Re: Bug in intarray?

Guillaume Lelarge <guillaume(at)lelarge(dot)info> writes:
> This query:
> SELECT ARRAY[-1,3,1] & ARRAY[1, 2];
> should give {1} as a result.

> But, on HEAD (and according to his tests, on 9.0.6 and 9.1.2), it
> appears to give en empty array.

Definitely a bug, and I'll bet it goes all the way back.

> Digging on this issue, another user (Julien Rouhaud) made an interesting
> comment on this line of code:

> if (i + j == 0 || (i + j > 0 && *(dr - 1) != db[j]))

> (line 159 of contrib/intarray/_int_tool.c, current HEAD)

> Apparently, the code tries to check the current value of the right side
> array with the previous value of the resulting array. Which clearly
> cannot work if there is no previous value in the resulting array.

> So I worked on a patch to fix this, as I think it is a bug (but I may be
> wrong). Patch is attached and fixes the issue AFAICT.

Yeah, this code is bogus, but it's also pretty unreadable. I think
it's better to get rid of the inconsistently-used pointer arithmetic
and the fundamentally wrong/irrelevant test on i+j, along the lines
of the attached.

regards, tom lane