Synchronisation status is based on document signature (sha) comparison.
Signature are stored each time a document is successfully synchronised.

Conflict are detected but reported as an error for now, as no solve process is implemented.

This (slow) method will be used to stabilised a storage (launch synchronisation process for example).

Use a new optional parameter: jIO.getAttachment(id, name, {format: blob})

Supported formats are: blob, text, json, array_buffer, data_url

Do not return the list of attachments in jIO.get.
Use a specific "allAttachments" method to fetch this information.

jIO.get now really returns the document as saved by jIO.put/post.

Modify jIO API to keep user data not modified.
    jIO.get({"_id": "foo"} -> jIO.get("foo")
    jIO.remove({"_id": "foo"} -> jIO.remove("foo")
    jIO.post({"title": "bar"}) -> jIO.post({"title": "bar"})
    jIO.put({"_id": "foo", "title": "bar"}) -> jIO.put("foo", {"title": "bar"})
    jIO.getAttachment({"_id": "foo", "_attachment": "enclosure"} -> jIO.get("foo", "enclosure")
    jIO.removeAttachment({"_id": "foo", "_attachment": "enclosure"} -> jIO.remove("foo", "enclosure")
    jIO.putAttachment({"_id": "foo", "_attachment": "enclosure", "_blob": blob} -> jIO.remove("foo", "enclosure", blob)

jIO.get does not return an _id attribute anymore.

Do not store data as encoded string but as data URL.
Set a default mime type for text attachment if not provided.

Allow to choose which action_reference will be used as default view.

openCursor only return one entry per index value by default.
Use IDBKeyRange to get the expected behaviour.

Simplify storage and tests code.

Change the database structure so that the "attachment" store contains one entry per attachment.

Add tests for memorystorage.
Drop copy/paste code in jio.post. Should be added in a standalone storage if needed.

All ERP5 callable can be represented as attachment.
Editing (FormData) is done with putAttachment.

It is not possible to directly create/modify.