Thanks, Craig. That makes sense. Yes, it's quite a number of work to do. :-) I'll take a look at the comments and code and try to understand it.
T On Thu, Nov 8, 2012 at 6:28 PM, Craig Ringer <cr...@2ndquadrant.com> wrote: > On 11/09/2012 09:37 AM, Tianyin Xu wrote: > > Thanks, Craig, > > > > Yes, I know "context diff". What I don't know is whether + or - some > > rows is a big problem, let's say correctness problem. I didn't write > > the test cases so I don't know what these test cases are exactly doing. > The SQL to the test cases is right there in the regress directory, > usually with comments explaining what each test is for. > > > If you tell me the failure of these test cases are severe and not > > acceptable, I'm fine with it. It means these configurations are not > > allowed. > It depends on the test. Some are testing whether the optimizer behaves > as expected, choosing a particular plan. Some tests assume rows are > being returned in a particular order, so if you change settings you can > see at a glance that while the test fails the output is actually OK. > Others are correctness tests and must produce exactly the the expected > results. > > This is generally clear from the test query and comments. > > which assigned a big number to the cpu_index_tuple_cost, affecting the > > query planner. > The planner tests are written for a particular configuration, so if you > change the configuration they won't produce the expected results. That's > OK; just make sure other tests are fine. > > It'd be nice to split the tests up into clearer groups - "will fail if > planner settings are changed; WARNING", "will fail only if incorrect > result is returned; FATAL" etc. Right now, AFAIK that hasn't been done. > > -- > Craig Ringer > -- Tianyin XU, http://cseweb.ucsd.edu/~tixu/
