Yeh, but if wrapping works, then the negative value of width/height can't be treated as zero. So, I think the width and height of this rectangle should be scalar.According to Nikolay Sivov's suggestion, if you set the WordWrapping of text_layout to DWRITE_WORD_WRAPPING_CHARACTER, then the wrapping will work properly for now. So the width/height of the rectangle should always be valid (except for zero), regardless of whether it is inverted or not. Henri Verbeet <hverbeet(a)gmail.com> 于2021年1月25日周一 下午10:08写道：
On Fri, 22 Jan 2021 at 03:21, 陈Daxia <chy2009123(a)gmail.com> wrote:
@@ -1161,7 +1161,6 @@ static void STDMETHODCALLTYPE d2d_device_context_DrawText(ID2D1DeviceContext *if IDWriteTextLayout *text_layout; IDWriteFactory *dwrite_factory; D2D1_POINT_2F origin; - float width, height; HRESULT hr;
TRACE("iface %p, string %s, string_len %u, text_format %p, layout_rect %s, " @@ -1176,14 +1175,13 @@ static void STDMETHODCALLTYPE d2d_device_context_DrawText(ID2D1DeviceContext *if return; }
- width = max(0.0f, layout_rect->right - layout_rect->left); - height = max(0.0f, layout_rect->bottom - layout_rect->top); if (measuring_mode == DWRITE_MEASURING_MODE_NATURAL) hr = IDWriteFactory_CreateTextLayout(dwrite_factory, string, string_len, text_format, - width, height, &text_layout); + fabs(layout_rect->right - layout_rect->left), fabs(layout_rect->bottom - layout_rect->top), &text_layout); else hr = IDWriteFactory_CreateGdiCompatibleTextLayout(dwrite_factory, string, string_len, text_format, - width, height, render_target->desc.dpiX / 96.0f, (DWRITE_MATRIX *)&render_target->drawing_state.transform, + fabs(layout_rect->right - layout_rect->left), fabs(layout_rect->bottom - layout_rect->top), + render_target->desc.dpiX / 96.0f, (DWRITE_MATRIX *)&render_target->drawing_state.transform, measuring_mode == DWRITE_MEASURING_MODE_GDI_NATURAL, &text_layout); IDWriteFactory_Release(dwrite_factory); if (FAILED(hr))
When Nikolay sent 9521c6a1eb781ff7cb75a83544bf9b78f889a9e2, he said his (private) tests showed negative width/height behave like zero width/height. Do you have tests that contradict that? Regardless of whether this patch is functionally correct, getting rid of the "width" and "height" variables doesn't seem like an improvement.