Re: Strange coding in _mdfd_openseg()

On Wed, Apr 3, 2019 at 5:34 PM Kyotaro HORIGUCHI
<horiguchi(dot)kyotaro(at)lab(dot)ntt(dot)co(dot)jp> wrote:
> I may be missing something, but it seems possible that
> _mdfd_getseg calls it with segno > opensegs.
>
> | for (nextsegno = reln->md_num_open_segs[forknum];

Here nextsegno starts out equal to opensegs.

> | nextsegno <= targetseg; nextsegno++)

Here we add one to nextsegno...

> | ...
> | v = _mdfd_openseg(reln, forknum, nextsegno, flags);

... after adding one to opensegs. So they're always equal. This fits
a general programming pattern when appending to an array, the next
element's index is the same as the number of elements. But I claim
the coding is weird, because _mdfd_openseg's *looks* like it can
handle opening segments in any order, except that the author
accidentally wrote "<=" instead of ">=". In fact it can't open them
in any order, because we don't support "holes" in the array. So I
think it should really be "==", and it should be an assertion, not a
condition.

--
Thomas Munro
https://enterprisedb.com