Consecutive per-frame settle timeouts after which the capture PAUSES
draining. A scene that can never settle (e.g. resident-byte thrash on a
partition that does not fit the budget) would otherwise pay
LOD_SETTLE_TIMEOUT_MS on EVERY remaining frame, silently multiplying the
capture's wall-clock. One successful settle resets the counter.
The pause RE-ARMS: a latched frame still spends a free, non-waiting probe of
the predicate (no rAF, no poll — so a latched frame costs exactly what it did
before the latch fired) and clears the latch the moment that probe reports
settled. Without the re-arm the latch was terminal, and on exactly the scenes
this drain targets — an over-budget adaptive / overview partition, or a
Capture pressed before the initial load finished — the first three frames
would burn their budget, the latch would fire at frame 3, and the remaining
hundreds of frames would be exported with the pre-#1695 behaviour even though
the scene settles seconds later.
Consecutive per-frame settle timeouts after which the capture PAUSES draining. A scene that can never settle (e.g. resident-byte thrash on a partition that does not fit the budget) would otherwise pay
LOD_SETTLE_TIMEOUT_MSon EVERY remaining frame, silently multiplying the capture's wall-clock. One successful settle resets the counter.The pause RE-ARMS: a latched frame still spends a free, non-waiting probe of the predicate (no rAF, no poll — so a latched frame costs exactly what it did before the latch fired) and clears the latch the moment that probe reports settled. Without the re-arm the latch was terminal, and on exactly the scenes this drain targets — an over-budget
adaptive/overviewpartition, or a Capture pressed before the initial load finished — the first three frames would burn their budget, the latch would fire at frame 3, and the remaining hundreds of frames would be exported with the pre-#1695 behaviour even though the scene settles seconds later.