AtlasEngine: Fix bugs introduced in 66f4f9d and d74b66a

src/renderer/atlas/AtlasEngine.cpp

@@ -360,6 +360,40 @@ try
         _api.dirtyRect = til::rect{ 0, _api.invalidatedRows.x, _api.cellCount.x, _api.invalidatedRows.y };
     }
 
+    // This is an important block of code for our TileHashMap.
+    // We only process glyphs within the dirtyRect, but glyphs outside of the
+    // dirtyRect are still in use and shouldn't be discarded. This is critical
+    // if someone uses a tool like tmux to split the terminal horizontally.
+    // If they then print a lot of Unicode text on just one side, we have to
+    // ensure that the (for example) plain ASCII glyphs on the other half of the
+    // viewport are still retained. This bit of code "refreshes" those glyphs and
+    // brings them to the front of the LRU queue to prevent them from being reused.
+    {
+        const std::array<til::point, 2> ranges{ {
+            { 0, _api.dirtyRect.top },
+            { _api.dirtyRect.bottom, _api.cellCount.y },
+        } };
+        const auto stride = static_cast<size_t>(_r.cellCount.x);
+
+        for (const auto& p : ranges)
+        {
+            // We (ab)use the .x/.y members of the til::point as the
+            // respective [from,to) range of rows we need to makeNewest().
+            const auto from = p.x;
+            const auto to = p.y;
+
+            for (auto y = from; y < to; ++y)
+            {
+                auto it = _r.cellGlyphMapping.data() + stride * y;
+                const auto end = it + stride;
+                for (; it != end; ++it)
+                {
+                    _r.glyphs.makeNewest(*it);
+                }
+            }
+        }
+    }
+
     return S_OK;
 }
 catch (const wil::ResultException& exception)
@@ -883,10 +917,12 @@ void AtlasEngine::_recreateSizeDependentResources()
         // This buffer is a bit larger than the others (multiple MB).
         // Prevent a memory usage spike, by first deallocating and then allocating.
         _r.cells = {};
+        _r.cellGlyphMapping = {};
         // Our render loop heavily relies on memcpy() which is between 1.5x
         // and 40x faster for allocations with an alignment of 32 or greater.
         // (40x on AMD Zen1-3, which have a rep movsb performance issue. MSFT:33358259.)
         _r.cells = Buffer<Cell, 32>{ totalCellCount };
+        _r.cellGlyphMapping = Buffer<TileHashMap::iterator>{ totalCellCount };
         _r.cellCount = _api.cellCount;
         _r.tileAllocator.setMaxArea(_api.sizeInPixel);
 
@@ -1062,6 +1098,13 @@ AtlasEngine::Cell* AtlasEngine::_getCell(u16 x, u16 y) noexcept
     return _r.cells.data() + static_cast<size_t>(_r.cellCount.x) * y + x;
 }
 
+AtlasEngine::TileHashMap::iterator* AtlasEngine::_getCellGlyphMapping(u16 x, u16 y) noexcept
+{
+    assert(x < _r.cellCount.x);
+    assert(y < _r.cellCount.y);
+    return _r.cellGlyphMapping.data() + static_cast<size_t>(_r.cellCount.x) * y + x;
+}
+
 void AtlasEngine::_setCellFlags(u16r coords, CellFlags mask, CellFlags bits) noexcept
 {
     assert(coords.left <= coords.right);
@@ -1399,9 +1442,9 @@ void AtlasEngine::_emplaceGlyph(IDWriteFontFace* fontFace, size_t bufferPos1, si
     attributes.cellCount = cellCount;
 
     AtlasKey key{ attributes, gsl::narrow<u16>(charCount), chars };
-    const AtlasValue* valueRef = _r.glyphs.find(key);
+    auto it = _r.glyphs.find(key);
 
-    if (!valueRef)
+    if (it == _r.glyphs.end())
     {
         // Do fonts exist *in practice* which contain both colored and uncolored glyphs? I'm pretty sure...
         // However doing it properly means using either of:
@@ -1440,26 +1483,24 @@ void AtlasEngine::_emplaceGlyph(IDWriteFontFace* fontFace, size_t bufferPos1, si
             coords[i] = _r.tileAllocator.allocate(_r.glyphs);
         }
 
-        const auto it = _r.glyphs.insert(std::move(key), std::move(value));
-        valueRef = &it->second;
+        it = _r.glyphs.insert(std::move(key), std::move(value));
         _r.glyphQueue.emplace_back(&it->first, &it->second);
     }
 
