Fix sink pad ordering #213

Manishearth · 2019-02-27T15:15:13Z

Fixes #210

Based on #208

This implements a version of the fix detailed in #210 (comment)

Instead of unlinking/relinking, we simply keep new streams unlinked until we decide to trigger an offer, at which point we link them to new request pads. The answerer side keeps them unlinked until it gets a remote offer, at which point it also requests pads and links them to the correctly numbered request pads.

A missing piece was that the payload number also has to match, so we keep track of that too.

r? @jdm @slomo (last two commits only)

Manishearth · 2019-02-27T15:24:49Z

I'm done for the day, but I do need to write some more comments for this. Will do tomorrow.

Manishearth · 2019-02-27T15:25:28Z

backends/gstreamer/src/webrtc.rs

+                    //
+                    // An alternate fix is to sort pending_streams according to the m-line index
+                    // and just do it in order. This also seems brittle.
+                    webrtc.get_request_pad("sink_%u").unwrap();


I feel like there's an underlying bug here, this should be unnecessary

cc @slomo

jdm · 2019-03-01T19:42:24Z

backends/gstreamer/src/media_stream.rs

@@ -21,7 +21,7 @@ lazy_static! {
            &[
                ("media", &"video"),
                ("encoding-name", &"VP8"),
-                ("payload", &(96i32)),
+                // ("payload", &(96i32)),


Remove these commented versions?

jdm · 2019-03-01T19:42:24Z

backends/gstreamer/src/webrtc.rs

    thread: WebRtcThread,
    signaller: Box<WebRtcSignaller>,
+    /// All the streams that are connected to the 


jdm · 2019-03-01T19:42:24Z

backends/gstreamer/src/webrtc.rs

+    ///
+    /// When we create an offer, we're controlling the pad order, so we set request_new_pads
+    /// to true and forcefully link all pending streams before generating the offer.
+    fn link_stream(&mut self, mut boxed_stream: Box<MediaStream>, request_new_pads: bool) {


Can we use an enum instead of a boolean here?

jdm · 2019-03-01T19:42:24Z

webrtc/src/thread.rs

@@ -90,6 +90,7 @@ pub enum InternalEvent {
    OnNegotiationNeeded,
    OnIceCandidate(IceCandidate),
    OnAddStream(Box<MediaStream>),
+    DescriptionAdded(SendBoxFnOnce<'static, ()>, bool, SdpType),


Maybe use an enum instead of a bool here?

jdm · 2019-03-01T19:42:24Z

backends/gstreamer/src/webrtc.rs

+        let element = stream.src_element();
+
+        if let Some(idx) = idx {
+            if idx >= self.request_pad_counter {


Is there any risk that we could call link_stream with !request_new_pads but end up running this code that requests pads? Would that be bad? I find the logic here a little bit obtuse.

Yes, this code requesting new pads is fine, request_new_pads is about requesting new pads when there is no remote m-line: in that case we normally want to just buffer it up for later, except for when we're making an offer, at which point we want to flush all buffered streams.

The behavior of this function with !request_new_pads is "request pads only if they have been set up by the remote already", which is too long to fit in a variable. I can document this more explicitly.

bors-servo · 2019-03-01T21:24:05Z

☔ The latest upstream changes (presumably #209) made this pull request unmergeable. Please resolve the merge conflicts.


        Add Debug to BundlePolicy


        Use correct webrtc sink and payloads

Manishearth · 2019-03-02T06:28:34Z

@bors-servo r=jdm

bors-servo · 2019-03-02T06:28:34Z

📌 Commit f0e7554 has been approved by jdm

bors-servo · 2019-03-02T06:28:37Z

⌛ Testing commit f0e7554 with merge fa92926...


        Auto merge of #213 - Manishearth:sink-pad-ordering, r=jdm

@jdm

Fix sink pad ordering Fixes #210 Based on #208 This implements a version of the fix detailed in #210 (comment) Instead of unlinking/relinking, we simply keep new streams unlinked until we decide to trigger an offer, at which point we link them to new request pads. The answerer side keeps them unlinked until it gets a remote offer, at which point it also requests pads and links them to the correctly numbered request pads. A missing piece was that the payload number also has to match, so we keep track of that too. r? @jdm @slomo (last two commits only)

bors-servo · 2019-03-02T06:41:08Z

☀️ Test successful - checks-travis
Approved by: jdm
Pushing fa92926 to master...

slomo · 2019-03-04T09:59:45Z

@Manishearth You are mentioning the wrong guy in your PR.

Manishearth · 2019-03-04T10:01:11Z

Oh, sorry about that!


        Auto merge of #273 - Manishearth:sdpmedia, r=ceyusa

@sdroege

Get rid of unsafe SDP media workaround Upstream had [a bug](https://gitlab.freedesktop.org/gstreamer/gstreamer-rs/issues/189) which we worked around in #213 by directly accessing the underlying GObject data. This is no longer necessary after the gstreamer update, pulling in #213 r? @sdroege @ferjm

Manishearth force-pushed the Manishearth:sink-pad-ordering branch from c413cf5 to 957a4cd Feb 27, 2019

Manishearth reviewed Feb 27, 2019

View changes

Manishearth force-pushed the Manishearth:sink-pad-ordering branch 3 times, most recently from 378076b to 1e15743 Feb 28, 2019

jdm reviewed Mar 1, 2019

View changes

Manishearth force-pushed the Manishearth:sink-pad-ordering branch 3 times, most recently from 4835ab0 to cec05f0 Mar 2, 2019

jdm approved these changes Mar 2, 2019

View changes

Manishearth added 2 commits Feb 27, 2019

Add Debug to BundlePolicy

b76b8a0

Use correct webrtc sink and payloads

Loading status checks…

f0e7554

Manishearth force-pushed the Manishearth:sink-pad-ordering branch from cec05f0 to f0e7554 Mar 2, 2019

bors-servo merged commit f0e7554 into servo:master Mar 2, 2019
2 checks passed

2 checks passed

Travis CI - Pull Request Build Passed
Details

homu Test successful
Details

Manishearth deleted the Manishearth:sink-pad-ordering branch Mar 4, 2019

Manishearth mentioned this pull request Jun 26, 2019

Get rid of unsafe SDP media workaround #273

Merged

servo / media

Fix sink pad ordering #213

Fix sink pad ordering #213

Manishearth commented Feb 27, 2019

Manishearth commented Feb 27, 2019

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

bors-servo commented Mar 1, 2019

Manishearth commented Mar 2, 2019

bors-servo commented Mar 2, 2019

bors-servo commented Mar 2, 2019

bors-servo commented Mar 2, 2019

slomo commented Mar 4, 2019

Manishearth commented Mar 4, 2019

servo / media

Join GitHub today

Fix sink pad ordering #213

Fix sink pad ordering #213

Conversation

Manishearth commented Feb 27, 2019

Manishearth commented Feb 27, 2019

This comment has been minimized.

Manishearth Feb 27, 2019

This comment has been minimized.

jdm Mar 1, 2019

This comment has been minimized.

jdm Mar 1, 2019

This comment has been minimized.

jdm Mar 1, 2019

This comment has been minimized.

jdm Mar 1, 2019

This comment has been minimized.

jdm Mar 1, 2019

This comment has been minimized.

Manishearth Mar 2, 2019

bors-servo commented Mar 1, 2019

Manishearth commented Mar 2, 2019

bors-servo commented Mar 2, 2019

bors-servo commented Mar 2, 2019

bors-servo commented Mar 2, 2019

slomo commented Mar 4, 2019

Manishearth commented Mar 4, 2019