1) it is a permenant storage of information persist beyond the current
execution (except when amode is DELETE_ON_CLOSE).
2) It does not have record size as in a Fortran file.
3) It supports random address access. So, sockets and Unix pipes are
excluded. (Is this really so?)
4) Smallest access unit is by bytes and it supports arbitary numbers
of bytes in each file access. It can grow and shrink in size as
directed.
5) It does not store access methods as in the sense of object-oriented
data model. An integer data written by one application
can be read as bytes by another application. It is up to each
application to interpret the meaning of data.
Let me put up a straw man definition and invite you all to "shot" at it.
A file is a permenant collection of ordered data bytes of 8 bits each and
it supports random address access to any non-negative integral bytes
of data.
--