On Jul 12, 2008, at 15:13, David E. Wheeler wrote:
>> 2. It's ridiculously slow; at least a factor of ten slower than doing
>> equivalent tests directly in SQL. This is a very bad thing. Speed
>> of
>> regression tests matters a lot to those of us who run them a dozen
>> times
>> per day --- and I do not wish to discourage any developers who don't
>> work that way from learning better habits ;-)
>
> Hrm. I'm wonder why it's so slow? The test functions don't really do
> a lot. Anyway, I agree that they should perform well.
Just as an FYI, I've just moved all the tests to regular SQL instead
of using pgTAP. The difference in runtime is:
psql -Xd try -f sql/citext.sql 0.03s user 0.02s system 19% cpu 0.253
total
psql -Xd try -f sql/citext.sql 0.03s user 0.02s system 4% cpu 1.298
total
So it's close to a factor of five, though subtract .125 for the time
to load the pgTAP functions. The pgTAP tests *are* doing a lot more
work, but I'm sure that they could be made a lot more efficient,
though of course the TAP functions will always introduce some
overhead. One just needs to decide whether the tradeoffs are worth it.
Best,
David
