<html> <head> <meta http-equiv="content-type" content="text/html; charset=ISO-8859-1"> </head> <body text="#000000" bgcolor="#FFFFFF"> Hello, in gstreamer-imx I have a "imxvpuvideotransform" element. This element uses the i.MX6 IPU for rotations, colorspace conversions, scaling, and deinterlacing. All of these operations can be done in one step. As an optimization, I check if a frame really needs to be transformed, or if it can be passed through. Cases where pass through is possible: - input and output width/height/pixel format are the same - video cropping is enabled, but the crop rectangle contains the entire frame (thus not actually cropping anything) - input frame is not interlaced - input frame is interlaced, but no deinterlacing will be performed - rotation is disabled Since rotation can be set at any time (it is a property), video crop rectangles can in theory change during playback, and deinterlacing might be required only for some frames in mixed mode, I need to evaluate this per-frame. I perform the evaluation in the prepare_output_buffer() function. Initially I called <meta http-equiv="content-type" content="text/html; charset=ISO-8859-1"> gst_base_transform_set_passthrough() in prepare_output_buffer(). However, this lead to problems in certain situations, such as when playbin tries to play next media. For this reason, I now call gst_base_transform_set_passthrough() once, during initialization, and set GstBaseTransform's passthrough flag to FALSE. I handle passthrough manually by setting the output buffer to the input buffer in prepare_output_buffer() if passthrough is required. transform() is still called, but does nothing then. Here is the relevant part in the code: <a class="moz-txt-link-freetext" href="https://github.com/Freescale/gstreamer-imx/blob/master/src/ipu/videotransform/videotransform.c#L1295">https://github.com/Freescale/gstreamer-imx/blob/master/src/ipu/videotransform/videotransform.c#L1295</a> My question now is: is this an acceptable solution? The documentation neither confirms nor denies this. It states: "Elements that only do analysis can return a subbuffer or even just return a reference to the input buffer (if in passthrough mode)." What if I do *outbuf = input; if I set GstBaseTransform's passthrough to FALSE? It works right now. I want to make sure this isn't just a fluke, and in some future GStreamer version, somebody puts an assert that fires in this case. cheers </body> </html>