Account incineration could leave orphaned records due to missing cascade
declarations and async jobs failing when the account was already gone.
Cascade fixes:
- Add Account::Searchable concern to clean up Search::Query (delete_all)
and Search::Record (destroy_all, respects SQLite FTS dependent: :destroy)
- Add before_destroy in Account::Storage to delete storage entries
- Suppress storage entry recording during incineration so attachment
purge callbacks don't create entries or enqueue materialize jobs
- Guard Access#clean_inaccessible_data_later with unless user.destroyed?
to avoid enqueuing pointless jobs during user cascade
Job tenancy:
- Extract AccountTenanted concern from the global ActiveJob initializer
into ApplicationJob (include) with targeted prepends for
ActionMailer::MailDeliveryJob and Turbo broadcast jobs
- Defer account resolution from deserialize to perform so that missing
accounts raise DeserializationError inside the execution path where
discard_on can handle it
Tests:
- Comprehensive incineration test covering 35 model types with
before/after assertions and full enqueued job processing
- Mailer deliver_later test verifying account context survives job
serialization for multi-account users
- Turbo broadcast test verifying account-scoped URLs in rendered partials
record.to_json serializes through model accessors. Column::Colored
overrides the color accessor to return a Color struct, which serializes
as {"name":"Blue","value":"var(--color-card-default)"} instead of the
raw CSS string.
Two fixes:
- Export raw DB values via record.attributes.slice(*attributes).to_json
- Color.for_value parses legacy JSON format so columns imported from
old exports self-heal when read
* Exclude export/import blobs and attachments from account exports
Export and import file blobs were being assigned the account's
account_id (via the Current.account default), causing subsequent
exports to include previous export/import ZIP files in the data.
This roughly doubled the export size each time and made imports
fail with an IntegrityError on unrecognized "Account::Export"
record type.
Filter out blobs attached to Export/Import records in
BlobRecordSet, FileRecordSet, and a new AttachmentRecordSet.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Use STI base type for export blob exclusion filter
Rails polymorphic_name returns base_class.name for STI models,
so ActiveStorage stores record_type as "Export" rather than the
concrete subclass names. Filter on the base type accordingly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Fix export filtering for mixed-use Active Storage blobs
Exclude only internal-only blobs for account export metadata/file sets, scope attachment subqueries by account, and add regression coverage for blobs shared between user records and export attachments.
* Simplify export blob filtering
Exports always create fresh blobs, so a blob can never be legitimately
shared between an Export/Import and a real account record. Replace the
internal_only/external split with a single excluded_blob_ids subquery
and drop the contrived shared-blob test.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Fix multi-term SQLite FTS5 search causing a 500 error
When filtering cards by more than one term, `matching` was called
once per term via an association join. Rails did not deduplicate
association joins, resulting in two JOINs to `search_records_fts`
in a single query. SQLite FTS5 then raised "ambiguous column name"
when evaluating the MATCH condition.
Fix by using a string join instead, which Rails deduplicates so
only one JOIN to `search_records_fts` is generated regardless of
how many terms are searched.
Fixes: https://github.com/basecamp/fizzy/discussions/2354
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test: verify multi-term search requires ALL terms to match
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Mike Dalessio <mike@37signals.com>
Add IMPORTABLE_MODEL_NAMES to RecordSet and verify polymorphic type
columns during import check against this allowlist before calling
constantize. This prevents arbitrary class instantiation from
untrusted import ZIP data.
BlobRecordSet#import_batch preserved blob keys from exported ZIP files
verbatim. A crafted ZIP with a traversal key like "../../config/deploy.yml"
would create a blob whose key resolves to an arbitrary filesystem path
when served by ActiveStorage's DiskService, allowing an attacker to read
local files.
Fix: generate fresh blob keys on import, discarding whatever key was in
the ZIP. This is safe because nothing else in the import pipeline
references blobs by key — attachments use blob_id, ActionText uses GIDs
based on record IDs, and only FileRecordSet used the old key for file
data lookup.
Update FileRecordSet to build an old_key→blob_id mapping from the blob
JSON metadata in the ZIP, then look up blobs by ID instead of by key.
Both check_record and import_batch are now fail-closed: they raise
IntegrityError for unmapped storage files, missing blobs, and duplicate
keys in the export.
Backfill test coverage for BlobRecordSet and FileRecordSet import, and
add a round-trip test verifying blob data survives export/import with
regenerated keys.
* Validate and normalize auto-postpone period to days
- Add Entropy::AUTO_POSTPONE_PERIODS and validate auto_postpone_period against the allowed set
- Introduce auto_postpone_period_in_days for forms and entropy update endpoints
- Fall back to index 0 in knob partial when current value isn't in options
- Remove entropy_auto_close_options helper in favor of model constant
- Update tests to use allowed period values
* Use auto_postpone_period_in_days for JSON entropy updates
The JSON API was accepting auto_postpone_period (seconds) which
bypassed the days normalization and validation. Switch to
auto_postpone_period_in_days consistently and add explicit
wrap_parameters so flat JSON params are wrapped correctly for
the virtual attribute.
* Default to account or 30-day fallback when knob value is invalid
* Address PR review feedback for entropy validation
- Fall back to first knob option (index 0) when persisted value isn't in
the allowed set, preventing nil index errors for legacy values
- Use integer division (1.day.to_i) for consistent day calculations
* Default to account entropy period instead of first option for knob fallback
* Test that default auto-postpone period is in the allowed periods
* Return 422 instead of 500 for invalid entropy auto-postpone values
Rescue ActiveRecord::RecordInvalid in entropy controllers so invalid
auto_postpone_period_in_days values return 422 Unprocessable Entity
instead of raising a 500.
