[RyanTheTide]

A second external display on the J616s (an MSI MPG 325CQRXF, Type-C to Type-C on the second external processor, DP-2) came up with a 0-byte EDID. GNOME showed no monitor name and saw none of its HDR or colour data, while the dock Acer on DP-1 had its 256-byte firmware-derived EDID.

Cause. The firmware-derived EDID path (from #530 and #553) works the same on both external processors. On DP-2 the parser rejected the sink's DisplayAttributes outright. The MSI reports its red primary as x 0.6904, y 0.3203. Those are the 10-bit values 707 and 328 from its own EDID, and they sum to more than 1. The parser refused any chromaticity summing above 1, so the whole dictionary was dropped and nothing was built. Feeding the captured DisplayAttributes blob through the host-selftest parser reproduces it: parse_display_metadata() returns -EINVAL at Chromaticity.

Fix.

• Primaries are passed on as the sink reports them, as a real EDID would carry them. The builder still requires each coordinate to be representable, the primaries to form a triangle and the white point to lie inside it. The white point, which the gamma belongs to, must still sum to 1 or less. The MSI's primaries pass those checks.
• The same sink reports Luminance without Min. A missing minimum now reads as 0 and a missing frame-average as the maximum, instead of the whole range being dropped (which would have lost its HDR block luminance).

Tests.

• apple-dcp host selftests: 80/80. The rate fixture gains a sink with the MSI's red primary and no Min luminance. It checks that the primaries survive into the EDID as 707/328 and that min is 0. The rejected-chromaticity case moves to the white point.
• The captured MSI DisplayAttributes now parse.
• Hardware check on the J616s with the MSI on DP-2: pending a slot. The module builds cleanly against the fleet4 kernel.

🤖 Generated with Claude Code

[RyanTheTide]

Hardware check on the J616s (fleet4 + this commit as a test module): DP-2 (MSI MPG 325CQRXF, Type-C on the second external processor) now logs using firmware-derived EDID metadata, its EDID is 256 bytes (name MPG 325CQRXF, serial DC1H514A00211, primaries 0.6904/0.3203 red as the sink reports them, HDR block), GNOME names it Microstep 32" and offers bt2100; 2560x1440@280 and 280+vrr are listed; HDR at 280 Hz applies as PQ BT.2020 YCbCr 4:4:4 (element 124) with no flip failures. DP-1 (Acer) and the internal panel are unchanged. Machine restored to stock afterwards.