Re: overly aggressive memory usage by sort.c

[Jeff Janes]

On Mon, Jun 11, 2012 at 3:02 AM, Paul Eggert <address@hidden> wrote:
> On 06/10/2012 11:30 PM, Jim Meyering wrote:
>> Any objection to reverting that and adjusting the comment to match?
>
> Yes, as this would run afoul of bug 10877
> <http://bugs.gnu.org/10877>.

My reading of bug 10877 was as a complaint about using too small of an
assumption for piped input, with the rest of the issue (using 1/16
rather than 1/8 of physmem, and coincidentally using avail/2 rather
than avail) just being a side-track.  So the patch applied there did
not fix a verified problem.  It changed a long-standing behavior in
order to make a long-erroneous comment become accurate--but since
there is no reason that I know of to think that the new behavior is
better than the old one, I think changing the comment would have been
better than changing the behavior.  On the other hand the "new"
behavior is already two releases old, so you could argue that
reverting it entirely at this point is not particularly conservative.
I do have some evidence that changing to use all of avail_mem was too
aggressive, but do we have evidence one way or the other about
physmem/8 versus physmem/16?

Cheers,

Jeff