As far as I can tell idb doesnt report the offset of the key when doing a keyrange query

will look into it later

IndexedDB doesnt doesnt allow for this api, this means replications will always create their target etc

I get this strange error message when there's no callback provided for the pouch.open function:

TypeError: Cannot read property 'http' of undefined

I know that calling this function without callback doesn't make sense, but the error message should indicate that this is the problem.

The current failing tests are due to deleteDatabase receiving an onblocked event and not completing, I dont know what is causing the onblocked event as there should be no open databases at the point it runs, and there isnt much documentation on how to unblock it

There is no API to list the currently defined IndexedDB's

This is just an issue to keep track of when browsers implement it, I am fairly sure they will have to at some point

i don't know if this was ready for testing or not but i am using all files from src folder with the following problem

queries that are set like http://localhost:3000/couch/testdb/....
are actually being sent like
http://localhost:3000/testdbcouch/ ......

Hi Guys,

I Start a portable version of futon to pouchdb. 
how think is a good idea?

cheers

Bug in the idb changes implementation

"a".localeCompare("A");
32

in chrome, whereas in firefox

"a".localeCompare('A');
-1

http://code.google.com/p/v8/issues/detail?can=2&start=0&num=100&q=&colspec=ID%20Type%20Status%20Priority%20Owner%20Summary%20HW%20OS%20Area%20Stars&groupby=&sort=&id=459

I implemented _junk as I was getting bugs in the firefox test suite relating to not having a unique string key, someone mentioned that may be fixed

No error is returned and you get a db object in the callback but when you try to write to it you get a "NOT_FOUND_ERR: DOM IDBDatabase Exception 3" - we should return a sensible error about valid database names

At one point, I believe that you could pass {continuous: true} as an options object to db.replicate.to and db.replicate.from. Currently it looks like the methods still take an opts argument, but now it is not used.

Are there plans to deprecate this functionality, or does it need to be reimplemented? Continuous replication seems to be a preferable approach to using a setInterval to regularly run replication.

When not doing continuous replication it would be extremely useful to know when it has completed.

Currently most PouchDB api calls are of the format

pouch.function(arg, [options], [callback])

This makes for a nice api however, I prefer a plain success callback than attaching one to options, however it does mean that pretty much every function needs to start with

if (opts instanceof Function) {
  callback = opts;
  opts = {};
}

We can provide a fixargs wrapper, so

pouch.get = fixArgs(function(doc, opts, callback ...

but that seems pretty ugly as well, any ideas?

Another issue is how we provide error callbacks, we currently use the node.js idiom of

pouch.get('id', function(err, doc) { 
} 

Which is nice enough, I am happy sticking with it, especially when there are people looking to implement a pouchdb server in node, but open for suggestions

The last issue is out of band data, I have been attaching things like status codes to the returned data however that isnt going to work in a lot of situations, so this will probably be changed to have an extra compatibility object in the callback such that:

pouch.get('id', function(err, doc, meta) {
  console.log(meta.status);
  console.log(meta.header('Content-Type')); 
}

We will also be moving towards a object model closer to Lawnchair of

new Pouch('name', fun....

which lets us have instance options for individual instances and handles the adapter model better, we will also use protocol based names so you can

Pouch.replicate('idb://test', 'http://127.0.0.1/test', ...

The mergeTrees code added by tilgovi has a number if syntax errors that completely break pouchdb causing it to fail all unit tests. I have commented mergeTrees out and now tests 1 and 2 pass, but it hangs on test 3 I think. txn.objectStore('ids').openCursor() could be the culprit. I don't know enough about IndexDB, nor the PouchDB code to be sure.

Can someone give me an idea of the status of the PouchDB project.

I think it would be brilliant to have CouchDB capabilities running in the browser (JS) with IndexedDB and Couch's Replication. I've yet to find any solution for sometimes connected apps with replication anywhere near as good as CouchDB. Nothing I'm aware of comes close.

Congrats to the folks working on PouchDB. I really would love to see this completed.

Neville

need to be able to replicate/use CouchDB instances that are hosted under a proxy like mydomain.com/couch

So I finally took a pass at writing a function to merge rev trees. I can push it to a branch for testing, but I want to unit test it outside the browser for reals.

I propose to break out the library stuff at the top of pouch.js into separate files. We should be able to compile the aggregate out to one file, optionally minified for production.

Eventually we will want this if we move forward with the LevelDB implementation so we can share code.

I was playing with browserify this week, but I don't know if it can export the result or just serve it from a running node process, but I'd love to use the require() module style.

