Re: varchar as primary key

Jeff Davis <pgsql(at)j-davis(dot)com> writes:
> $ ./cmp
> locale set to: en_US.UTF-8
> strcmp time elapsed: 2034183 us
> strcoll time elapsed: 2019880 us

It's hardly credible that you could do either strcmp or strcoll in 2 nsec
on any run-of-the-mill hardware. What I think is happening is that the
compiler is aware that these are side-effect-free functions and is
removing the calls entirely, or at least moving them out of the loops;
these times would be credible for loops consisting only of an increment,
test, and branch.

Integer overflow in your elapsed-time calculation is probably a risk
as well --- do the reports add up to something like the actual elapsed
time?

I tried a modified form of your program (attached) on an FC6 machine
and found that at any optimization level above -O0, that compiler
optimizes the strcmp() case into oblivion, even with code added as below
to try to make it look like a real operation. The strcoll() call without
any following test, as you had, draws a warning about "statement with
no effect" which is pretty suspicious too. With the modified program
I get

$ gcc -O1 -Wall cmptest.c
$ time ./a.out
locale set to: en_US.UTF-8
strcmp time elapsed: 0 us
strcoll time elapsed: 67756363 us

real 1m7.758s
user 1m7.746s
sys 0m0.006s

$ gcc -O0 -Wall cmptest.c
$ time ./a.out
locale set to: en_US.UTF-8
strcmp time elapsed: 4825504 us
strcoll time elapsed: 68864890 us

real 1m13.692s
user 1m13.676s
sys 0m0.010s

So as best I can tell, strcoll() is pretty dang expensive on Linux too.

regards, tom lane