[PATCH 0/2] MR8175: mf: Clear pending MFT stream requests when flushing.

[Charlotte Pabst (＠CharlottePabst)]

Before this, stopping and restarting the media session
could cause MFT streams to get stuck in the "already pending"
state, which would lead to no more samples getting delivered.

This also reverts 18c0dd7390b354d2e505e54cd3f3f9b08226b265,
my previous attempt to fix the problem which for some reason
seemed to help - but with the new fix, it does not make any
difference.

The reverted commit is faulty since the Flush() method - which is called at session Stop, but runs after sample_grabber_set_state - already releases pending items, while processing markers immediately.

Sorry for the trouble.

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/8175

[PATCH 1/2] mf: Clear pending MFT stream requests when flushing.

[Charlotte Pabst]

From: Charlotte Pabst <cpabst(a)codeweavers.com>

Before this, stopping and restarting the media session
could cause MFT streams to get stuck in the "already pending"
state, which would lead to no more samples getting delivered.
---
 dlls/mf/session.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/dlls/mf/session.c b/dlls/mf/session.c
index 483ea6f904f..a9213d2cf18 100644
--- a/dlls/mf/session.c
+++ b/dlls/mf/session.c
@@ -1014,8 +1014,13 @@ static void session_flush_nodes(struct media_session *session)
     {
         if (node->type == MF_TOPOLOGY_OUTPUT_NODE)
             IMFStreamSink_Flush(node->object.sink_stream);
-        else if (node->type == MF_TOPOLOGY_TRANSFORM_NODE)
+        else if (node->type == MF_TOPOLOGY_TRANSFORM_NODE) {
+            UINT i;
+            for (i = 0; i < node->u.transform.output_count; i++)
+                node->u.transform.outputs[i].requests = 0;
+
             IMFTransform_ProcessMessage(node->object.transform, MFT_MESSAGE_COMMAND_FLUSH, 0);
+        }
     }
 }
 
-- 
GitLab


https://gitlab.winehq.org/wine/wine/-/merge_requests/8175

[PATCH 2/2] Revert "mf: Release pending items when sample grabber is stopped."

[Charlotte Pabst]

From: Charlotte Pabst <cpabst(a)codeweavers.com>

This reverts commit 18c0dd7390b354d2e505e54cd3f3f9b08226b265.

The actual bug that broke sample delivery was in mf session code.

This commit is faulty since the Flush() method - which is called
at session Stop, but runs after sample_grabber_set_state - already
releases pending items, while processing markers immediately.
---
 dlls/mf/samplegrabber.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/dlls/mf/samplegrabber.c b/dlls/mf/samplegrabber.c
index 016c68f98b8..de599139736 100644
--- a/dlls/mf/samplegrabber.c
+++ b/dlls/mf/samplegrabber.c
@@ -1191,7 +1191,6 @@ static HRESULT sample_grabber_set_state(struct sample_grabber *grabber, enum sin
                 sample_grabber_cancel_timer(grabber);
                 release_samples(grabber);
                 grabber->sample_count = MAX_SAMPLE_QUEUE_LENGTH;
-                sample_grabber_release_pending_items(grabber);
             }
 
             if (state == SINK_STATE_RUNNING && grabber->state != SINK_STATE_RUNNING)
-- 
GitLab

https://gitlab.winehq.org/wine/wine/-/merge_requests/8175

[Nikolay Sivov (＠nsivov)]

Please apply this formatting change to the first commit, but otherwise sure, let's try it.

[fmt.diff](/uploads/c024bde2e0f1eb269c580f2a7dc08d07/fmt.diff)

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/8175#note_107077