Skip to content

Add bdt cleanup: prune merged worktrees and orphaned pgdevkit test DBs - #22

Merged
aersam merged 11 commits into
mainfrom
cleanup-worktrees-testdb
Sep 17, 2026
Merged

Add bdt cleanup: prune merged worktrees and orphaned pgdevkit test DBs#22
aersam merged 11 commits into
mainfrom
cleanup-worktrees-testdb

Conversation

@aersam

@aersam aersam commented Sep 14, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Summary

  • New bdt cleanup worktrees command: finds git worktrees merged into main/test across a directory tree and offers removal, dropping each worktree's pgdevkit test DB(s) along with it
  • New bdt cleanup orphaned-dbs command: finds pgdevkit-managed test DBs whose worktree no longer exists, and offers removal
  • New bdt cleanup db command: drops the current worktree's own pgdevkit test DB(s) in place, without touching the worktree itself -- run from inside the worktree, gated by an interactive confirmation unless --confirm is passed
  • New bdt cleanup worktree command: removes one specific still-live worktree (regardless of merge status) plus its pgdevkit test DB(s) in one go -- also run from inside it, same --confirm/interactive-prompt gate; refuses the main checkout, a protected branch (main/test), a locked worktree, or a dirty one (submodules included); removal reuses the same --force retry as cleanup worktrees so a worktree containing submodules is handled correctly, covered by a real submodule integration test
  • Safety net across every command that drops a database (worktrees, orphaned-dbs, db, worktree): a non-local --pg-host (anything other than localhost/127.0.0.1/::1) now always requires a separate interactive "type yes" confirmation that no flag can bypass, not even --yes/--confirm -- a script/agent can never drop a database on a shared/remote Postgres instance without a human actually confirming it. Non-interactive input (no stdin, e.g. CI) fails closed.
  • DB-naming/orphan-detection is delegated entirely to pgdevkit (workspace_db_names()/find_orphaned_dbs(), see Add find_orphaned_dbs() / clean_testdb(orphaned=True) pgdevkit#34/Enforce conventional commits #35) instead of reimplementing the naming algorithm here -- the one thing this PR started out doing that we specifically wanted to avoid duplicating
  • db_nested_projects (e.g. MDMApp's akeneo_editor/ sharing its parent worktree's branch) stays bdt's own concern via [tool.bdt.worktree], since pgdevkit has no notion of a wholly separate nested project
  • Bulk commands (worktrees, orphaned-dbs) stay non-interactive by design (--yes to act, otherwise dry-run print), matching this codebase's existing bdt issue delete --yes convention; the two single-target commands (db, worktree) instead default to an interactive confirmation, since they're meant to be run by hand against one worktree at a time (--confirm opts back into non-interactive use)
  • Version bumped to 0.17.0

Notes

  • Temporarily depends on pgdevkit via a pinned commit SHA ([tool.uv.sources]) until pgdevkit#35 (workspace_db_names()) merges and ships as 0.7.1 on PyPI -- remove that override once it does.
  • requires-python moved >=3.11 to >=3.14 since pgdevkit requires 3.14; pgdevkit's testdb module also unconditionally pulls in its db extra's heavy dependency chain (psycopg-pool/pydantic) even though this PR only needs the DB-naming/testdb pieces -- noted upstream, not fixed here.

Test plan

  • 366 tests pass locally (uv run -m pytest -q), including a real git submodule integration test for cleanup worktree
  • ruff check and ty check clean
  • Multiple /code-review passes addressed, including a broken --pg-user env fallback, missing host/password on the DROP step, unescaped SQL interpolation of DB names, and a slugification mismatch in the caution-DB heuristic -- all verified against this sandbox's real Postgres, not just mocks

🤖 Generated with Claude Code

aersam and others added 3 commits September 14, 2026 19:13
@aersam @claude
chore: start work on worktree+orphaned-test-db cleanup subcommand
1f6ce3d 
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@aersam @claude
feat(worktree): add bdt cleanup worktrees / orphaned-dbs subcommands
a06d3f7 
Ports a personal script's git-worktree + pgdevkit test-DB cleanup logic into
bdt as a proper, tested subcommand: `bdt cleanup worktrees` prunes worktrees
merged into main/test (and their pgdevkit test DB(s)), and `bdt cleanup
orphaned-dbs` sweeps for test DBs whose worktree is already gone. Both are
non-interactive (require --yes), matching `bdt issue delete`'s pattern since
bdt is also invoked by AI-agent callers. Per-repo sibling/nested DB
conventions (e.g. CCMT2's _onetrade, MDMApp's _akeneo/akeneo_editor) are
configured via `[tool.bdt.worktree]` in that repo's own pyproject.toml rather
than hardcoded here, so other repos can add their own later.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@aersam @claude
fix(cleanup): address code-review findings on psql/pg-user defaults a…
aba4017 
…nd messaging

