Publish: M153, the sender slice, and what the rollover cost
Branch selection had M152 as the chosen stable milestone, which it stopped
being on 2026-09-19, and gains a section on what a rollover does to a project
mid-flight.
Troubleshooting gains the three M153 failure modes, none of which was findable
from its symptom: the Windows SDK pin mismatch, disk exhaustion presenting as a
linker crash that says nothing about disk, and an Android archive that builds
clean and is then rejected by its consumer over bytecode level. Two of those
entries say what not to do, because both were tried.
Consuming the artifacts gains how to check that bytecode level, and the Mac
Catalyst slice writing PrivacyInfo.xcprivacy one directory too deep.
The Interop ABI page catches up with the sender parameter and capture format
work, including the measurement that contradicted the issue describing it:
scale_resolution_down_by survives a capturer that ignores adaptation, because
the encoder scales too, so outbound-rtp frameWidth is no evidence about the
source. max_framerate is the one only the source can apply.
Slice one is complete
All twenty-five interop functions are implemented and covered by three
tests.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The frame sink is implemented
Twenty-three of twenty-five functions now work. Records the sink's
lifetime rules and the measured frame rate.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Peer connection and negotiation are implemented
Adds the Winsock note: without WinsockInitializer every UDP socket fails
quietly, ICE reaches "checking" and stops there. Records
WEBRTC_INTEROP_VERBOSE as the way to see the ICE layer at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Devices and tracks are implemented
Video enumeration and both track creators work; audio enumeration is
deferred with the reason recorded. rtc_video_track_create gained a label
parameter after the device id turned out to leak into the SDP msid.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Windows interop builds; correct the exception guidance
The ABI page prescribed try/catch, which cannot compile under WebRTC's
-fno-exceptions and is unnecessary anyway. Also records the keep-both
decision for webrtc.dll and WebRtcInterop.dll, and gives Windows a partial
tier 3 in the figure.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Record the verified ABI layout
Interop.h now exists and compiles; adds the measured struct sizes and
offsets, notes every struct is blittable, and generalises the ownership
rule to cover handles delivered through callbacks.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Add Interop ABI
The C contract between WebRtcInterop.dll and the Windows binding:
conventions, ownership rule, threading, and the twenty-three functions of
slice one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>