On Tue Jul 2 17:46:02 2024 +0000, Elizabeth Figura wrote:
Where does this come from? Unless I've missed something, there are no tests for mismatched content sizes (only mismatched frame sizes with the same content size). If that's the case, and I'm not missing a test somewhere, can we please either add some tests, or stop pretending to handle mismatched content size, instead just erroring out with a FIXME?
There are some tests, in `test_color_convert`, and enforcing a matching input/output content size makes them fail to pass.
The tests use a 96x96 frame size + 82x84 aperture input type (set at `mf/tests/transform.c:7259`), and a 96x96 frame size without aperture output type. The result is that the full image is output, with the padding data becoming part of the frame.
Then there isn't any equivalent tests with the video processor, although some of the test SetInputType / SetOutputType call sequence also temporarily end up with mismatched content size, and they all succeed so we should not return an error.
Further checking that case locally it looks like the video processor doesn't behave the same as the color converter. I haven't seen that being used anywhere yet though, so I'm only going to add a FIXME here.