On Wed, Jul 20, 2022 at 4:52 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I think we could probably just drop fls() entirely. It doesn't look
> to me like any of the existing callers expect a zero argument, so they
> could be converted to use pg_leftmost_one_pos32() pretty trivially.
> I don't see that fls() is buying us anything that is worth requiring
> readers to know yet another nonstandard function.
That was not true for the case in contiguous_pages_to_segment_bin(),
in dsa.c. If it's just one place like that (and, hrrm, curiously
there is an open issue about binning quality on my to do list...),
then perhaps we should just open code it there. The attached doesn't
trigger the assertion that work != 0 in a simple make check.