Cool, thanks for this pull request! I plan to move the tests to mocha anyway, so I should probably do that before merging the domains patch?

No problem!

Re: mocha, it looks like it may have the same issue as expresso -- again, it'll probably have to be fixed upstream. On the upside, at a glance it seems like the hack to make expresso work with domains will also work with mocha.

One other thing to note: MakeCallback (in C++-land) was introduced in 0.8 -- so it won't compile against <0.8 with this patch. I'm looking into wrapping the TRY_CATCH_CALL in a #if statement so we can use the old definition on pre-0.8.X code -- I just need to figure out what I should be comparing against (if node supplies a WITH_DOMAINS directive or if we should supply our own.) The JS code should be unaffected -- the statement.domain = process.domain; line should effectively be a no-op on pre-0.8.X code. I'll update the PR when I figure out a way forward.

Okay, I feel better about this now: it should work on node <0.8.X (falling back to the old TRY_CATCH_CALL), as well as supporting domains where available. If domains aren't available, the domain tests are skipped.

Some questions about the patch:

• why do you set statement.domain to the current domain? I couldn't find a "best practice" or sth about this. If you save the current domain to .domain on Statementobjects, why not also set them on Database objects?
• it seems you set the current domain to each statement object after creation -- can't you put that logic in C++'s Statement construction code? If not -- why now wrap Statement construction in a js function, that would set .domain properly.

@ichernev .domain gets automatically set on descendants of EventEmitter -- as you can see, I'm calling EventEmitter against the Database object here.

To set domain properly through C++ would require looking at the global process object in C++ (which seemed hairy to me). If I follow your alternate suggestion, would it look something like:

function createStatement() {
  var stmt = new Statement
  stmt.domain = process.domain
  return stmt
}

? If so, I'm +1 on that change.

@chrisdickinson thanks for the explanation! So the EventEmitter constructor does that assignment. If so why not use it, inside createStatement than an explicit stmt.domain = process.domain? It'll be clearer that way IMHO.

That makes sense to me too

FYI, node 0.9 has changed so that an uncaughtException is not emitted if there is an active domain

this pull is stale. If anyone is keen on updating this work to support both node v0.8.x and v0.10.x and the new mocha tests then please open a new pull request. closing now without merging.