I'm using datomic on the server side, with multiple reagent atoms on the client, and now looking at trying datascript on the client.
Currently, I'm passing across a nested structure via an initial api load, which contains the result of a datomic pull query. It's pretty concise, and works fine.
However, now looking to explore the potential benefits of datascript. Selling point there is it seems to allow to retain normalisation right down to the attribute level. However, I come across an initial hurdle. Datascript isn't, as I'd imagined (perhaps, hoped...), a way to just subset your datomic db and replicate it on the client. Problem is, datomic's entity ids cannot be shared to datascript, specifically - when you transact! entities into datascript, a new eid (datascript's) is issued for each entity.
I haven't worked through all of the consequences yet but it appears it would be necessary to store :datomic-id in datascript, in addition to datascript's own newly issued :db/id, and ref types are going to use datascript's id, not datomics. This potentially complicates synchronisation back to datomic, feels like it could create a lot of potential gotchas, and isn't as isomorphic as I'd hoped. But still working on it. Can anyone share experience here? Maybe there's a solution...
Update:
wonder if a solution is to ban use of datomic's :db/id on the client, enforcing this by filtering them out of initial load; not passing them to client at all. Then any client -> server communication would have to use the (server-generated) slugs instead, which are passed in the initial load.
So, all entities would have different ids on the client, but we ban passage of server id to client, so client id if accidentally passed to server should probably say eid not found. There are likely more issues with this, haven't worked it right through yet.
You also have to think in entities, not datoms, when passing to & inserting to client, so as to create the correct refs there (or perhaps could insert a tree, if wrangle that).
So I've discovered that the datomic/datascript partnership certainly isn't just a case of 'serialise a piece of your database' - that might work if using datascript on the server, which is not the use case here at all (db persistence being required).
If I remember correctly, Datomic uses all 64 bits for entity ids, but in JavaScript (and by extension in DataScript) there’re only 53-bit integers max. So some sort of translation layer is necessary either way, no way around it.
P.S. you can totally specify :db/id to whatever you want in DataScript and it’ll use that instead of generating its own. Just make sure to fit in 53 bits.