[Bug 774649] glupload: expose memory:DMABuf capsfeature

[GStreamer (GNOME Bugzilla)]

https://bugzilla.gnome.org/show_bug.cgi?id=774649

--- Comment #5 from Matthew Waters (ystreet00) <ystreet00 at gmail.com> ---
(In reply to Julien Isorce from comment #3)
> When the dmabuf is mappable the user does not want the gst elements to work
> in pass-through mode (and these elements do not want to have conscience this
> is dmabuf) in between the export element (upstream elem) and the importer
> element (downstream elem).

Huh?

"the gst elements to work in pass-through mode"? What are we talking about
here?  Are we talking about other non-dmabuf aware elements like
videoconvert/scale/etc that will passthrough caps-features?  Then I bring up
the relevance to dmabuf.  Unless you still want them to work as usual?

"and these elements do not want to have conscience this is dmabuf"

"conscience this is"?  Maybe you mean "knowledge of"?, "code for"

Maybe an example would help here?

> Having it after systemMem tells the exporter to follow the rule above. If we
> put it after then it will not let the exporter the chance to push the dma
> buffers without the caps feature.

Assuming "it" = sysmem caps feature.

So, you want to add a caps feature but not use the caps feature?  That sounds
backwards.  Why can't the source choose based on the available features?

My main problem with this argument is that you're moving who chooses the caps
feature into an element that doesn't actually know very much about the
properties of the dmabuf, sysmem-mappable or not.  The source has a much better
idea on what is possible with the dma-buf they are pushing and has the ultimate
say on what they produce.

(In reply to Nicolas Dufresne (stormer) from comment #4)
> For clarity, dmabuf passed as SystemMemory can fallback if the import fails,
> while dmabuf with the CapsFeature set can't. You cannot trust the result of
> mapping the dmabuf when the caps feature is set. It may fail, or it may work
> and give you something black (e.g. ION secured surfaces)

Sure.  That doesn't really explain why the dmabuf capsfeature is after the
sysmem one though.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.