Re: Named scratch files



In article <1haqqe2.8cxt5r12bonweN%nospam@xxxxxxxxxxxxx>,
nospam@xxxxxxxxxxxxx (Richard E Maine) wrote:

How would you place a scratch file on a particular device? Your
solution does not solve the problem.

That is pretty fundamentally a non-portable issue anyway, since you need
to know what devices are available, what the appropriate one is (i.e.
which one has enough space), and what the path name syntax is. Realize
that the Fortran standard doesn't even have a concept of a device or a
path name. There is nothing but the file name. It is system-dependent
that syntax within the file name might relate to devices.

Yes, but that is the same kind of nonportability that we address all
the time with named nonscratch files. I just want the same
flexibility, nonportable as it is, for scratch files, along with the
advantages that 'scratch' brings.

An environment variable to redirect all unnamed scratch files to
specific place does not address the issue either. In my
applications over the past 30 years, I have always addressed
problems that are right at the border of what is and isn't possible.
This includes file space limitations. If I have access to a
half-dozen disks (or network file systems, or whatever), then I will
find a way to use all of them at the same time. With named files, I
can do that. With the limitations of fortran scratch files, I have
never been able to do that. An environment variable to redirect all
scratch files to the same device does not have sufficient
flexibility either. There are some advantages of scratch files that
I would have liked to have exploited over the past 30 years, but I
have not been able to do so. If the standard would have simply
allowed the filename to be specified for scratch files, then my
problem would have been solved in, as far as the source code is
concerned, a portable way.

I do admit that disk space is cheap enough these days, and
filesystems are finally getting beyond the 2GB limit, so that this
is not such a pressing issue. Now for a few thousand bucks you can
buy a multi-TB RAID system, and you don't have to worry so much
about spreading specific files among specific locations. But for
the past 30 years, the decision made in the f77 standard has
hampered my progress. That was an extremely shortsighted decision.
30 years was too long, that problem should have been corrected long
ago. Or, IMO, it should never have been a problem in the first
place.

$.02 -Ron Shepard
.