[patch, fortran] PR22539 implement FSEEK intrinsic

[Jerry DeLisle]

Daniel Franke wrote:
> On Tuesday 01 May 2007 19:35:02 Brooks Moses wrote:
>>> +/* FSEEK intrinsic */
>>>
>>> +extern void fseek_sub (int *, GFC_IO_INT *, int *, int *);
>>> +export_proto(fseek_sub);
>>> +
>>> +void
>>> +fseek_sub (int * unit, GFC_IO_INT * offset, int * whence, int * status)
>>> +{
>>> +  gfc_unit * u = find_unit (*unit);
>>> +
>>> +  if (status) *status = 0;

A minor nit, coding convention puts this on two lines like so:

if (status)
   *status = 0;

>>> +
>>> +  if (u == NULL || !is_seekable(u->s))
>>> +    {
>>> +      if (status) *status = -1; /* EBADF */
Same nit here
>>> +    }
>>> +  else if (*whence == 0)
>>> +    sseek(u->s, *offset);                       /* SEEK_SET */
>>> +  else if (*whence == 1)
>>> +    sseek(u->s, file_position(u->s) + *offset); /* SEEK_CUR */
>>> +  else if (*whence == 2)
>>> +    sseek(u->s, file_length(u->s) + *offset);   /* SEEK_END */
>>> +  else if (status)
>>> +    *status = -1;  /* EINVAL */
>>> +}
>> I don't see how this returns a nonzero status if *offset is outside the
>> limits of the file such that the sseek fails.  How does that work?
> 
> It does not fail. 
> The sseek macro, eventually, calls lseek(2):
> 
>   "The lseek() function allows the file offset to be set beyond the end of
>   the file (but this does not change the size of the file).  If  data  is
>   later written at this point, subsequent reads of the data in the gap (a
>   "hole") return null bytes ('\0') until data is  actually  written  into
>   the gap."
> 
> 
>> Also, what happens if sseek fails and STATUS isn't supplied; is this a
>> runtime error, or a silent failure?
> 
> If an error condition is met, no seek is performed. Possible errors of 
> sseek/lseek are:
> 
>  * EBADF:     bad file descriptor (u == NULL)
>  * EINVAL:    whence < 0 || whence > 2
>  * EOVERFLOW: offset to large (handled by largest int kind GFC_IO_INT)
>  * ESPIPE:    file not seekable (!is_seekable(u->s))
> 
> None of these conditions will ever be propagated to lseek. G77 does not error 
> out if no label is specified. Nonetheless, a runtime error might be useful in 
> some cases. Maybe Jerry has a more definite opinion on this?!
> 
> I will clarify the documentaion according to your suugestions.
> 
> Regards
> 	Daniel
> 
If you look in unix.c, the sseek function calls fd_seek which is of type "try". 
  So one can test the return from sseek for SUCCESS or FAILURE.  In STREAM mode 
we use this to test for the EOF condition.

I would suggest a variable be declared:  try result

Then use:  result = sseek (...)

status = 0 if result is SUCCESS. otherwise -1.

Generally, short of a hardware failure, you can't get an error.  I have not seen 
what happens if we try to seek a read only file past its current end.  That 
would be worth checking.

Sorry I was slow looking this over.  Been busy :)

Jerry