* Fix NoMethodError when board entropy falls back to account default
Use container.account.entropy.auto_postpone_period_in_days instead of
container.account.auto_postpone_period_in_days since Account doesn't
delegate that method.
* Test board entropy fallback to account default for invalid periods
Publishing or unpublishing a board didn't bust the `json.cache! board`
fragment cache because `Board::Publication` didn't touch its parent
board. The board's `updated_at` stayed stale, so subsequent GET requests
continued serving cached JSON without `public_url`.
Clear assignee's existing notifications in setup since Notifier now
uses create_or_find_by instead of create, and reload the association
to avoid caching after destroy_all.
Update fixture references after rebase: use `logo_assignment_kevin`
as base notification and `logo_mentioned_david` for mention tests.
Update source to `logo_published` in tests that need generic card events.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add comprehensive integration test covering card assignment, comments,
mentions, email bundling, and edge cases (system user, inactive user)
- Inline push notification test helpers into the only test that uses them
- Remove separate PushNotificationTestHelper module
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Each target now implements process directly with its own logic,
rather than using processable?/perform_push hooks. The pushable?
check is done once in Notification#push before iterating targets.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
"Push to target" reads naturally - we push the notification to the
target. "Target processes" also makes sense - the target receives
and handles the notification in its own way.
- Add class method PushTarget.process(notification) that instantiates
and calls the instance method
- Rename instance method from push to process
- Add private push_to helper in Pushable for readable iteration
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Replace separate WebPushJob and NativePushJob with a single PushJob
that calls notification.push, which iterates over registered targets.
Each target handles its own delivery - Web pushes synchronously via
the pool, Native enqueues device-level jobs via deliver_later_to.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
PushTarget#push reads more naturally than Push#push. The push target
is the thing that pushes notifications to a specific destination.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Replace NotificationPusher with a cleaner architecture:
- Add Notification::Pushable concern with push target registry
- Add Notification::Push base class with template methods
- Add Notification::Push::Web for web push (OSS)
- Add Notification::Push::Native for native push (SaaS)
- Add Notification::WebPushJob and Notification::NativePushJob
Key design:
- Registry pattern: Notification.register_push_target(:web)
- Template method: push calls should_push? then perform_push
- Subclasses override should_push? (with super) and perform_push
- Each target handles its own job enqueueing
Also:
- Add Notification#pushable? for checking push eligibility
- Add Notification#identity delegation to user
- Reorganize tests to match new class structure
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Tidy up saas engine a bit more
Event notifiers used the `mentionees` DB association to exclude mentioned
users from comment/card notifications. Since mentions are created async
via Mention::CreateJob, a race condition meant the mentionee list could
be empty when the event notification job ran first, causing the user to
receive both a comment and a mention push notification.
Use `scan_mentionees` instead, which scans the rich text body directly
for mentioned users without depending on Mention records existing yet.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Rails only applies the last callback when `after_create_commit` and
`after_update_commit` reference the same method name [1]:
> However, if you use the `after_create_commit` and the
`after_update_commit` callback with the same method name, it will only
allow the last callback defined to take effect, as they both internally
alias to `after_commit` which overrides previously defined callbacks
with the same method name.
- Push notifications were never sent when a notification was first
created — only when the source was updated
- Replaced the two callbacks with a single `after_save_commit`, which
fires on both create and update, with the
`source_id_previously_changed?` guard (true for both new records and
source changes)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
[1] https://guides.rubyonrails.org/active_record_callbacks.html#aliases-for-after-commit
Cover the stemmer tokenizing hyphenated input into separate words and
end-to-end search finding cards with hyphenated titles.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
ZipKit::FileReader raises MissingEOCD for truncated, corrupted, or
non-zip files. This exception is a direct StandardError subclass, not
a subclass of InvalidStructure or ReadError, so it escaped as an
unhandled job failure instead of being caught and surfaced to the user.
Broaden the rescue in ZipFile::Reader#initialize to catch ReadError
(parent of InvalidStructure), MissingEOCD, and UnsupportedFeature.
When concurrent NotifyRecipientsJobs process an Event and a Mention for
the same user+card, Rails' dirty tracking can skip writing source_type
in the UPDATE if it hasn't changed from the stale in-memory value,
leaving source_type and source_id mismatched (e.g. source_type='Event'
with a Mention's source_id, resulting in a nil source).
Force source_type to always be included in the UPDATE via
source_type_will_change! to prevent this.
Here's a sample timeline of this race condition happening in the real
world (with simplified IDs):
Two `NotifyRecipientsJob` involving notification `03fklpu`, same card,
same user, same comment — enqueued within 50ms of each other:
1. Both jobs load notification `03fklpu` — it has source_type='Mention'
(from a previous job)
2. `EventNotifier` writes at 21:22:36.051: ```sql SET
source_type='Event', source_id=<event_id>, unread_count=1 ```
3. `source_type` included because it changed ('Mention' → 'Event')
4. `MentionNotifier` writes at 21:22:36.057 (~6ms later): ```sql SET
source_id=<mention_id>, unread_count=1 ```
5. No `source_type`! It was `'Mention'` when loaded and `'Mention'` is
what it's setting → not dirty → skipped
6. Final DB state: `source_type='Event'` (from step 2, untouched),
`source_id=<mention_id>` (from step 3)
`Notification.source` now does `Event.find(<mention_id>) → nil.`
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The convert_gids_to_sgids method resolved GIDs globally and minted
valid SGIDs without verifying the record belongs to the importing
account. This adds the same account_id check that the export path
already has, plus RecordNotFound handling for invalid GIDs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
We had client-side notification stacking in the tray since launch, but now we want to stack notifications in the notifications page, in API responses and in email bundles.
Rob Zolkos