Each test that has 2 database that interact with each other should automatically tun against each combination of the adapters

idb <-> http, idb <-> idb, http <-> idb, http <-> http

etc

This is the easiest way to test the completeness of each adapter without massively redundant tests

How much of this is usable for development/playing around (obviouly not production yet)? Known issues?

on chrome 13.0.782.107

there isnt a test to test it either

you can see this fail by running test.html a few times and watching the number of docs go up by 100 each time you reload

hi guys

i'm trying using chrominium for use the dev tool to inspect indexeddb but nothing :(
insted that show some warnings and an error

Numeric transaction modes are deprecated in IDBDatabase.transaction. Use "readonly" or "readwrite".
Numeric direction values are deprecated in IDBObjectStore.openCursor. Use"next", "nextunique", "prev", or "prevunique".
Uncaught Error: NOT_ALLOWED_ERR: DOM IDBDatabase Exception 6 pouch.idb.js:394

what i do wrong?

So we dont need to wait for setTimeouts in the tests (also generally useful)

I have been using

{error: 'error', reason: 'reason', status: 400}

to report errors, I find it handy to copy the http status codes, need to ensure bulkdocs report these for things like invalid revisions

Hi guys,

When i try make the master branch, and include the pouch.alpha.js on my html file i had an error on chrome:

Uncaught TypeError: Cannot set property 'merge' of undefined
resultspouch.alpha.js:602
(anonymous function)pouch.alpha.js:613
(anonymous function)pouch.alpha.js:1806

some one knows why this?

the settimeout in the auth_replication test is a workaround

It would be possible to implement cross-domain CouchDB access by using a simple iframe trick.

The first step would be placing a proxy.html on the target CouchDB as a document. On the client website, a PouchDB instance would create a hidden iframe with this proxy.html in it.

proxy.html can make XMLHttpRequests to the target CouchDB instance (since it is on the same domain), and PouchDB can communicate with proxy.html using the postMessage function. The message format could be JSON-RPC.

From the user's perspective would look like new Pouch('iframe://couch.com/db/proxy', function() {}).

I know that this is a hacky workaround, but the CORS ticket on CouchDB's bug tracker is 3 years old now, and I don't expect it to be resolved soon.

couch.js is missing ??

I think it may be a bug related to objectstores without a keypath that autoIncrement

I would like to mirror the api for communicating with a remote couchdb server with the same api

pouch.open('http://mycouch.com/test', function ....

This allows a fallback for when idb support isnt around, without having to change any user code

Not planning to implement it until the idb version is somewhat complete, but just making a note

In order to avoid writing a replicator (and any other higher level modules) twice there should be some API exposed which is common between both the browser and the node environment.

I propose the following hybrid of DOM Events and Node.JS style APIs:

• The last argument is always either an Events-style options Object or callback Function.
• If there are no options passed the argument may be left out and replaced with a callback Function.
• If both an options Object and a callback Function are specified (in that order), the callback should supersede any event handlers in the options and the callback should be called according to Node.JS idioms.
• All events and callbacks receive JavaScript objects which are the native equivalent of the JSON for the matching Apache CouchDB request.

I think this can make everyone happy. It's trivial to write a function that consumes options and a callback and turns the possible combinations into a single options argument (properly wrapping the callback in event handlers) so that external code can write to either API and internal code can do what makes sense for the environment.

if tests crash, leaves us no way to clean up the database, firefox has some fairly small limit to the number of databases to a host, I havent hit chromes limit yet

I am trying to use it, and i get a lot of "not implemented" and "implement X
ASAP" messages and it only managed to create records, tests also fail.

is there any prevision for pouchDB to be usable?

by the way great idea!.

Happening occasionally in chrome

test.bulk_docs.js:28 Uncaught TypeError: Cannot read property 'length' of undefined

We could do with lots of adapters, but localstorage is probably the most important

"If you need to fetch a document and all its attachments in one request, add an ?attachments=true URL parameter to the GET request. The resulting JSON will include the base64-encoded contents in the "data" property of each attachment, just as in an inline upload." [http://wiki.apache.org/couchdb/HTTP_Document_API#Getting_Attachments_With_a_Document]

Support for that inline attachments api would be great.

And if so, could you put in a LICENSE file?

Only the first few tests have been converted, so the rest don't even show up. This is because qunit's start() function is never called for those.

When running the test for the first time on a brand new install, the "Remove a Pouch" test fails.

When running the tests again, it passes fine.

it is quite rare to need to replicate the whole database for a user, filtered replications IMHO is a must have

Currently Couch doesnt support CORS, so either a cors proxy is needed, or you need to serve the webpage from couchdb (or anywhere from the same domain)