[GRASS-user] i.rectify - ERROR: error while writing to temp file

[Glynn Clements]

Hamish wrote:

> > I'm trying to rectify some scanned maps, but I've
> > this error:
> > 
> > ERROR: error while writing to tmp file
> > 
> > I've checked the write permission of .tmp directory and
> > it's ok...
> > 
> > I've 15 GB of free space.
> > 
> > How can I solve this problem?
> > 
> > The .tmp directory where the .tmp files are saved, are the
> > source LOCATION (xy) or target LOCATION?
> 
> 
> No idea, but the help page says this:
> ---
> If i.rectify starts normally but after some time the following text is seen:
> GIS ERROR: error while writing to temp file
> the user may try the flag -c (or the module needs more free space on the hard drive).
> ---
> 
> The -c flag: "Use curr. region settings in target location (def.=calculate smallest area)"
> 
> 
> 
> the code that triggers that is this, in i.rectify/write.c:
> ---
> offset = ((off_t) row++ * target_window.cols + col) *
>           G_raster_size(map_type);
> lseek(temp_fd,offset,SEEK_SET);
> 
> if(write(temp_fd,cell_buf[n],G_raster_size(map_type)*matrix_cols) 
>                           != G_raster_size(map_type)*matrix_cols)
> {
>     unlink(temp_name);
>     G_fatal_error("error while writing to temp file");
> }
> ---

However: note that nothing prior to the above checks that the temp_fd
returned from creat() is valid. So, if creat() fails, the first
indication will be that write() fails (with EBADF).

I suggest adding an explicit check for the return value from creat(),
i.e.:

        temp_fd = creat(temp_name,0660);
+	if (temp_fd < 0)
+		G_fatal_error("unable to open temporary file");

This will make it possible to distinguish between an error when
opening the file (e.g. EPERM = "permission denied") and an error when
writing to the file (e.g. ENOSPC = "no space left on device").

> ?does the (off_t) cast only affect nearest row++ and not the whole
> (row*col+col)? or does the whole thing get promoted once one part of
> the calc is cast to off_t? or is that ok?

In C, types propagate upwards from the leaves. The result type of a
diadic operator (addition, multiplication) etc is equal to the
"largest" operand type, where long long > long > int, and double is
deemed larger than all integral types.

When dealing with off_t, the general rule is to cast one "limb" of any
multiplication to off_t (it's invariably multiplications where any
truncation will occur). The result of the multiplication will then be
off_t, as will the result of any enclosing expression involving that
value.

So, in the above code, the off_t propagates up to the result of the
first multiplication, and then to the addition, then to the second
multiplication, and then to the assignment.

OTOH, this won't work:

	offset = (off_t) ((row++ * target_window.cols + col) * G_raster_size(map_type));

as all of the calculations will be truncated to lesser types (int,
long, size_t), and by the time that the value is cast to off_t
(immediately before the assignment), it has already been truncated.

-- 
Glynn Clements <glynn at gclements.plus.com>