(#2685) - fix bogus conflicts, remove 20x speedup

[nolanlawson]

@daleharvey Correct me if I'm wrong, but shouldn't the _conflicts array be empty in these cases? For the remote db it's returning ['1-cf398aa8cceefcf9b8484d02bdaee3fe'] and for the local one, ['2-cf398aa8cceefcf9b8484d02bdaee3fe']. Very bizarre.

[nolanlawson]

Well crap. Removing the 20x include_docs=true speedup fixes the bug. Looks like we can have fast replication, we can have correct replication, but we can't have both. :(

[nolanlawson]

This is one of the saddest commits I've had to write. Also, it's broken in LevelDB, and I don't have time to fix it right now.

[nolanlawson]

Nevermind, it's passing.

[nolanlawson]

Sigh. There's an intermittent:

{ failures: 
   [ { message: 'res.rows[0] is undefined',
       title: 'Replicates modified docs (issue #2636)',
       stack: '@http://127.0.0.1:8000/tests/test.replication.js:1058\nunwrap/<@http://127.0.0.1:8000/dist/pouchdb.js:8070\ndrainQueue@http://127.0.0.1:8000/dist/pouchdb.js:8101\n' } ],
  passed: 485,
  failed: 1,
  lastPassed: 'Replicates deleted docs w/ compaction' }

[nolanlawson]

OK, figured it out. Apparently open_revs="all" doesn't mean what I think it means, and it doesn't necessarily return all the non-leaf revisions (only the leaves). So to avoid the potential URL-too-long error (which is tested btw), I had to implement a batching solution instead.

On the bright side, we have more stable and informative tests now. On the downside, we are back to slowsville for replication.

[nolanlawson]

^ This used to have an if (src.type() === 'http') around it, so we only used this optimization for http adapters. But now that allDocs is fixed in all three adapters, we can use this optimization everywhere.

[nolanlawson]

the default is 30000 anyway. this is a mistake

[nolanlawson]

Travis would be green, but this is blocked by an unrelated issue: #2690. Because when it rains, it pours.

[nolanlawson]

Green. Looking for a +1, then a 3.0.3 release.

@calvinmetcalf I know you have some other commits waiting, but I can look at them after the 3.0.3 release. This is important enough that I think it should get out soon, if you agree.

[calvinmetcalf]

+1

[calvinmetcalf]

Though the changes canceling one that should also be in
On Aug 28, 2014 11:28 PM, "Nolan Lawson" notifications@github.com wrote:

Green. Looking for a +1, then a 3.0.3 release.

@calvinmetcalf https://github.com/calvinmetcalf I know you have some
other commits waiting, but I can look at them after the 3.0.3 release. This
is important enough that I think it should get out soon, if you agree.

—
Reply to this email directly or view it on GitHub
#2685 (comment).

[nolanlawson]

I'll look at it, but I don't think it should block 3.0.3. We can always do a 3.0.4.

[nolanlawson]

ded1b43