::exception_ptr &ept
         // rather than replacing it: a triangle the texture does not colour keeps its existing filament,
         // and one the user never painted at all stays at NONE, which already means "the volume's own
         // filament". That is what confines the effect to the painted area without having to invent a
-        // colour for everything outside it. Safe to index straight onto the new mesh - the bake is
-        // topology-preserving, so triangle i is still triangle i.
+        // colour for everything outside it. What was already painted comes from m_carried_mmu, which
+        // process() brought onto this mesh's topology; this bake's own colour goes on top of it, so it
+        // wins wherever it painted and leaves the rest alone.
         if (!m_triangle_color.empty() && m_triangle_color.size() == volume->mesh().its.indices.size()) {
+            if (mmu_changed && !live_mmu.bitstream.empty())
+                m_carried_mmu = pre_bake_mesh.indices.size() == volume->mesh().its.indices.size()
+                                    ? live_mmu
+                                    : TriangleSelector::remap_painting(pre_bake_mesh, live_mmu,
+                                                                       volume->mesh().its,
+                                                                       Transform3d::Identity(), {});
             TriangleSelector selector(volume->mesh());
-            const TriangleSelector::TriangleSplittingData &existing = volume->mmu_segmentation_facets.get_data();
-            if (!existing.bitstream.empty())
-                selector.deserialize(existing, false);
+            if (!m_carried_mmu.bitstream.empty())
+                selector.deserialize(m_carried_mmu, false);
             for (size_t i = 0; i < m_triangle_color.size(); ++i) {
                 const size_t entry = size_t(m_triangle_color[i]);
                 if (entry > 0 && entry <= filament_of.size() && filament_of[entry - 1] >= 0)
@@ -222,6 +258,8 @@ void queue_texture_displacement_bake(const ModelVolume &volume, const TextureCol
     input.layers     = volume.texture_displacement_layers;
     input.options    = volume.texture_displacement_options;
     input.volume_to_world = texture_displacement_volume_to_world(volume);
+    input.mmu_segmentation           = volume.mmu_segmentation_facets.get_data();
+    input.mmu_segmentation_timestamp = volume.mmu_segmentation_facets.timestamp();
     for (int i = 0; i < int(TEXTURE_DISPLACEMENT_MAX_LAYERS); ++i)
         input.facets_data[size_t(i)] = volume.texture_displacement_facet(i).get_data();
 

src/slic3r/GUI/Jobs/TextureDisplacementBakeJob.hpp

@@ -10,6 +10,7 @@
 #include "libslic3r/ObjectID.hpp"
 #include "libslic3r/TextureDisplacement.hpp"
 #include "libslic3r/TriangleMesh.hpp"
+#include "libslic3r/TriangleSelector.hpp"
 
 #include "Job.hpp"
 
@@ -34,6 +35,13 @@ struct TextureDisplacementBakeInput
     // writes the volume's mmu_segmentation_facets - the same per-triangle filament assignment the MMU
     // paint gizmo writes - alongside the displaced geometry.
     TextureColorSettings                  color;
+    // The colour already painted on the volume, indexed against `base_mesh`. Carried here rather than
+    // read at commit time because the one-run pipeline rebuilds the mesh, and bringing this across to
+    // the new topology is spatial work (see process()) that has no business on the UI thread.
+    TriangleSelector::TriangleSplittingData mmu_segmentation;
+    // What `mmu_segmentation` was taken from, so the commit can tell whether the volume was painted
+    // while the bake ran and fall back to the live data if it was.
+    ObjectBase::Timestamp                   mmu_segmentation_timestamp = 0;
     // Whether this job pushes its own undo step when it commits. False when the caller has already
     // taken one that is meant to cover the displacement as well - Standard mode's Bake, which remeshes
     // and subdivides first and has to undo as a single action.
@@ -60,6 +68,9 @@ private:
     // Empty unless a layer asked for colour. See TextureColorRequest; finalize() turns these into
     // filaments.
     std::vector<uint8_t>         m_triangle_color;
+    // m_input.mmu_segmentation brought onto m_result's topology, so finalize() can lay this bake's
+    // colour over the colour that was already there. Indexed against m_result, not base_mesh.
+    TriangleSelector::TriangleSplittingData m_carried_mmu;
     std::function<void()>        m_on_finished;
 };
 

src/slic3r/GUI/Widgets/WebView.cpp

@@ -69,10 +69,6 @@ webkit_web_view_run_javascript_finish                (WebKitWebView
 						      GError                    **error);
 WEBKIT_API void
 webkit_javascript_result_unref              (WebKitJavascriptResult *js_result);
-WEBKIT_API guint
-webkit_get_major_version                     (void);
-WEBKIT_API guint
-webkit_get_minor_version                     (void);
 }
 #endif
 