- psql calls now fail with a friendly "not found on PATH" message (matching
  the existing git/az/gh pattern) instead of a raw FileNotFoundError.
- --pg-user's "current OS user" default now actually resolves via
  getpass.getuser() when PGUSER/USER/LOGNAME are all unset, instead of
  silently substituting the literal role "postgres".
- `bdt cleanup orphaned-dbs` now distinguishes "no repos found under root"
  from "repos scanned, no orphans found", matching `cleanup worktrees`.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@aersam
aersam force-pushed the cleanup-worktrees-testdb branch from 0a447d4 to aba4017 Compare September 14, 2026 17:14
aersam and others added 4 commits September 14, 2026 20:16
@aersam @claude
Delegate testdb naming/orphan-detection to pgdevkit instead of reimpl…
3b31cc0 
…ementing it

testdb.py's slugify/workspace_db_name/expected_db_names and worktree.py's
own live-worktree-vs-Postgres diffing reimplemented pgdevkit's naming
algorithm "without ever importing pgdevkit itself" (per testdb.py's old
module docstring) -- a drift risk called out explicitly when that code
was written. pgdevkit 0.7.1 now exposes find_orphaned_dbs()/
workspace_db_names() directly, so this replaces the reimplementation with
real calls into pgdevkit:

- pgdevkit>=0.7.1 (with the [db] extra, needed to import pgdevkit.testdb
  at all -- see testdb.py's comment) is now a real dependency. Requires
  bumping devtools' own requires-python to >=3.14 (pgdevkit's own
  minimum) and the CI Python matrix accordingly.
- [tool.bdt.worktree].db_sibling_suffixes is gone; that's now
  [tool.pgdevkit].extra_db_suffixes in the *consuming* repo (pgdevkit's
  own config key, folded into its own naming/orphan-detection).
  db_nested_projects stays bdt's own concern (pgdevkit has no notion of a
  wholly separate nested project); verified against the real MDMApp
  checkout that pgdevkit's load_config() falls back to the directory name
  when a nested project's own pyproject.toml has no [tool.pgdevkit]
  section at all -- exactly akeneo_editor's actual current state -- so no
  changes needed there for the nested lookup itself to work.
- bdt cleanup orphaned-dbs no longer takes --remote (pgdevkit's orphan
  detection only cares about live-worktree existence, not merge status);
  its per-repo diffing is entirely pgdevkit.testdb.find_orphaned_dbs()
  now, fanned out across discovered repos (and configured nested
  projects) instead of bdt's own by-hand diff.
- --pg-port/--pg-user now default to pgdevkit's own test-container
  port/user instead of Postgres' standard 5432/the OS user: dropping is
  still bdt's own psql-based DROP DATABASE (pgdevkit has no public
  "drop this arbitrary already-known name" call, and its plain
  clean_testdb() doesn't fold in extra_db_suffixes), but the listing half
  now always goes through pgdevkit's own PGDEVKIT_TESTDB_*-driven
  connection -- defaulting the two halves differently risked silently
  targeting different Postgres instances.
- Fixed a latent bug surfaced along the way: has_pgdevkit_project (nee
  read_pgdevkit_project) treated an entirely empty but present
  [tool.pgdevkit] section as "no section" (empty dict is falsy); fixed to
  check for absence explicitly.
- require_psql() (cli_tools.py) was dead code with testdb.py's _run_psql
  reimplementing the identical check inline; wired it in instead.

Tests reworked throughout: what tested the removed naming/diffing
reimplementation is gone; what's left covers bdt's own remaining
logic (db_nested_projects discovery, caution-DB flagging, CLI
option wiring) plus a few real (unmocked) calls into pgdevkit's naming
for the nested-project nuance specifically, since that's the part most
likely to silently drift again.

