wegamekinglc / calc-flow / 35497788794
93%

DEFAULT BRANCH: main
Ran
Jobs 3
Files 200
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 93.125%. Remained the same
35497788794

push

github

web-flow 
test: cover native stream join evidence (DAL-290) (#303)

* test: cover native stream join in benchmark suite (DAL-290)

The engine-side stream_join output batching (#298, #300) removed the
row-per-message bottleneck that kept `join` out of the native streaming
column. Measure it through the bounded temporal join: the dimension side
completes at the stream origin and the inclusive `before` bound spans the
workload, so every quote row matches exactly its symbol's factor row and
the suite's shared join oracle applies unchanged.

The 10M tier of this one combination stays out of the catalog: the join
retains one state row per matched input row for the whole run, so a
sample needs roughly 200 seconds and 44 interleaved executions cannot
fit the 180-minute shard budget. The exclusion is explicit — a catalog
constant honored by the baseline-id parser and a marked
`excluded (suite budget)` cell in the cross-library table — instead of a
silent skip, and the smaller decades carry full evidence.

Co-authored-by: multica-agent <github@multica.ai>

* test: cap native stream join evidence at 100k rows (DAL-290)

User-directed pacing constraint (2026-09-20): the stream join's
per-sample performance above the 100k tier (roughly 5 s at 1M and 200 s
at 10M, because the join retains one state row per matched input row for
the whole run) would slow the whole suite's cadence. Replace the 10M
budget exclusion with STREAM_JOIN_MAX_ROWS so the 1M and 10M tiers stay
unsupported in the catalog and render as unsupported cells, with the
performance reason documented. The chunk-boundary correctness fixture
builds its case directly so join keeps its multi-batch finalization
coverage at the synthetic 64,001/128,000 sizes, which are correctness
probes rather than suite tiers.

Co-authored-by: multica-agent <github@multica.ai>

* fix: clear Copilot and Codacy findings on the join benchmark (DAL-290)

Copilot: the stream test double only opened, readied, and EOF-handled one
source, s... (continued)

105295 of 113068 relevant lines covered (93.13%)

103334.12 hits per line

Coverage Regressions

Lines Coverage ∆ File
1
94.99
0.11% crates/calc-flow-python/src/continuous.rs
1
99.25
-0.37% crates/calc-flow-python/src/runtime/dispatch.rs
1
95.23
-0.11% crates/calc-flow-python/src/runtime.rs
1
93.32
-0.02% crates/calc-flow/src/runtime/streaming/progress/driver.rs
1
94.75
-0.01% crates/calc-flow/src/runtime/streaming/runner.rs
Jobs
ID Job ID Ran Files Coverage
1 studio - 35497788794.1 10
92.22
GitHub Action Run
2 rust - 35497788794.2 149
92.97
GitHub Action Run
3 python - 35497788794.3 41
95.32
GitHub Action Run
Source Files on build 35497788794