u1db-discuss team mailing list archive

[Alex Besogonov <alex.besogonov@xxxxxxxxx>]

On Mon, Nov 21, 2011 at 6:08 PM, John Rowland Lenton
<john.lenton@xxxxxxxxxxxxx> wrote:
> On Mon, 21 Nov 2011 17:17:51 -0500, Alex Besogonov <alex.besogonov@xxxxxxxxx> wrote:
> on the one hand, we're not writing a database; we're writing an API
> anybody can implement ontop of existing databases to make them
> syncable. On the other hand, the hard part is the sync protocol, not the
> store itself :)
I've actually done this before (for an unreleated closed source
project). The problem is that you just
devolve it into a simple key/value storage. Lists and recursive
structures in RDBs suck :(

Then there's a problem with conflicts - the sort of schema constraints
to manage them properly
makes it not really worth the bother. So that's why I'm using a dumb
key-value storage (so it's
trivial to substitute it with something else) and use JSON blobs for
documents. That automatically
leads to something like CouchDB views for indexing.

So I pretty much get the design of CouchDB automatically. Except that
I'm planning to make it
suck less at speed.

> Glad to have you in u1db-discuss,
Thanks!