* Use new fetcher implementation for post/comment
* rewrite person fetch to use new fetcher
* rewrite community to use new fetcher
* rename new_fetcher to dereference_object_id
* make ObjectId a newtype
* handle deletion in new fetcher
* rewrite apub object search to be generic
* move upsert() method out of ApubObject trait
* simplify ObjectId::new (and fix clippy)
* Use URL type in most outstanding struct fields
This fixes all known remaining cases where url fields are stored as
plain strings, with the exception of form fields where empty strings
are used as sentinels (see `diesel_option_overwrite_to_url`).
Tested for regressions in the federated docker setup attempting to
exercise all changed fields, including through apub federation.
Fixes#1385
* Add migration to fix blank-string post.url values to be null
This also then fixes#602
* Address review feedback
- Fixed some unwraps and err message formatting
- Bumped the `url` library to 2.2.1 to fix a bug with serde error
messages
- Add unit tests for the two diesel option override functions
- Fix migration teardown by adding a no-op
* Rename lemmy_db_queries::Url to lemmy_db_queries::DbUrl
* fix compile error
* box PostOrComment variants
Since DB tests execute diesel migrations automatically, concurrent
execution causes flaky failures from simultaneous migrations. This can
be worked around using `cargo test --workspace -- --test-threads=1`,
which is what the CI config does, but this is not intuitive for
newcomer developers and unnecessarily slows down the test suite for
the majority of tests which are safe to run concurrently. This fixes
this issue by integrating with the small test crate `serial_test` and
using it to explicitly mark DB tests to run sequentially while
allowing all other tests to run in parallel.
Additionally, this greatly improves the speed of `cargo test` by
disabling doc-tests in all crates, since these are aren't currently
used and cargo's doc-test pass, even when no doc-tests exist, has
significant overhead. On my machine, this change significantly
improves test suite times by about 85%, making it much more practical
to develop with tools like `cargo watch` auto-running tests.
