Published and verified green on the pod rig: Condenser/CondenserState,
Generator/GeneratorState, Emitter/LaserOn, ControlsMapper/TargetRangeExponent,
plus the Torso and Myomers tables from the previous pass.
Reservoir/ReservoirState CRASHES the pod on the DOS extender, and a bisect
pins it to exactly that change: revert it and the run returns to a clean
Fail, re-apply it alone and the crash returns. It is reverted; the tree is
green and the ladder is blocked there.
The important lesson is general: AttributeWatcherOf<T> does
currentValue = *(T*)attributePointer AT CONSTRUCTION, so a published name is
read the moment its watcher is built. My earlier staging note claimed the
provisional types could not matter because nothing drives the values yet --
that is wrong, and this is the counterexample.
Ruled out by measurement and recorded so they are not retried: the
AlarmIndicator-vs-int type (a plain int got further, 232 -> 749 bytes of log,
but still crashed), static-init order (reservr.obj sorts last, after heat),
an id gap (contiguous at HeatSink::NextAttributeID), and the gauge rig (same
binary runs clean there -- only the pod/arena context faults).
Crash signature for whoever picks it up: 0044B4AD, mov eax,[edx+0x18] then
call [eax+4], EAX=0x15, fault at 0x19 -- a small integer called through as an
object, i.e. the AttributeIndexSet::Build uninitialised-slot pattern.
Condenser is the control: same base class, plain int, no crash.
Also fixed on the way: staged members must be appended at the END of a class
(offset-sensitive readers exist -- condenserNumber is reached as
master+0x1d4) and never added to a resource struct, which is a wire format.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CydandClaude Fable 5
0d97d8c38b