@@ -405,29 +401,6 @@ bool WebView::NeedsRecreateOnShow()
     return false;
 #endif
 }
-
-bool WebView::NeedsCompositingWorkaround()
-{
-#ifdef __linux__
-    const guint major = webkit_get_major_version();
-    const guint minor = webkit_get_minor_version();
-    return major < 2 || (major == 2 && minor < 42);
-#else
-    return false;
-#endif
-}
-
-bool WebView::SupportsForceSharedMemory()
-{
-#ifdef __linux__
-    const guint major = webkit_get_major_version();
-    const guint minor = webkit_get_minor_version();
-    return major > 2 || (major == 2 && minor >= 44);
-#else
-    return false;
-#endif
-}
-
 #if wxUSE_WEBVIEW_EDGE
 bool WebView::CheckWebViewRuntime()
 {

src/slic3r/GUI/Widgets/WebView.hpp

@@ -30,16 +30,6 @@ public:
     // ignoring every navigation. A panel that gets true here recreates its view on first Show().
     static bool NeedsRecreateOnShow();
 
-    // True when the linked WebKitGTK predates 2.42, where accelerated compositing
-    // could fail silently on XWayland sessions, so it is forced off as a fallback.
-    // False on every other platform. Safe to call before GTK initialises.
-    static bool NeedsCompositingWorkaround();
-
-    // True when the linked WebKitGTK is new enough (>= 2.44) to honour
-    // WEBKIT_DMABUF_RENDERER_FORCE_SHM, the shared-memory transport switch.
-    // False on every other platform. Safe to call before GTK initialises.
-    static bool SupportsForceSharedMemory();
-
     static void RecreateAll();
 };
 

tests/libslic3r/test_texture_displacement.cpp

@@ -35,10 +35,12 @@
 #include <boost/filesystem.hpp>
 
 #include "libslic3r/AABBMesh.hpp"
+#include "libslic3r/AABBTreeIndirect.hpp"
 #include "libslic3r/TextureDisplacement.hpp"
 #include "libslic3r/TextureBake/TextureBakeDecimate.hpp"
 #include "libslic3r/TextureBake/TextureBakeFlip.hpp"
 #include "libslic3r/TextureBake/TextureBakeMesh.hpp"
+#include "libslic3r/TextureBake/TextureBakeRepair.hpp"
 #include "libslic3r/TriangleMesh.hpp"
 #include "libslic3r/TriangleSelector.hpp"
 #include "libslic3r/PNGReadWrite.hpp"
@@ -2174,3 +2176,151 @@ TEST_CASE("whether a layer's texture has colour agrees with its decode", "[Textu
 
