Merge remote-tracking branch 'origin/master' into claude/issue-47

# Conflicts:
#	interfaces/Crafter.Graphics-UI.cppm
#	project.cpp

implementations/Crafter.Graphics-Font.cpp

@@ -56,6 +56,23 @@ Font::Font(const std::filesystem::path& fontFilePath) {
     this->ascent = ascent;
     this->descent = descent;
     this->lineGap = lineGap;
+
+    asciiAdvance_.fill(-1);
+}
+
+std::int32_t Font::AdvanceUnits(std::uint32_t cp) {
+    if (cp < asciiAdvance_.size()) {
+        if (asciiAdvance_[cp] < 0) {
+            int advance = 0, lsb = 0;
+            stbtt_GetCodepointHMetrics(&font, static_cast<int>(cp), &advance, &lsb);
+            asciiAdvance_[cp] = advance;
+        }
+        return asciiAdvance_[cp];
+    }
+    if (auto it = advanceUnits_.find(cp); it != advanceUnits_.end()) return it->second;
+    int advance = 0, lsb = 0;
+    stbtt_GetCodepointHMetrics(&font, static_cast<int>(cp), &advance, &lsb);
+    return advanceUnits_.emplace(cp, advance).first->second;
 }
 
 std::uint32_t Font::GetLineWidth(const std::string_view text, float size) {
@@ -65,13 +82,33 @@ std::uint32_t Font::GetLineWidth(const std::string_view text, float size) {
     while (i < text.size()) {
         std::uint32_t cp = DecodeUtf8(text, i);
         if (cp == 0) break;
-        int advance, lsb;
-        stbtt_GetCodepointHMetrics(&font, static_cast<int>(cp), &advance, &lsb);
-        lineWidth += (int)(advance * scale);
+        lineWidth += (int)(AdvanceUnits(cp) * scale);
     }
     return lineWidth;
 }
 
+std::size_t Font::NearestCursorByte(std::string_view text, float size, float target) {
+    if (target <= 0.0f) return 0;
+    float scale = stbtt_ScaleForPixelHeight(&font, size);
+    std::size_t best     = 0;        // byte offset 0 → caret before the first glyph
+    float       bestDist = target;   // |target - 0|, and target > 0 here
+    float       cumWidth = 0.0f;
+    std::size_t i = 0;
+    while (i < text.size()) {
+        std::uint32_t cp = DecodeUtf8(text, i);   // i advances to the next boundary
+        if (cp == 0) break;
+        cumWidth += (int)(AdvanceUnits(cp) * scale);  // match GetLineWidth's per-glyph rounding
+        float d = std::abs(target - cumWidth);
+        if (d < bestDist) {
+            bestDist = d;
+            best     = i;            // byte offset just past this codepoint
+        } else {
+            break;                   // monotonic advance ⇒ already past the closest boundary
+        }
+    }
+    return best;
+}
+
 float Font::LineHeight(float size) {
     float scale = stbtt_ScaleForPixelHeight(&font, size);
     return (ascent - descent + lineGap) * scale;

implementations/Crafter.Graphics-InputField.cpp

@@ -117,23 +117,12 @@ std::size_t Crafter::InputField_HitTestCursor(const InputField& f,
                                               Font& font, float fontSize,
                                               const InputFieldColors& colors)
 {
+    // Single O(n) cumulative-advance pass over the value, keyed at codepoint
+    // byte boundaries — matching cursorPos, which is a byte offset. The old
+    // loop re-walked the prefix for every boundary (O(n^2) glyph metric
+    // lookups) and could even land mid-codepoint on multibyte input.
     float target = clickX - (rect.x + colors.paddingX);
-    if (target <= 0.0f) return 0;
-
-    std::size_t best     = 0;
-    float       bestDist = std::abs(target);
-    for (std::size_t i = 1; i <= f.value.size(); ++i) {
-        std::string_view sub(f.value.data(), i);
-        float w = static_cast<float>(font.GetLineWidth(sub, fontSize));
-        float d = std::abs(target - w);
-        if (d < bestDist) {
-            bestDist = d;
-            best     = i;
-        } else {
-            break;
-        }
-    }
-    return best;
+    return font.NearestCursorByte(f.value, fontSize, target);
 }
 
 void Crafter::DrawInputField(UIBuffer& buf, const InputField& f, Rect rect,

implementations/Crafter.Graphics-UI-Shared.cpp

@@ -38,11 +38,27 @@ UIDispatchHeader UIRenderer::FillHeader(std::uint32_t itemBufferSlot,
 #else
     h.frameIdx  = 0;
 #endif
-    h.flags     = flags;
+    h.flags     = ClipFlags(clipRectPx, window_->width, window_->height, flags);
     h._pad      = 0;
     return h;
 }
 
