> -----Original Message-----
> From: Branko Čibej [mailto:brane_at_wandisco.com]
> Sent: woensdag 26 augustus 2015 10:19
> To: Subversion Development <dev_at_subversion.apache.org>
> Subject: Re: svn commit: r1694533 -
> /subversion/trunk/subversion/libsvn_ra_svn/marshal.c
>
> On 25.08.2015 23:08, Stefan Fuhrmann wrote:
> >> All right, so I figured that the difference is that apr_array_make does
> >> two allocations compared to one in this code. Still: relying on
> >> knowledge of APR implementation details is a really bad idea.
> >
> > The structure definition of apr_array_header_t is part of
> > the public API, i.e. will never change.
>
> The semantics might, even if the shape of the structure itself doesn't.
> I still think this kind of performance hack belongs in APR. Users that
> don't have a new-enough APR won't get the performance boost, but on the
> other hand, the kind of bug that started this discussion will stay out
> of our code.
>
> I think we've had our fair share of alignment bugs with all the
> hand-crafted allocations etc.; we may as well stop now.
+1
Bert
Received on 2015-08-26 11:10:03 CEST