-    // For some reason MSVC doesn't understand that valueRef is overwritten in the branch above, resulting in:
-    //   C26430: Symbol 'valueRef' is not tested for nullness on all paths (f.23).
-    __assume(valueRef != nullptr);
-
-    const auto valueData = valueRef->data();
+    const auto valueData = it->second.data();
     const auto coords = &valueData->coords[0];
-    const auto data = _getCell(x1, _api.lastPaintBufferLineCoord.y);
+    const auto cells = _getCell(x1, _api.lastPaintBufferLineCoord.y);
+    const auto cellGlyphMappings = _getCellGlyphMapping(x1, _api.lastPaintBufferLineCoord.y);
 
     for (u32 i = 0; i < cellCount; ++i)
     {
-        data[i].tileIndex = coords[i];
+        cells[i].tileIndex = coords[i];
         // We should apply the column color and flags from each column (instead
         // of copying them from the x1) so that ligatures can appear in multiple
         // colors with different line styles.
-        data[i].flags = valueData->flags | _api.bufferLineMetadata[static_cast<size_t>(x1) + i].flags;
-        data[i].color = _api.bufferLineMetadata[static_cast<size_t>(x1) + i].colors;
+        cells[i].flags = valueData->flags | _api.bufferLineMetadata[static_cast<size_t>(x1) + i].flags;
+        cells[i].color = _api.bufferLineMetadata[static_cast<size_t>(x1) + i].colors;
     }
+
+    std::fill_n(cellGlyphMappings, cellCount, it);
 }

src/renderer/atlas/AtlasEngine.h

@@ -175,19 +175,24 @@ namespace Microsoft::Console::Render
                 _data{ allocate(size) },
                 _size{ size }
             {
+                std::uninitialized_default_construct_n(_data, size);
             }
 
             Buffer(const T* data, size_t size) :
                 _data{ allocate(size) },
                 _size{ size }
             {
-                static_assert(std::is_trivially_copyable_v<T>);
-                memcpy(_data, data, size * sizeof(T));
+                // Changing the constructor arguments to accept std::span might
+                // be a good future extension, but not to improve security here.
+                // You can trivially construct std::span's from invalid ranges.
+                // Until then the raw-pointer style is more practical.
+#pragma warning(suppress : 26459) // You called an STL function '...' with a raw pointer parameter at position '3' that may be unsafe [...].
+                std::uninitialized_copy_n(data, size, _data);
             }
 
             ~Buffer()
             {
-                deallocate(_data);
+                destroy();
             }
 
             Buffer(Buffer&& other) noexcept :
@@ -199,7 +204,7 @@ namespace Microsoft::Console::Render
 #pragma warning(suppress : 26432) // If you define or delete any default operation in the type '...', define or delete them all (c.21).
             Buffer& operator=(Buffer&& other) noexcept
             {
-                deallocate(_data);
+                destroy();
                 _data = std::exchange(other._data, nullptr);
                 _size = std::exchange(other._size, 0);
                 return *this;
@@ -288,6 +293,12 @@ namespace Microsoft::Console::Render
             }
 #pragma warning(pop)
 
+            void destroy() noexcept
+            {
+                std::destroy_n(_data, _size);
+                deallocate(_data);
+            }
+
             T* _data = nullptr;
             size_t _size = 0;
         };
@@ -553,21 +564,28 @@ namespace Microsoft::Console::Render
 
         struct TileHashMap
         {
+            using iterator = std::list<std::pair<AtlasKey, AtlasValue>>::iterator;
+
             TileHashMap() noexcept = default;
 
-            AtlasValue* find(const AtlasKey& key)
+            iterator end() noexcept
+            {
+                return _lru.end();
+            }
+
+            iterator find(const AtlasKey& key)
             {
                 const auto it = _map.find(key);
                 if (it != _map.end())
                 {
                     // Move the key to the head of the LRU queue.
-                    _lru.splice(_lru.begin(), _lru, *it);
-                    return &(*it)->second;
+                    makeNewest(*it);
+                    return *it;
                 }
-                return nullptr;
+                return end();
             }
 
-            std::list<std::pair<AtlasKey, AtlasValue>>::iterator insert(AtlasKey&& key, AtlasValue&& value)
+            iterator insert(AtlasKey&& key, AtlasValue&& value)
             {
                 // Insert the key/value right at the head of the LRU queue, just like find().
                 //
@@ -579,6 +597,11 @@ namespace Microsoft::Console::Render
                 return it;
             }
 
