You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
CLAUDE.md claimed javadoc was "build-enforced: failOnError +
failOnWarnings". Neither was configured anywhere and the plugin was not
bound to any phase, so nothing ran it. That is how the seven broken
references fixed in the previous commit accumulated unnoticed.
Binds javadoc-no-fork to verify with failOnError=true, so a dangling
[Type#member] link fails the build. verify rather than test, so
contributors running ./mvnw test do not pay for it; the whole reactor
costs ~3s. Confirmed to bite by pointing StructArray at a non-existent
type and watching the build go red.
failOnWarnings stays off globally: 313 warnings predate the check
(performance 100, reader 82, writer 42, fbs-gen 35, csv 34, parquet 14,
cli 6), and gating on them would mean a 313-item cleanup before anything
else can merge. core and fsst are already at zero, so their POMs raise
failOnWarnings to hold that line — a ratchet other modules can join as
they are cleared, rather than one flag day. Confirmed to bite with a
probe class carrying an undocumented parameter.
CLAUDE.md now describes what is actually enforced, where, and why the
two levels differ. It also records that bare `./mvnw javadoc:javadoc`
fails on dependency resolution rather than javadoc (no lifecycle phase
runs, so reactor jars are absent and install is forbidden), and that a
target the module cannot see is an error too — the writer/reader and
fbs-gen/core.fbs cases from the previous commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
0 commit comments