[PATCH 0/4] MR6950: winegstreamer: Handle null wg_transform in several functions.

[PATCH 1/4] winegstreamer: Handle null transform in video IMediaObject::Flush().

[Conor McCarthy]

From: Conor McCarthy

Other methods for the video object return DMO_E_TYPE_NOT_SET if
wg_transform is null. A race condition in Nioh can result in Flush()
being called when the transform is not set.
---
 dlls/winegstreamer/video_decoder.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/dlls/winegstreamer/video_decoder.c b/dlls/winegstreamer/video_decoder.c
index 891066dac3b..ce716dd0004 100644
--- a/dlls/winegstreamer/video_decoder.c
+++ b/dlls/winegstreamer/video_decoder.c
@@ -1315,6 +1315,9 @@ static HRESULT WINAPI media_object_Flush(IMediaObject *iface)

TRACE("iface %p.\n", iface);

+    if (!decoder->wg_transform)
+        return DMO_E_TYPE_NOT_SET;
+
     if (FAILED(hr = wg_transform_flush(decoder->wg_transform)))
         return hr;

[PATCH 2/4] winegstreamer: Handle null transform in video IMFTransform::ProcessMessage() DRAIN.

[Conor McCarthy]

From: Conor McCarthy

ProcessMessage() can return MF_E_TRANSFORM_TYPE_NOT_SET if necessary.
---
 dlls/winegstreamer/video_decoder.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/dlls/winegstreamer/video_decoder.c b/dlls/winegstreamer/video_decoder.c
index ce716dd0004..d071b009a49 100644
--- a/dlls/winegstreamer/video_decoder.c
+++ b/dlls/winegstreamer/video_decoder.c
@@ -791,7 +791,7 @@ static HRESULT WINAPI transform_ProcessMessage(IMFTransform *iface, MFT_MESSAGE_
         return S_OK;

case MFT_MESSAGE_COMMAND_DRAIN:
-        return wg_transform_drain(decoder->wg_transform);
+        return decoder->wg_transform ? wg_transform_drain(decoder->wg_transform) : MF_E_TRANSFORM_TYPE_NOT_SET;

case MFT_MESSAGE_COMMAND_FLUSH:
         return wg_transform_flush(decoder->wg_transform);

[PATCH 3/4] winegstreamer: Handle null transform in video IMFTransform::ProcessMessage() FLUSH.

[Conor McCarthy]

From: Conor McCarthy

---
 dlls/winegstreamer/video_decoder.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/dlls/winegstreamer/video_decoder.c b/dlls/winegstreamer/video_decoder.c
index d071b009a49..83fb713d5da 100644
--- a/dlls/winegstreamer/video_decoder.c
+++ b/dlls/winegstreamer/video_decoder.c
@@ -794,7 +794,7 @@ static HRESULT WINAPI transform_ProcessMessage(IMFTransform *iface, MFT_MESSAGE_
         return decoder->wg_transform ? wg_transform_drain(decoder->wg_transform) : MF_E_TRANSFORM_TYPE_NOT_SET;

case MFT_MESSAGE_COMMAND_FLUSH:
-        return wg_transform_flush(decoder->wg_transform);
+        return decoder->wg_transform ? wg_transform_flush(decoder->wg_transform) : MF_E_TRANSFORM_TYPE_NOT_SET;

case MFT_MESSAGE_NOTIFY_START_OF_STREAM:
         decoder->sample_time = -1;

[PATCH 4/4] winegstreamer: Handle null transform in WMA IMediaObject::Flush().

[Conor McCarthy]

From: Conor McCarthy

---
 dlls/winegstreamer/wma_decoder.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/dlls/winegstreamer/wma_decoder.c b/dlls/winegstreamer/wma_decoder.c
index c141b7e26fe..09b2ee647b3 100644
--- a/dlls/winegstreamer/wma_decoder.c
+++ b/dlls/winegstreamer/wma_decoder.c
@@ -897,6 +897,9 @@ static HRESULT WINAPI media_object_Flush(IMediaObject *iface)

TRACE("iface %p.\n", iface);

+    if (!decoder->wg_transform)
+        return DMO_E_TYPE_NOT_SET;
+
     if (FAILED(hr = wg_transform_flush(decoder->wg_transform)))
         return hr;

[Rémi Bernon (＠rbernon)]

> A race condition in Nioh can result in Flush() being called when the transform is not set.

Should we consider adding a CS to these objects?

[Conor McCarthy (＠cmccarthy)]

On Wed Dec  4 00:34:31 2024 +0000, Rémi Bernon wrote:
> > A race condition in Nioh can result in Flush() being called when the
>
> transform is not set.
> Should we consider adding a CS to these objects?

Implement the `Lock()` and `Unlock()` methods? From my understanding of the documentation, all locking should go through these, but they are never called.

[Rémi Bernon (＠rbernon)]

On Wed Dec  4 00:34:30 2024 +0000, Conor McCarthy wrote:
> Implement the `Lock()` and `Unlock()` methods? From my understanding of
> the documentation, all locking should go through these, but they are
> never called.

No, I meant: if there's a race condition does this mean that functions are called concurrently from two different threads? In which case we should maybe add a CS. Or if it's just the consequence of a race condition elsewhere but functions still called from a single thread, these fixes are enough.

[Conor McCarthy (＠cmccarthy)]

On Wed Dec  4 09:07:24 2024 +0000, Rémi Bernon wrote:
> No, I meant: if there's a race condition does this mean that functions
> are called concurrently from two different threads? In which case we
> should maybe add a CS. Or if it's just the consequence of a race
> condition elsewhere but functions still called from a single thread,
> these fixes are enough.

Functions are not called concurrently. The game creates a media object, never sets a transform output type, and eventually calls `Flush()` on the object. For some reason this doesn't happen when tracing is not enabled. Not a race condition that causes invalid behaviour, since it's not invalid to call `Flush()`.

[PATCH 0/4] MR6950: winegstreamer: Handle null wg_transform in several functions. - approved

[Rémi Bernon (＠rbernon)]

This merge request was approved by Rémi Bernon.