Hello Alvaro,
Here is a v6 with most of your suggestions applied.
> On top of evaluateExpr() we need a comment (generally I think pgbench
> could do with more comments; not saying your patch should add them, just
> expressing an opinion.) Also, intuitively I would say that the return
> values of that function should be reversed: return true if things are
> good.
Comment & inverted return value done.
> I wonder about LOCATE and LOCATION. Can we do away with the latter, and
> keep only LOCATE perhaps with a better name such as PRINT_ERROR_AT or
> similar? I would just expand an ad-hoc fprintf in the single place
> where the other macro is used.
I've used just one PRINT_ERROR_AT() macro consistently.
> Are we okay with only integer operands? Is this something we would
> expand in the future? Is the gaussian/exp random stuff going to work
> with integer operands, if we want to change it to use function syntax,
> as expressed elsewhere?
Nothing for now, I feel it is for a later round.
> [other mail] bring ERROR() macro back
I also prefer the code with it, but the cost-benefit of a pre-C99
compatible implementation seems quite low, and it does imply less (style)
changes with the previous situation as it is.
--
Fabien.