[tx] Handle map assertion shorthand

[ncalexan]

Right now, we don't handle assertions in the map shorthand, like

[{:db/id IDENT :db/attr1 VALUE1 :db/attr2 VALUE2 ...}]

That is equivalent to

[[:db/add IDENT :db/attr1 VALUE1]]
[[:db/add IDENT :db/attr2 VALUE2]]

This ticket tracks parsing such assertions into Entity instances, and transacting such Entity instances.

Datomic allows to drop the :db/id IDENT in nested entities (marked with :db/isComponent?) like

[{:db/id IDENT :db/ref-attr {:db/attr1 VALUE1 ...} ...}]

A tempid in the same partition as IDENT will be generated for the inner entity. And it's possible that even the outer :db/id can be dropped, automatically selecting :db.part/user, although I haven't tested this.

In the Clojure implementation, I expanded from a map into a vector before transacting. I'd very much like to avoid doing that, both to save allocations, and to potentially save tempid lookups; but it's not critical that we're optimal initially.

[ncalexan]

This relies on #169, but the tempid parsing should land in #283 so we should just do the map parsing as part of this ticket. This is similar to #284; do that one first, since it's simpler.

In the Clojure implementation, I expanded from a map into a vector before transacting. I'd very much like to avoid doing that, both to save allocations, and to potentially save tempid lookups; but it's not critical that we're optimal initially.

This is no longer relevant -- the approach taken by #283 is to share Rc handles to expensive-to-clone things (like lookup refs and tempid strings), so it's reasonable to just expand map entities into many assertions and be done with it. However, this is still a multi-part process. You'll need to:

1. Add a Map case to

   mentat/tx/src/entities.rs

   Line 57 in 6831925

   },

   . Map notation is recursive, so the transaction processor will have to handle arbitrary nested edn::Value::Map instances, so there's no advantage to specializing the representation here -- just make it Map { map: edn::Value }.
2. Parse maps around

   mentat/tx-parser/src/lib.rs

   Line 140 in 6831925

   let mut p = Tx::<I>::add()

   .
3. Expand maps in the transactor into multiple Term instances around

   mentat/db/src/tx.rs

   Line 133 in 6831925

   Entity::AddOrRetract { op, e, a, v } => {

   .
   This is conceptually a flat_map, but Rust makes it hard to flat_map when different branches produce iterators with different type signatures, so instead make the code a loop and push into a Vec of terms. That'll let you recursively handle destructuring a map and accumulating the resulting assertions, as described in [tx] Handle map assertion shorthand #180 (comment).
   Be aware that :db/id is optional, and if not specified should be a fresh tempid allocated by the transactor. See Accept :db/id in nested maps. #206 for a Clojure implementation bug and test for this issue.
4. Add test fixtures asserting the success and some of the failure cases around https://github.com/ncalexan/mentat/tree/6831925af0eafe9e647a462364b6188748f1e840/tx/fixtures. You might learn from the tests at https://github.com/mozilla/mentat/blob/master/test/datomish/db_test.cljc#L482.

For bonus points: Datomic allows nested anonymous (no :db/id) maps only when the attribute is :db/isComponent true or the inner map might upsert (i.e., has an [a v] pair where a is :db/unique :db.unique/identity). See http://docs.datomic.com/transactions.html, the section "Nested maps in transactions". Enforcing these rules could be a Post: commit, or a follow-up issue.

[ncalexan]

@joe this is a good second ticket in the transactor. It's conceptually similar to #284, but more involved.