[core] standardize on uint16_t for texture sizes
This commit is contained in:
@@ -208,7 +208,7 @@ void NativeMapView::render() {
|
||||
// take snapshot
|
||||
const unsigned int w = fbWidth;
|
||||
const unsigned int h = fbHeight;
|
||||
mbgl::PremultipliedImage image { w, h };
|
||||
mbgl::PremultipliedImage image { static_cast<uint16_t>(w), static_cast<uint16_t>(h) };
|
||||
MBGL_CHECK_ERROR(glReadPixels(0, 0, w, h, GL_RGBA, GL_UNSIGNED_BYTE, image.data.get()));
|
||||
const size_t stride = image.stride();
|
||||
auto tmp = std::make_unique<uint8_t[]>(stride);
|
||||
|
||||
Reference in New Issue
Block a user