TextureImplDefault::copyToImage calls glGetTexImage into a buffer sized for size (public size), but glGetTexImage reads the full underlying texture dimensions (which may be larger due to NPOT padding). This can overflow pixels and is also missing the previous logic that trimmed padding/flipped rows. Store the actual texture size in the impl (or query it via glGetTexLevelParameteriv) and only copy the valid sub-rectangle into the output buffer, applying vertical flip when requested.

TextureImplDefault::update(const Window&, ...) no longer activates the window’s context before calling glCopyTexSubImage2D. Previously this code ensured window.setActive(true) succeeded; without it the copy may read from the wrong context/backbuffer or fail. Call window.setActive(true) (and handle failure) before copying.

VertexBufferImpl::isAvailable() uses GLEXT_vertex_buffer_object but never calls ensureExtensionsInit(). If extensions haven’t been initialized yet, this can incorrectly report that VBOs are unavailable (or rely on uninitialized globals). Mirror the pattern used by ShaderImplDefault::isAvailable() by calling ensureExtensionsInit() under the transient context lock before checking the extension flags.

The previous RenderTarget logic always re-bound (and later forcibly unbound) textures that are FBO attachments (Texture::m_fboAttachment) to ensure cross-context visibility and avoid driver clear issues. The new setupDraw/cleanupDraw paths dropped this special-casing, which can reintroduce those known driver bugs. Restore the FBO-attachment handling (always rebind in setupDraw, and unbind in cleanupDraw when an FBO attachment was used).

Shader::getNativeHandle() now unconditionally dereferences m_impl. With the move operations defaulted, moved-from Shader instances will have m_impl == nullptr, so calling getNativeHandle() can crash. Return 0 when m_impl is null (matching the previous moved-from behavior).

The old code enabled the texture-coordinate array when either a texture or a shader was used (states.texture || states.shader). The new implementation toggles GL_TEXTURE_COORD_ARRAY based only on states.texture, which can break shaders that rely on gl_MultiTexCoord0 / texture coordinates even when no texture is bound. Include states.shader in the condition for enabling/disabling the texcoord array (both for immediate vertices and VBO drawing).

RenderTarget::draw(const VertexBuffer&, ...) now calls m_impl->setupDraw(...) but doesn’t check whether it actually activated the target/context (the impl can early-return if activation fails). The function then proceeds to bind VBOs and issue draw calls unconditionally. Either ensure activation succeeds before continuing (e.g., early-return on failure) or move the entire VBO draw path into the impl so it can manage activation consistently.

There’s a malformed Doxygen tag here (/// �…rief contains a stray control character) and the class closing brace is jammed onto the same line as the last pure-virtual (= 0;};). This will render incorrectly in docs and is easy to miss in reviews. Fix the tag to \brief and put the closing }; on its own line.

This Doxygen comment contains a stray control character in \brief (/// �…rief). It will break documentation generation/rendering. Replace it with a normal /// \brief ... line.

Texture::getNativeHandle() now unconditionally dereferences m_impl. After a move, the moved-from Texture will have m_impl == nullptr (due to the defaulted move operations), so calling getNativeHandle() can crash. Preserve the previous behavior by returning 0 when m_impl is null (and consider doing the same for other const accessors).