Temporary: [tool.uv.sources] pins pgdevkit to its own
testdb-cli-orphaned-followups branch, since 0.7.1 isn't on PyPI yet
(that PR is open but unmerged). Remove once it's merged and published.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@aersam @claude
Pin the temporary pgdevkit uv.sources override to a commit SHA, not a…
87a3cc5 
… branch

/code-review flagged the branch-name pin as fragile: if
testdb-cli-orphaned-followups gets force-pushed, rebased, or deleted
(e.g. after its PR merges and the branch is cleaned up), uv sync/CI here
would start failing or silently resolve different code with no way to
reproduce what was actually reviewed/tested. Pinning to the exact commit
reviewed here avoids that; still temporary until pgdevkit 0.7.1 publishes.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@aersam @claude
fix(cleanup): fix broken --pg-user fallback, missing host/password on…
3c74b4b 
… DROP, SQL escaping, and caution-check slugging

Fixes four issues surfaced by an independent code-review pass on the
pgdevkit-delegation refactor, none of which the reviewing agent's own
"finding fixed" summary actually addressed in the pushed code:

- `--pg-user`'s envvar list included USER/LOGNAME, which are set on
  virtually every real shell, so `pg_user or pgdevkit_constants.USER`'s
  fallback to pgdevkit's own test-container user never fired in practice.
  Now only PGUSER feeds it.
- `drop_database`/`_run_psql` never passed -h or PGPASSWORD, so it defaulted
  to unix-socket peer auth instead of the TCP+password path pgdevkit's own
  find_orphaned_dbs()/workspace_db_names() use -- the listing and dropping
  halves could target different auth paths (or just fail outright). Added
  a --pg-host option (default pgdevkit's own HOST) and PGPASSWORD env.
- `drop_database` interpolated the DB name into `DROP DATABASE IF EXISTS
  "{name}"` unescaped; a name containing a literal `"` (from a live
  pg_database listing, not a validated slug) could break out of the quoted
  identifier. Now escapes embedded `"` per Postgres identifier quoting.
- `is_caution_db` compared db_name's prefix against the raw project name,
  but db names are always pgdevkit's *slugified* name -- a project name
  with uppercase/special characters (e.g. "MDMApp") never matched its own
  DBs ("mdmapp_main"), so the caution flag silently never fired for them.

All four verified against the real sandbox Postgres, not just mocks.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@aersam @claude
Depend on the real pgdevkit 0.7.1 release, drop the temporary git-sou…
54d3a0d 
…rce pin

pgdevkit#35 (workspace_db_names()) merged and 0.7.1 is now published on
PyPI, so the commit-SHA [tool.uv.sources] override this PR added as a
stopgap is no longer needed -- the >=0.7.1 constraint now resolves for real.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@aersam
aersam marked this pull request as ready for review September 14, 2026 18:51
aersam and others added 4 commits September 17, 2026 09:40
@aersam
Merge origin/main into cleanup-worktrees-testdb
c353df9
@aersam @claude
feat(cleanup): add bdt cleanup db, guard all db drops against non-l…
422c97a 
…ocal hosts

`bdt cleanup db` drops the current worktree's own pgdevkit test DB(s) in
place (unlike `cleanup worktrees`, it never touches the worktree itself),
guarded by an interactive confirmation unless --confirm is passed.

Every cleanup command that drops a database (`worktrees`, `orphaned-dbs`,
`db`) now also requires an unconditional, un-overridable interactive
confirmation whenever --pg-host isn't localhost/127.0.0.1/::1 -- --yes/
--confirm alone can no longer drop anything on a remote/shared Postgres
instance.

Also bumps the package version to 0.17.0.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XiiWxLhjWygyfEs7H2WnVr
@aersam @claude
feat(cleanup): add bdt cleanup worktree to remove a single live wor…
b734ad3 
…ktree + its db

Unlike `cleanup worktrees` (bulk, merged-into-main/test only), this tears
down one specific still-live worktree on demand -- run from inside it,
default path '.' -- plus its pgdevkit test DB(s), gated by the same
--confirm/interactive-prompt pattern as `cleanup db`.

Refuses the main checkout, a protected branch, a locked or dirty worktree.
Removal reuses `remove_worktree()`, so a worktree containing submodules is
handled the same way as the bulk command (git's blanket refusal retried
once with --force) -- covered by a real submodule integration test.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XiiWxLhjWygyfEs7H2WnVr
@aersam @claude
docs(readme): document bdt cleanup db/worktree and fix --pg-user fall…
d0c5b78 
…back claim

Documents the two new single-worktree cleanup commands and the non-local
--pg-host confirmation gate, and corrects a stale claim that --pg-user
falls back to \$USER/\$LOGNAME -- it deliberately never did (found by
/code-review).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XiiWxLhjWygyfEs7H2WnVr
@aersam
aersam merged commit 2f28162 into main Sep 17, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@aersam