Repository navigation
Weave 1.14.0
Fixed
-
quality.confconfig format —[exclude]parser and[classify]parser both now strip
inline#comments. The README incorrectly showed INI-stylepatterns = \n valmulti-line
syntax; corrected to one-glob-per-line. -
File classification heuristics table — README listed
migrations/,vendor/,
node_modules/,.min.jsas generated patterns; none exist inclassification.py. Table now
matches the code exactly (4 errors across 3 rows corrected). -
ScanMetadocstring — said "Two scans retained" butdb.pyuses_MAX_SCANS=5for trend
data and_FILES_SCANS=2for raw file data. Docstring anddb.pyheader updated. -
fn_ccEAV key collisions —FunctionCC.to_eav_row()keyed onfn_cc:<name>, colliding on
(path, scan_id, metric)when two methods share a name in different classes within one file. Key
is nowfn_cc:<name>@<line_start>. Python identifiers cannot contain@so the split is
unambiguous. -
Gini penalty dead-letter at N=3 — the concentration penalty threshold 0.7 is mathematically
unreachable for exactly 3 functions (max Gini = 0.667). Minimum raised toN >= 4, where max Gini
= 0.75. -
make checkon Python-3-only systems —PYTHON ?= pythonfailed on Debian/Ubuntu after
Python 2 EOL (nopythonsymlink). Changed topoetry run python.PYLINTaligned to
$(PYTHON) -m pylintfor consistency.
Added
scanner_versioninscan_meta— each scan now records the Weave version that produced it
(_migrate_v4). If a scan detects the previous scan used a different version, it automatically
forces a full re-scan so no stale metrics are carried forward. No manualwv quality resetneeded
after upgrading.
Documentation
-
Known Limitations — two new entries: hotspot threshold instability (absolute cutoff on min-max
normalised scores) andwv quality diffgit-stats caveat (previous score retroactively
recalculated with current churn data). -
Proposal docs superseded —
PROPOSAL-wv-quality-depth.md,PROPOSAL-wv-quality-perf.md, and
PROPOSAL-wv-mccabe-review.mdeach carry aSUPERSEDEDbanner pointing to
scripts/weave_quality/README.mdas the single source of truth.