From octave-maintainers-request at bevo dot che dot wisc dot edu Thu Nov 20 23:51:46 2003
Subject: file offsets used by fseek and ftell
From: "John W. Eaton" <jwe at bevo dot che dot wisc dot edu>
To: octave-maintainers mailing list <octave-maintainers at bevo dot che dot wisc dot edu>
Date: Thu, 20 Nov 2003 23:51:44 -0600

In Matlab, fseek and ftell use integer values stored in double
objects to represent file positions.  In C, fseek and ftell use long.
So this combination works OK (if you don't care about large files) on
systems with 32-bit longs and IEEE 64-bit doubles (which can represent
52-bit integer values).  But it doesn't work so well if you care about
large files (the 32-bit signed value can only handle 1GB files).

In Octave, we use C++ streams instead of using fseek and ftell from
the C library.  So we have to deal with streamoff and streampos
objects, which are not guaranteed to have any conversions to integer
values.  So what should we do?  If we introduce new streampos and
streamoff types, I imagine it would be useful to have arrays of them,
but I don't think we will be able to stuff these new values into
double precision values reliably.  I suggest a new type that works
like streamoff, which will be returned from ftell.  You can convert an
integer value into a streamoff object, but if you have a streamoff
object, you can't see what is inside it.   Also, you can add,
subtract, and compare streamoff objects.

Having this new type would allow us to handle large files (and also
build Octave with gcc 3.4, which no longer defines streampos and
streamoff with simple typedefs to long).

Comments or suggestions?

jwe