[PATCH v2 0/1] MR8933: mf: Update the stream sink input type when a transform output type changes.

[Conor McCarthy (＠cmccarthy)]

--
  v2: mf: Update the stream sink input type when handling an output node format change.

https://gitlab.winehq.org/wine/wine/-/merge_requests/8933

[PATCH v2 1/1] mf: Update the stream sink input type when handling an output node format change.

[Conor McCarthy]

From: Conor McCarthy

---
 dlls/mf/session.c | 35 ++++++++++++++++++++++++++++++-----
 1 file changed, 30 insertions(+), 5 deletions(-)

diff --git a/dlls/mf/session.c b/dlls/mf/session.c
index abc77d9f143..a06f106f4dc 100644
--- a/dlls/mf/session.c
+++ b/dlls/mf/session.c
@@ -3765,18 +3765,43 @@ static HRESULT transform_node_handle_format_change(struct media_session *session
     return transform_stream_update_input_type(topo_node, input, media_type);
 }

+/* For H.264, the sink's input media type must be set to the aligned frame size
+ * before MESessionStarted is sent. The decoder sends MF_E_TRANSFORM_STREAM_CHANGE
+ * to trigger a refresh of its output type once the aligned frame size is known.
+ * This occurs before any call to ProcessOutput() can succeed on the transform.
+ * Satisfactory is known to use the sink frame size. */
+static HRESULT sink_node_update_input_type(struct topo_node *topo_node, IMFMediaType *media_type)
+{
+    IMFMediaTypeHandler *handler;
+    HRESULT hr;
+
+    if (FAILED(hr = IMFStreamSink_GetMediaTypeHandler(topo_node->object.sink_stream, &handler)))
+    {
+        WARN("Failed to get type handler, hr %#lx.\n", hr);
+        return hr;
+    }
+    if (FAILED(hr = IMFMediaTypeHandler_SetCurrentMediaType(handler, media_type)))
+        WARN("Failed to set type, hr %#lx.\n", hr);
+    IMFMediaTypeHandler_Release(handler);
+
+    return hr;
+}
+
 static HRESULT session_handle_format_change(struct media_session *session, struct topo_node *topo_node,
         UINT input, IMFMediaType *media_type)
 {
-    HRESULT hr;
+    HRESULT hr = S_OK;

switch (topo_node->type)
     {
         case MF_TOPOLOGY_OUTPUT_NODE:
-            if (!topo_node->u.sink.allocator)
-                return S_OK;
-            if (SUCCEEDED(hr = IMFVideoSampleAllocator_UninitializeSampleAllocator(topo_node->u.sink.allocator)))
-                hr = IMFVideoSampleAllocator_InitializeSampleAllocator(topo_node->u.sink.allocator, 4, media_type);
+            if (topo_node->u.sink.allocator)
+            {
+                if (SUCCEEDED(hr = IMFVideoSampleAllocator_UninitializeSampleAllocator(topo_node->u.sink.allocator)))
+                    hr = IMFVideoSampleAllocator_InitializeSampleAllocator(topo_node->u.sink.allocator, 4, media_type);
+            }
+            if (SUCCEEDED(hr))
+                hr = sink_node_update_input_type(topo_node, media_type);
             return hr;

case MF_TOPOLOGY_TRANSFORM_NODE:

[Conor McCarthy (＠cmccarthy)]

On Thu Oct 16 14:12:15 2025 +0000, Nikolay Sivov wrote:
> The way it works now, if I'm reading this part correctly:
> MF_E_TRANSFORM_STREAM_CHANGE (happens due to decoder internal logic) ->
> transform_node_format_changed -> transform_stream_update_output_type +
> transform_stream_push_format_change -> MEStreamFormatChanged -> session_handle_format_change.
> Now, session_handle_format_change() then branches depending on
> destination node type. For output nodes it only cares about
> re-initializing sample allocator. My point is, shouldn't we amend
> session_handle_format_change() instead and keep existing flow, instead
> of fixing up sink types that early?

I moved it to `session_handle_format_change()`.

[Nikolay Sivov (＠nsivov)]

Nikolay Sivov (@nsivov) commented about dlls/mf/session.c:
> +static HRESULT sink_node_update_input_type(struct topo_node *topo_node, IMFMediaType *media_type)
> +{
> - IMFMediaTypeHandler *handler;
> - HRESULT hr;
> -
> - if (FAILED(hr = IMFStreamSink_GetMediaTypeHandler(topo_node->object.sink_stream, &handler)))
> - {
> -
>    WARN("Failed to get type handler, hr %#lx.\n", hr);
> -
>    return hr;
> - }
> - if (FAILED(hr = IMFMediaTypeHandler_SetCurrentMediaType(handler, media_type)))
> -
>    WARN("Failed to set type, hr %#lx.\n", hr);
> - IMFMediaTypeHandler_Release(handler);
> -
> - return hr;
>
> +}

This could be reduce to 3 lines with topology_node_get_type_handler() - _get_() -> SetCurrentMediaType() -> Release(). It will be short enough to avoid another helper.

[Nikolay Sivov (＠nsivov)]

Nikolay Sivov (@nsivov) commented about dlls/mf/session.c:
>  switch (topo_node->type)
>  {
>      case MF_TOPOLOGY_OUTPUT_NODE:
> -
>        if (!topo_node->u.sink.allocator)
> -
>            return S_OK;
> -
>        if (SUCCEEDED(hr = IMFVideoSampleAllocator_UninitializeSampleAllocator(topo_node->u.sink.allocator)))
> -
>            hr = IMFVideoSampleAllocator_InitializeSampleAllocator(topo_node->u.sink.allocator, 4, media_type);
> -
>        if (topo_node->u.sink.allocator)
> -
>        {
> -
>            if (SUCCEEDED(hr = IMFVideoSampleAllocator_UninitializeSampleAllocator(topo_node->u.sink.allocator)))
> -
>                hr = IMFVideoSampleAllocator_InitializeSampleAllocator(topo_node->u.sink.allocator, 4, media_type);
> -
>        }
> -
>        if (SUCCEEDED(hr))
> -
>            hr = sink_node_update_input_type(topo_node, media_type);
>        return hr;

Would it make sense to first update the type and then reinitialize allocator, if type update succeeded? Allocator is normally our code, while sink could be generally anything.