pub static SCAN_PATH_ENTERED: AtomicU64Expand description
v7.39 (round 485) — how many projected rows the single-table scan builds, and how many of those the DISTINCT probe throws away again.
The round-485 profile of SELECT DISTINCT g FROM h ORDER BY g put
21 % of all samples in malloc/free called straight from the scan
closure. The closure’s one per-row allocation is the projected
Vec<Value>, and under DISTINCT most of those are discarded a few
instructions later — but “most” is a guess until it is a number, so
these count it. (Round 480 was spent acting on an inference about a
branch that turned out never to run.)
v7.39 (round 488) — reachability counters for round 487’s projection
binding. The interleaved panel says round 487 costs group_500k 13 %,
and a never-called-function probe rules out code layout — so the
question is whether that shape reaches this code at all, which is a
number, not an inference.