From 06b26535657b4c15e9541282c45db3766909a1ad Mon Sep 17 00:00:00 2001 From: thekevinbot <248217931+thekevinbot@users.noreply.github.com> Date: Fri, 18 Sep 2026 13:12:55 -0400 Subject: [PATCH 1/2] config: re-pin the rust coverage floors to current actuals Measured on 59474c3 with `cargo +nightly llvm-cov --lib --features cli --branch`: lines 94.54, regions 93.67, functions 93.16, branch 79.49. Each floor moves to the highest integer strictly under its actual, per the rule the file's own header states, leaving under a point of slack on every dimension again. No test is added, deleted, or weakened; this measures the tree as it already is. The sub-90 `branch` exception is unchanged in kind -- 72 simply left seven points unguarded. skip-changelog: config-only, no shipped behavior change Fixes #1136 --- testing-conventions.toml | 37 +++++++++++++++++++++++++++++++------ 1 file changed, 31 insertions(+), 6 deletions(-) diff --git a/testing-conventions.toml b/testing-conventions.toml index 0c74cede5..187710fec 100644 --- a/testing-conventions.toml +++ b/testing-conventions.toml @@ -190,9 +190,34 @@ features = ["cli"] # those branches hidden, which is a flattering number rather than a stronger # gate. # -# Actuals at the time of writing: lines 92.46, regions 91.95, functions 91.72, -# branch 72.24. -lines = 92 -functions = 91 -regions = 91 -branch = 72 +# Every dimension ratchets UP for the first time (lines 92->94, regions +# 91->93, functions 91->93, branch 72->79), re-pinned against a fresh +# measurement on 59474c3 (#1136). The numbers moved up because the floors were +# never re-measured after the two downward nudges above, while the tree kept +# gaining unit-reachable code. The refactors of the last stretch pulled pure +# logic OUT of the effectful modules the unit tier structurally cannot reach +# and into modules that are wholly unit-testable -- extension resolution into +# extension_resolution.rs (99.8% regions / 100% branch), RowEvent flattening +# into row_event_flat.rs (100% across the board, #1146), the path vtab's +# arithmetic into path_table.rs (98.8%) -- each landing with its own inline +# tests. That both raises the covered numerator and, for `branch`, adds +# fully-covered branch regions to the denominator, which is why `branch` gains +# the most (7pt) despite reason (2) above still applying. +# +# Nothing here reconsiders the sub-90 `branch` exception: 79 is still short of +# the >=90 ideal for exactly the two reasons documented above, and raising the +# floor to the actual only stops leaving the gap between 72 and 79 unguarded. +# A floor that far under actual does not ratchet against regression, which is +# the whole purpose of the table. +# +# The header's fragility warning applies again in full: slack is back under a +# point on every dimension (0.5 / 0.7 / 0.2 / 0.5), so a PR that merely adds +# well-formed unit tests can lower `branch` and trip the floor. The fix then is +# to nudge the affected floor DOWN to the new actual, not to chase the dip. +# +# Actuals at the time of writing: lines 94.54, regions 93.67, functions 93.16, +# branch 79.49. +lines = 94 +functions = 93 +regions = 93 +branch = 79 From bbb7e4b1033fd36de1609960320000c4edeef45c Mon Sep 17 00:00:00 2001 From: thekevinbot <248217931+thekevinbot@users.noreply.github.com> Date: Fri, 18 Sep 2026 13:18:28 -0400 Subject: [PATCH 2/2] docs(config): trim the rust coverage floor commentary The rationale for the upward move lives in the PR body; the file keeps the fresh actuals and one line on direction. skip-changelog: config comment only, no shipped behavior change --- testing-conventions.toml | 28 ++++------------------------ 1 file changed, 4 insertions(+), 24 deletions(-) diff --git a/testing-conventions.toml b/testing-conventions.toml index 187710fec..436672cfd 100644 --- a/testing-conventions.toml +++ b/testing-conventions.toml @@ -190,30 +190,10 @@ features = ["cli"] # those branches hidden, which is a flattering number rather than a stronger # gate. # -# Every dimension ratchets UP for the first time (lines 92->94, regions -# 91->93, functions 91->93, branch 72->79), re-pinned against a fresh -# measurement on 59474c3 (#1136). The numbers moved up because the floors were -# never re-measured after the two downward nudges above, while the tree kept -# gaining unit-reachable code. The refactors of the last stretch pulled pure -# logic OUT of the effectful modules the unit tier structurally cannot reach -# and into modules that are wholly unit-testable -- extension resolution into -# extension_resolution.rs (99.8% regions / 100% branch), RowEvent flattening -# into row_event_flat.rs (100% across the board, #1146), the path vtab's -# arithmetic into path_table.rs (98.8%) -- each landing with its own inline -# tests. That both raises the covered numerator and, for `branch`, adds -# fully-covered branch regions to the denominator, which is why `branch` gains -# the most (7pt) despite reason (2) above still applying. -# -# Nothing here reconsiders the sub-90 `branch` exception: 79 is still short of -# the >=90 ideal for exactly the two reasons documented above, and raising the -# floor to the actual only stops leaving the gap between 72 and 79 unguarded. -# A floor that far under actual does not ratchet against regression, which is -# the whole purpose of the table. -# -# The header's fragility warning applies again in full: slack is back under a -# point on every dimension (0.5 / 0.7 / 0.2 / 0.5), so a PR that merely adds -# well-formed unit tests can lower `branch` and trip the floor. The fix then is -# to nudge the affected floor DOWN to the new actual, not to chase the dip. +# All four ratchet UP for the first time (lines 92->94, regions 91->93, +# functions 91->93, branch 72->79): recent refactors moved pure logic out of +# the effectful modules the unit tier cannot reach and into wholly +# unit-testable ones. # # Actuals at the time of writing: lines 94.54, regions 93.67, functions 93.16, # branch 79.49.