It seems that python 2.4.2 file reading stuff (for l in
f) and (f.read()) cache EOF state. For f.next
documtation hints that there is internal buffer so I
guess that this is somewhat acceptable there. But for
f.read() there is no hint about internal buffering of
the state.

This is somewhat unfortunate as tail -f like
functionality is impossible to implement without doing
lowlevel io (os.read) (as traddional unix has no
select-like functionality on files).

Tested on solaris8. This is either library or
documentation bug.

Code below

def iterline(f):
    """ Own function because the internal seems to
buffer EOF
    """
    b=[]
    while 1:
        data=os.read()
        print `data`,`b`,f.tell(), \
               os.fstat(f.fileno()).st_size
        if len(data) == 0:
            time.sleep(1)
            continue
        idx=data.find('\n')
        while idx > 0:
            line=data[:idx]
            if b:
                yield ''.join(b) + line
                b = []
            else:
                yield line
            data=data[idx+1:]
            idx=data.find('\n')
        b.append(data)
    raise StopIteration

Output
'' [''] 1699424 1699424
(log lines appended here)
'' [''] 1699424 1699647

In Linux, it seems to be the behaviour of the underlying C function 'fread'.

Do you think it's ok to add the following line in the read() documentation?

"""
As this function depends of the underlying C function :cfunc:`fread`,
inheritates its behaviour in details like caching EOF and others.
"""

Assigning this to George, to reword that in nicer English and apply.

Regards,

Added note in r64635.