+            void makeNewest(const iterator& it)
+            {
+                _lru.splice(_lru.begin(), _lru, it);
+            }
+
             void popOldestTiles(std::vector<u16x2>& out) noexcept
             {
                 Expects(!_lru.empty());
@@ -602,7 +625,7 @@ namespace Microsoft::Console::Render
             // If you need a LRU hash-map, write a custom one with an intrusive
             // prev/next linked list (it's easier than you might think!).
             std::list<std::pair<AtlasKey, AtlasValue>> _lru;
-            std::unordered_set<std::list<std::pair<AtlasKey, AtlasValue>>::iterator, AtlasKeyHasher, AtlasKeyEq> _map;
+            std::unordered_set<iterator, AtlasKeyHasher, AtlasKeyEq> _map;
         };
 
         // TileAllocator yields `tileSize`-sized tiles for our texture atlas.
@@ -844,6 +867,7 @@ namespace Microsoft::Console::Render
         IDWriteTextFormat* _getTextFormat(bool bold, bool italic) const noexcept;
         const Buffer<DWRITE_FONT_AXIS_VALUE>& _getTextFormatAxis(bool bold, bool italic) const noexcept;
         Cell* _getCell(u16 x, u16 y) noexcept;
+        TileHashMap::iterator* _getCellGlyphMapping(u16 x, u16 y) noexcept;
         void _setCellFlags(u16r coords, CellFlags mask, CellFlags bits) noexcept;
         void _flushBufferLine();
         void _emplaceGlyph(IDWriteFontFace* fontFace, size_t bufferPos1, size_t bufferPos2);
@@ -911,6 +935,7 @@ namespace Microsoft::Console::Render
             wil::com_ptr<IDWriteTypography> typography;
 
             Buffer<Cell, 32> cells; // invalidated by ApiInvalidations::Size
+            Buffer<TileHashMap::iterator> cellGlyphMapping; // invalidated by ApiInvalidations::Size
             f32x2 cellSizeDIP; // invalidated by ApiInvalidations::Font, caches _api.cellSize but in DIP
             u16x2 cellSize; // invalidated by ApiInvalidations::Font, caches _api.cellSize
             u16x2 cellCount; // invalidated by ApiInvalidations::Font|Size, caches _api.cellCount

src/renderer/atlas/AtlasEngine.r.cpp

@@ -293,11 +293,16 @@ void AtlasEngine::_drawCursor()
     // At 150% scale lineWidth thus needs to be 1.33333... because at a zoom scale of 1.5 this results in a 2px wide line.
     const auto lineWidth = std::max(1.0f, static_cast<float>((_r.dpi + USER_DEFAULT_SCREEN_DPI / 2) / USER_DEFAULT_SCREEN_DPI * USER_DEFAULT_SCREEN_DPI) / static_cast<float>(_r.dpi));
     const auto cursorType = static_cast<CursorType>(_r.cursorOptions.cursorType);
-    D2D1_RECT_F rect;
-    rect.left = 0.0f;
-    rect.top = 0.0f;
-    rect.right = _r.cellSizeDIP.x;
-    rect.bottom = _r.cellSizeDIP.y;
+
+    // `clip` is the rectangle within our texture atlas that's reserved for our cursor texture, ...
+    D2D1_RECT_F clip;
+    clip.left = 0.0f;
+    clip.top = 0.0f;
+    clip.right = _r.cellSizeDIP.x;
+    clip.bottom = _r.cellSizeDIP.y;
+
+    // ... whereas `rect` is just the visible (= usually white) portion of our cursor.
+    auto rect = clip;
 
     switch (cursorType)
     {
@@ -328,6 +333,9 @@ void AtlasEngine::_drawCursor()
     }
 
     _r.d2dRenderTarget->BeginDraw();
+    // We need to clip the area we draw in to ensure we don't
+    // accidentally draw into any neighboring texture atlas tiles.
+    _r.d2dRenderTarget->PushAxisAlignedClip(&clip, D2D1_ANTIALIAS_MODE_ALIASED);
     _r.d2dRenderTarget->Clear();
 
     if (cursorType == CursorType::EmptyBox)
@@ -346,5 +354,6 @@ void AtlasEngine::_drawCursor()
         _r.d2dRenderTarget->FillRectangle(&rect, _r.brush.get());
     }
 
+    _r.d2dRenderTarget->PopAxisAlignedClip();
     THROW_IF_FAILED(_r.d2dRenderTarget->EndDraw());
 }