This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: -frecord-marker
- From: Bud Davis <bdavis9659 at sbcglobal dot net>
- To: Adrian Umpleby <apumpleby at yahoo dot co dot uk>, fortran at gcc dot gnu dot org
- Date: Wed, 26 Jul 2006 06:21:26 -0700 (PDT)
- Subject: Re: -frecord-marker
- Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=sbcglobal.net; h=Message-ID:Received:Date:From:Subject:To:In-Reply-To:MIME-Version:Content-Type:Content-Transfer-Encoding; b=ebjqMzgNJNgma3QX4XXzTVoenQ2LJwEBI4CHWDIYXfo9iel9HxKCP8K3wC7ad9GhpMbLD6uIgJxQQzfwVt9UJvaVXPnx5GL/+A5SdBthhSHEFLnlVyME1reodxv4za9xZeNOHwwyoh9OJqIZA5w1RYZK8pXbXrY8jrKzrcRgJ6U= ;
adrian,
your analysis is correct AFAICT.
my goal is to convince someone to have the front end
figure out how many bytes are going to be in the
record on the first I/O call..then no seek is needed,
and the solution is optimal. But that is not as easy
as it sounds.
(At may day job this is a time honored technique known
as "allocating something you don't know how to do
someone else...problem solved" :):)
any seek takes time, on a locally mounted file system
they are neglible compared to everything else, but
over NFS or other networked file systems they can be
very expensive. i rememeber some g77 bug reports
filed about poor performance of I/O over NFS and they
were caused by the seek, and of course, they were
unfixable without changing the mechanisation of the
record markers.
in this thread there was some talk about detecting the
record marker format automatically. i don't think
that can be done reliably enough to be used in a
compiler / runtime. there will be user data sets that
confuse the algorithm. IMHO, this is something that
has to be correct 100% of the time.
this is my second posting in a week. maybe i have
time to start writing code ?
take care,
bud