Bug in roaring bitmaps

[tgruben]

Seems to be an off by one kinda problem that shows up in the iterator. I've managed to narrow it to somewhere in the container allocation i think. I've got two import files that only differ by one line and one works and the other is broken. I happens when the highbits it the magic number of 1024, specifically the j value is between 65536 and 65599 inclusive.

broke.txt
works.txt

[tgruben]

you have a quick thought on that @benbjohnson ? specifically around line hb := int(itr.j / 64)
Line 373(roaring/roaring.go). Basically what should be done when hb>1023. Sorry for the crunch, just trying to get some numbers together for a meeting and this was kinda a road blocker

[tgruben]

I think i managed to fix it @benbjohnson, the behavior seems to match what's going on over in the roaring project.

[benbjohnson]

@tgruben Awesome, sorry I wasn't able to dive into it earlier. What was the fix?

[tgruben]

i went to the next container if the highbits where too big. I can't seem to find out why the bits would be to big, but in the other roaring project, they handled that situation the same way. @benbjohnson I'm "pretty" confident it was a good fix, all the other roaring test work. i'm gonna see if i can make a test that produces the same scenario our sample data presented.