In some places the IO library simply assumes that the IO request is 'small
enough', and happily allocates a buffer big enough for the request to be
handled in one go. Unformatted IO already uses sread()/swrite() which bypasses
buffering if the request is bigger than the buffer size. However, this has not
been done for formatted IO. In some cases this results in excessive memory
usage. A rather extreme (and admittedly contrived) example is

program bigformdir
  integer :: i = 10**9
  open (10, file="foo", form="formatted", access="direct", recl=i)
  write (10, '(i3)', rec=1) 1
  close (10)
end program bigformdir

The above program will cause libgfortran to allocate 1 GB, fill it with ' ',
and then write it. Clearly the allocation is not necessary.


-- 
           Summary: Unbounded I/O buffer memory usage for formatted IO
           Product: gcc
           Version: 4.2.0
            Status: UNCONFIRMED
          Severity: minor
          Priority: P3
         Component: libfortran
        AssignedTo: jb at gcc dot gnu dot org
        ReportedBy: jb at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25949

Reply via email to