Here you go.

[JulianBirch]

I thought I should document what I've done and why.

In client integration tests, helper.client now automatically adds a handler to client.on 'error'. This means that connection problems and other unexpected errors get printed out. It also means that the tests don't crash out the moment they hit an incorrect SQL syntax. Next, I did the same thing to client.query. This enabled me to actually see what errors I was getting, which revealed that (no real surprise here) "Relation person did not exist".

It's then that I found the comment "Make sure you run the script create-test-tables". :) So, reading the documentation correctly would have saved me a lot of hassle.

After doing that, I finally had 25 != 26 like we discussed last night. I've left these changes to the tests in because I think it's easier for someone to debug when they start out this way. Oh yes, it also prints out the test names. Your mileage may vary :)

I changed the query to select name from person order by name, simply because I had some (SQL Server) code recently that assumed the order by, which broke after a year in production. Wasn't very funny...

So finally I got around to reproducing my issue, which it turns out was harder than I thought. The basic problem was that it broke eyes, which isn't very testable. Instead, I went for the fact that iterating through the columns and calling "length" doesn't agree. Of course, if you use Object instead of Array, there isn't a length property anyway, but I think it might also be possible to get confused if there actually was a column called 'length'.

[brianc]

Thanks mucho. Soon after we discussed this yesterday I was building out postgres-session and noticed exactly what you were talking about. It all made sense, and thanks again for the patches.