wegamekinglc / calc-flow / 35684523595

94%

Build Type

push

github

Committed by web-flow

Commit Message

perf: bound Join materialization for DAL-300 (#321)

* perf: materialize DAL-300 Join output in bounded chunks

Preflight every matched range before emitting independently allocated chunks so wide fan-out output no longer retains one full output buffer. Preserve atomic admission, output sequence checks and checkpoint recovery.

Maintain a bounded-edge Join workload with allocation, RSS and slow-sink observations, retaining new target coverage in the unified collector.

* fix: rebuild product libraries between DAL-300 benchmark sides

The sealed native check reproduced a shared-target collision: Cargo reused the baseline calc-flow library when linking the candidate. Invalidate only product rlib/rmeta artifacts once per source, keeping dependencies and sealed executables.

* fix: restore DAL-300 benchmark validation gates

Include the maintained Join target in the exact release inventory assertion.
Extract case normalization so the collector stays within the complexity
limit while preserving every evidence validation and diagnostic field.

* refactor: resolve DAL-300 native quality gates

Extract the nested Join range budget predicate without changing preflight
order, saturating arithmetic or recursive payload validation.

Document and narrowly mark the benchmark argv audit finding: arguments
select local modes and report destinations, never authorization decisions.

* docs: explain Join chunking and measurement boundaries

Document complete output preflight, independent chunk materialization and
checkpoint replay after interrupted delivery. Align the maintained command
and report contract while preserving separate memory costs and the original
source identities of paired, candidate-only and oracle evidence.

* fix: scope Join costs and benchmark evidence for DAL-300

Cache preflight costs only for matched indices and scope dependency stamps to each compiled benchmark target. Reject invalid workload and output row counts while preserving aggregate prove... (continued)

Coverage Stats

494 of 512 new or added lines in 3 files covered. (96.48%)

1 existing line in 1 file now uncovered.

106987 of 114766 relevant lines covered (93.22%)

91418.7 hits per line