+std::uint32_t UIRenderer::ClipFlags(std::array<float,4> clipRectPx,
+                                    std::uint32_t surfaceWidth,
+                                    std::uint32_t surfaceHeight,
+                                    std::uint32_t flags) noexcept {
+    // The reserved kUIFlagClip bit signals "clip rect is narrower than the
+    // surface". When the rect already covers the whole surface (the common
+    // {0,0,1e9,1e9} default), the four per-pixel clip compares in
+    // uiResolveScreenPixel can never reject an in-surface pixel, so we leave
+    // the bit clear and the shader skips them — pixel-identical output.
+    const bool clipCoversSurface =
+        clipRectPx[0] <= 0.0f && clipRectPx[1] <= 0.0f
+        && clipRectPx[0] + clipRectPx[2] >= static_cast<float>(surfaceWidth)
+        && clipRectPx[1] + clipRectPx[3] >= static_cast<float>(surfaceHeight);
+    return clipCoversSurface ? (flags & ~kUIFlagClip) : (flags | kUIFlagClip);
+}
+
 std::uint32_t UIRenderer::ShapeText(Font& font, float pxSize,
                                     float x, float baselineY,
                                     std::string_view utf8,

implementations/Crafter.Graphics-UI.cpp

@@ -219,7 +219,15 @@ void UIRenderer::DispatchFused(GraphicsCommandBuffer cmd,
     pc.outImage       = outImageSlot_;
     pc.fontTexture    = fontAtlasImageSlot_;
     pc.fontSampler    = fontAtlasSamplerSlot_;
-    pc.flags          = 0;
+    // Per-category clip-active bits, reusing the same surface-coverage test the
+    // standard path uses (ClipFlags). A category whose clip already covers the
+    // surface clears its bit so the kernel skips that category's per-pixel clip
+    // compares (the common full-surface case) — pixel-identical either way.
+    auto clipBit = [&](const FusedBatch& b, std::uint32_t bit) -> std::uint32_t {
+        return (ClipFlags(b.clipRectPx, window_->width, window_->height, 0) & kUIFlagClip) ? bit : 0u;
+    };
+    pc.flags          = clipBit(quads,   0x1u) | clipBit(circles, 0x2u)
+                      | clipBit(images,  0x4u) | clipBit(text,    0x8u);
     pc.surfaceWidth   = window_->width;
     pc.surfaceHeight  = window_->height;
     pc.frameIdx       = window_->currentBuffer;
@@ -239,15 +247,41 @@ void UIRenderer::Dispatch(GraphicsCommandBuffer cmd, const GraphicsComputeShader
                           const void* push, std::uint32_t pushBytes,
                           std::uint32_t gx, std::uint32_t gy, std::uint32_t gz) {
     if (!firstDispatchThisFrame_) {
-        VkMemoryBarrier mb {
-            .sType         = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
-            .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
+        // Every UI pass read-modify-writes the same swapchain storage image
+        // (later passes overdraw earlier ones), so each dispatch must observe
+        // the previous dispatch's writes. The only resource the hazard touches
+        // is that one image, so scope the dependency to its subresource with a
+        // VkImageMemoryBarrier rather than a queue-wide VkMemoryBarrier: same
+        // correctness, but only this image's shader caches are flushed —
+        // unrelated buffers/images are no longer invalidated. The image stays
+        // in VK_IMAGE_LAYOUT_GENERAL (bound as a storage image), so this is a
+        // pure memory dependency, no layout transition.
+        //
+        // The execution dependency is still COMPUTE->COMPUTE over the whole
+        // queue, so the passes do NOT overlap — this is a narrower cache flush
+        // only. Eliminating the barriers entirely requires fusing the passes
+        // into one dispatch (tracked in #47); do not attempt that here.
+        VkImageMemoryBarrier ib {
+            .sType               = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .srcAccessMask       = VK_ACCESS_SHADER_WRITE_BIT,
+            .dstAccessMask       = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
+            .oldLayout           = VK_IMAGE_LAYOUT_GENERAL,
+            .newLayout           = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image               = window_->images[window_->currentBuffer],
+            .subresourceRange    = {
+                .aspectMask     = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel   = 0,
+                .levelCount     = 1,
+                .baseArrayLayer = 0,
+                .layerCount     = 1,
+            },
         };
         vkCmdPipelineBarrier(cmd,
             VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
             VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
-            0, 1, &mb, 0, nullptr, 0, nullptr);
+            0, 0, nullptr, 0, nullptr, 1, &ib);
     }
     firstDispatchThisFrame_ = false;
     shader.Dispatch(cmd, push, pushBytes, gx, gy, gz);