SNAPSHOT Drift: Ten Imports, Two Lessons
The CI failure that kicked off this session was a compile error in casehub-work-ledger: cannot find symbol: class ActorType, location: package io.casehub.ledger.api.model. The cause took thirty seconds to find — casehub-ledger#88 had moved ActorType and ActorTypeResolver from that package to io.casehub.platform.api.identity. Both types are shared across multiple repos and belong in the platform module. Our DTOs and audit service still had the old import.
The fix should have been: grep the whole codebase for io.casehub.ledger.api.model.ActorType, fix everything at once, run all tests, push once. What I actually did was fix the ledger module, push, watch CI fail in examples, fix examples, push, watch CI fail again on a transient Testcontainers timeout. Claude grepped for the old package path after the first round and found the remaining two files immediately. Eight imports in the ledger module, two in examples — ten total across three pushes when one would have done. When a package moves, grep is the blast radius tool. CI is not.
The second break was subtler. casehub-ledger#95 had moved the base Flyway migrations from classpath:db/migration to classpath:db/ledger/migration. Every consumer’s test application.properties needed both paths, otherwise V2001+ migrations fail at startup with Table "LEDGER_ENTRY" not found. The error looks like a migration ordering problem. It’s actually a classpath gap — the base tables (ledger_entry, ledger_attestation) were never created because Flyway didn’t scan the right location.
quarkus.flyway.locations=db/migration,db/ledger/migration
What made this hard to catch locally: Maven had already resolved and cached the old SNAPSHOT. Tests passed. CI pulled a fresh copy and hit it immediately. Classic SNAPSHOT drift scenario — the local cache confirms a state that no longer exists upstream.
There was one other mistake worth recording. While updating PLATFORM.md in the parent repo, we were on issue-60-sync-all-repo-lists, not main. I ran git push upstream main. It appeared to succeed — 393ca03..d500aa3 main -> main. What it pushed was local main, not the current branch. Git did exactly what I asked. The SHA range in the output was the tell: it showed where local main was, not where the current branch was. The fix was a cherry-pick to local main. It’s in the garden now.
Full test suite green locally. CI confirmed clean on the rerun.