Skip to main content

atomic_tree

Function atomic_tree 

Source
pub async fn atomic_tree<T, F>(pool: &Pool, f: F) -> Result<T, TreeError>
where T: Send + 'static, F: Send + 'static + for<'tx> FnOnce(&'tx mut PoolTx<'_>) -> Pin<Box<dyn Future<Output = Result<T, TreeError>> + Send + 'tx>>,
Expand description

Run f inside a framework rustango::sql::atomic scope while letting it return TreeError — the bridge that lets the on_commit cache-invalidation pattern reach the tree-write paths.

rustango::sql::atomic fixes its closure’s error to ExecError, but the tree helpers (cascade_url_path_to_descendants, move_to, …) speak TreeError, and there’s no TreeError -> ExecError conversion (only the reverse, via TreeError::Db). This adapter closes that gap: a TreeError from f is stashed and a sentinel ExecError is returned to drive atomic’s rollback, then the real error is recovered afterwards. Recovery keys off the stash, not the sentinel value, so a genuine ExecError raised by BEGIN / COMMIT itself is never misread as a domain error.

on_commit callbacks queued by f (e.g. via crate::cache_invalidate::invalidate_urls_on_commit) fire only on the commit path — a TreeError rolls the transaction back and drops them, exactly like a native atomic block.

The cleaner long-term fix is a generic atomic in the framework (closure returning Result<T, E: From<ExecError>>); until that lands, this adapter keeps the rollout unblocked.

§Errors

The first TreeError returned by f, or a BEGIN / COMMIT driver error surfaced as TreeError::Db.