Just checked in SVN R501. The details of the changes are given below.
John
On 3/23/12 2:47 PM, Dominique Prunier wrote:
> Hey John,
>
> I'm testing r500 right now and i have a few issues:
>
> 1. In category::stringSearch:
> else if (ind == 1 && dic.size() == 1) { // special case with one value
> getNullMask(hits); // all valid entries
> }
> We can't do this shortcut, this is not nulls that make an error but also
> empty strings (which is what i have in my test case). When i remove these 3
> lines, it works again
Your are absolutely right. In fact both special cases need to be
removed. This has been done now.
>
> 2. Most of my select distinct (using bundles) that returns empty strings (key
> 0) on categories were broken. I think this is related to the fact that the
> bundle used to write "<NULL>" for key 0 instead of "". Unfortunately, in the
> C API (fastbit_result_set_getString), the empty string is the mark of an
> error, so there is no way to distinguish them anymore. In this particular
> case, this is probably the CAPI which is wrong, maybe we should return a null
> pointer on error instead of an empty string. If i restore the else in
> colUInts::write (else out << "<NULL>";), my queries works again
Printing out an empty string as "<NULL>" was a total random action on
my part. One rationale thing is to do would be give back exactly what
was on input, i.e., "". I am giving this a try now. Please let me
know how it goes for you.
>
> 3. This is not really an issue for me, but it seems also that the quoting of
> string in colUInts::write got removed. I'm perfectly fine with that, i used
> to try to strip quotes. It actually happened at r468 but my code was checking
> if the string was quoted or not before trying to remove them, so it didn't
> break anything.
>
The problem is somewhat unclear to me. The C API is using a string
object to hold the strings. An empty string should return a valid
pointer const char*. Now that the empty strings are actually printed
as "", we should not have a problem with empty strings in the
std::string objects. Hopefully the problem will go away. If it
persists, I will need some details..
_______________________________________________
FastBit-users mailing list
[email protected]
https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users
