[whoisaldo]

What changed

Three small fixes to iPad support-log lines, one commit each.

Each MetricKit frame shows its address (Core/CrashReporting.swift)

Early MetricKit payloads can put the binary's load address in offsetIntoBinaryTextSegment, and then the offset alone can't place the frame. Each frame in frames= is now binary+0x<offset>@0x<address>, using the frame's address from the call stack tree (0 when MetricKit leaves it out, as it does for a missing offset).

The final scrub keeps the address: the line formats it from a UInt64, so it is at most 16 hex digits. The exemption is a lookbehind for +0x<offset>@, which only the frame list produces, because binary names are tokens without @. The same hex in a binary name is still <id>. Eight frames with 64-character names and maximum offsets and addresses stay under 2 KiB.

The termination code survives the scrub

64-bit codes such as EXC_GUARD and JETSAM reasons have 16 hex digits with a letter. The scrub turned them into code:<id>, so the reason the app was killed was lost. The hex scrub now skips the value after termination=namespace:<NAMESPACE> code: and nowhere else. termination() has already reduced that value to 0x plus at most 16 hex digits, or at most 20 decimal digits. The line has no code= field (exception_code= is a decimal Int64, which the scrub never matched).

Before and after, with synthetic data:

before: … frames=libsystem_kernel.dylib+0x1a2b00,EternalMonitor+0x104a2c000,EternalMonitor+0x1a2b20 termination=namespace:GUARD code:<id>
after:  … frames=libsystem_kernel.dylib+0x1a2b00@0x1e8f3ab00,EternalMonitor+0x104a2c000@0x104a3d2b4,EternalMonitor+0x1a2b20@0x104bcf2c8 termination=namespace:GUARD code:0xc00000000000ab12

The rest of the line (MetricKit crash app=0.3.4 build=370 os=ipados:26.1 os_build=23B85 at=2026-10-26T07:33:20Z exception_type=12(EXC_GUARD) exception_code=0 signal=9(SIGKILL) objc_exception=- thread=attributed) is unchanged.

A failed USB connection is logged once (Network/FramedLink.swift)

When the connection reported .failed, FramedLink logged NWConnection failed: …. fail() then logged the same failure again: through the connection that took the link (state=… USB connection failed: …), or as the link's own line while it waited for its preamble. The send path did the same with NWConnection send: …. #126 removed this extra line from the receive path; the other two now log only through fail(). The removed lines added nothing, because localizedDescription on an NWError already carries the errno and its text ("… (Network.NWError error 54 - Connection reset by peer)").

In practice the .failed handler rarely ran: with a receive outstanding, a reset reaches the receive callback first, and its fail() clears the state handler. FramedLink now takes its support-log writer as an init parameter (default SupportDiagnostics.shared), and the state switch moved into handle(_:), so a test can drive it.

Tests

• CrashReportingTests: existing expectations now include each frame's @0x<address>, and a frame without one shows @0x0.
• testEachFrameShowsItsAddress: a 16-hex-digit address is kept while the same digits in a binary name become <id>. Eight maximum-length frames stay under 2 KiB.
• testALongTerminationCodeIsKept: a 16-hex-digit GUARD code is kept in termination=, and the same hex as a binary name is still scrubbed. A bare code:<hex> outside the termination field is still scrubbed, and a JETSAM code survives.
• FramedLinkTests.testAFailedConnectionIsLoggedOnce: a .failed state gives one error [usb] USB connection failed: … line while no connection owns the link. With an owner it gives no line and one onError, and onClosed fires once either way.
• scripts/test_ios.sh -only-testing:EternalMonitorTests on an iPad Pro 11-inch (M4) simulator: 349 tests, 0 failures on iOS 26.3 / Xcode 26.3, and 349 tests, 0 failures on iOS 18.6 / Xcode 16.4.
• The new crash lines don't match the session_summary contract regex (they are crash events, not summaries), SupportDiagnostics.redact leaves them unchanged, and they still start MetricKit crash , which the daily check matches.

Release notes

iPad crash reports now include more detail to help us find the cause, and a failed USB connection is no longer logged twice.