     CHECK_FALSE(height_texture_has_color(TextureDisplacementLayer{}));
 }
+
+TEST_CASE("Baking twice keeps the mesh closed and carries the first bake's colours across",
+          "[TextureDisplacement][Regression]")
+{
+    // Bake, colour part of what came back, bake again. The one-run pipeline rebuilds the topology, so
+    // the colour recorded against the first bake's mesh indexes triangles that no longer mean the same
+    // thing. Taken across by triangle id it lands on unrelated faces - a bake almost always grows the
+    // triangle count, so every stale id passes the only bound check deserialize() makes - which both
+    // scatters colour over surfaces nothing painted and destroys what the first pass laid down.
+    // TriangleSelector::remap_painting() carries it by position instead, which is what makes successive
+    // bakes accumulate. See TextureDisplacementBakeJob, which is where the two meet.
+    constexpr float            side    = 20.f;
+    constexpr float            refine  = 0.5f;
+    constexpr float            radius  = 4.f;
+    const indexed_triangle_set cube    = its_make_cube(side, side, side);
+    const Vec2f                first_stroke(6.f, 6.f), second_stroke(14.f, 14.f);
+
+    TextureDisplacementLayer layer;
+    layer.slot          = 0;
+    layer.image_data    = make_rgb_png_2x2();
+    layer.tiling_scale  = 5.f;
+    layer.depth_mm      = 0.5f;
+    layer.color_enabled = true;
+
+    TextureDisplacementOptions options;
+    options.pipeline_v2        = true;
+    options.v2_refine_mm       = refine;
+    options.v2_max_triangles_k = 750;
+
+    // Three primaries, so a palette entry is legible in the assertions below.
+    const std::array<Vec3f, 3> palette = { Vec3f(1, 0, 0), Vec3f(0, 1, 0), Vec3f(0, 0, 1) };
+    TextureColorRequest        request;
+    request.quantize = [&palette](const Vec3f &rgb) {
+        int   best = 0;
+        float bd   = std::numeric_limits<float>::max();
+        for (int i = 0; i < 3; ++i)
+            if (const float d = (palette[size_t(i)] - rgb).squaredNorm(); d < bd) { bd = d; best = i; }
+        return best;
+    };
+
+    const auto bake = [&](const indexed_triangle_set &mesh, const Vec2f &stroke, std::vector<uint8_t> &color) {
+        TextureDisplacementFacetsData facets;
+        facets[0] = paint_top_disc(mesh, side, stroke, radius);
+        TextureColorRequest req = request;
+        req.out_triangle        = &color;
+        return build_texture_displacement(mesh, { layer }, facets, options, {}, &req);
+    };
+    // The pipeline's own counters, over welded positions, so a mesh watertight only after welding
+    // still reads as closed.
+    const auto defects = [](const indexed_triangle_set &mesh) {
+        return TextureBake::count_edge_defects(TextureBake::to_soup(mesh));
+    };
+
+    std::vector<uint8_t>       color1;
+    const indexed_triangle_set first = bake(cube, first_stroke, color1);
+    REQUIRE(first.indices.size() > cube.indices.size());
+    REQUIRE(color1.size() == first.indices.size());
+    const TextureBake::EdgeDefects d1 = defects(first);
+    INFO("first bake: " << d1.triangles << " triangles, " << d1.open << " open, " << d1.non_manifold << " non-manifold");
+    CHECK(d1.open == 0);
+    CHECK(d1.non_manifold == 0);
+
+    // The colour the first bake committed, as TextureDisplacementBakeJob writes it.
+    TriangleSelector::TriangleSplittingData painted_first;
+    {
+        const TriangleMesh mesh(first);
+        TriangleSelector   selector(mesh);
+        for (size_t i = 0; i < color1.size(); ++i)
+            if (color1[i] != 0)
+                selector.set_facet(int(i), EnforcerBlockerType(color1[i]));
+        painted_first = selector.serialize();
+    }
+    REQUIRE_FALSE(painted_first.bitstream.empty());
+
+    std::vector<uint8_t>       color2;
+    const indexed_triangle_set second = bake(first, second_stroke, color2);
+    REQUIRE(color2.size() == second.indices.size());
+    REQUIRE(second.indices.size() != first.indices.size()); // the topology really did change
+    const TextureBake::EdgeDefects d2 = defects(second);
+    INFO("second bake: " << d2.triangles << " triangles, " << d2.open << " open, " << d2.non_manifold << " non-manifold");
+    CHECK(d2.open == 0);
+    CHECK(d2.non_manifold == 0);
+
+    // The commit: the earlier colour carried onto the new topology, this bake's own colour over it.
+    std::vector<uint8_t> committed(second.indices.size(), 0);
+    {
+        const TriangleMesh mesh(second);
+        TriangleSelector   selector(mesh);
+        const TriangleSelector::TriangleSplittingData carried = TriangleSelector::remap_painting(
+            first, painted_first, second, Transform3d::Identity(), {});
+        REQUIRE_FALSE(carried.bitstream.empty());
+        selector.deserialize(carried, false);
+        for (size_t i = 0; i < color2.size(); ++i)
+            if (color2[i] != 0)
+                selector.set_facet(int(i), EnforcerBlockerType(color2[i]));
+        for (int state = 1; state <= int(palette.size()); ++state) {
+            std::vector<int> src;
+            if (selector.get_facets_strict(EnforcerBlockerType(state), &src).indices.empty())
+                continue;
+            for (const int s : src)
+                if (s >= 0 && size_t(s) < committed.size() && committed[size_t(s)] == 0)
+                    committed[size_t(s)] = uint8_t(state);
+        }
+    }
+
+    // Just the triangles the first bake coloured, to ask "how far is this from colour that was really
+    // laid down" of anything the commit painted.
+    indexed_triangle_set coloured_first;
+    coloured_first.vertices = first.vertices;
+    for (size_t i = 0; i < color1.size(); ++i)
+        if (color1[i] != 0)
+            coloured_first.indices.push_back(first.indices[i]);
+    const AABBTreeIndirect::Tree3f tree = AABBTreeIndirect::build_aabb_tree_over_indexed_triangle_set(
+        coloured_first.vertices, coloured_first.indices);
+
+    // A triangle may carry colour the second bake did not produce only where the first bake's colour
+    // actually was. Carrying the mask by position can widen that region by a triangle, so the margin
+    // is the edge length the bake was told to refine to; landing anywhere else means the mask was read
+    // against a mesh it no longer matches.
+    size_t own = 0, carried_over = 0, stranded = 0;
+    float  worst = 0.f;
+    for (size_t i = 0; i < second.indices.size(); ++i) {
+        if (committed[i] == 0)
+            continue;
+        if (i < color2.size() && color2[i] != 0) {
+            CHECK(committed[i] == color2[i]); // this bake's own colour wins where it painted
+            ++own;
+            continue;
+        }
+        const stl_triangle_vertex_indices &t = second.indices[i];
+        const Vec3f g = (second.vertices[size_t(t[0])] + second.vertices[size_t(t[1])] +
+                         second.vertices[size_t(t[2])]) / 3.f;
+        size_t      hit = 0;
+        Vec3f       foot;
+        const float d = std::sqrt(std::max(0.f, AABBTreeIndirect::squared_distance_to_indexed_triangle_set(
+                                                    coloured_first.vertices, coloured_first.indices, tree, g, hit, foot)));
+        worst = std::max(worst, d);
+        if (d <= 2.f * refine)
+            ++carried_over;
+        else
+            ++stranded;
+    }
+    INFO("committed: " << own << " from this bake, " << carried_over << " carried over, " << stranded
+                       << " stranded, worst distance from the first bake's colour " << worst << " mm");
+    CHECK(own > 0);
+    CHECK(carried_over > 0);   // the first bake's colour survived the second
+    CHECK(stranded == 0);      // and nothing was painted where neither bake put colour
+}