Re: [patch,libgfortran] PR45723 opening /dev/null for appending writes fails

[Jerry DeLisle <jvdelisle at frontier dot com>]

On 09/20/2010 09:56 PM, Steve Kargl wrote:
> On Mon, Sep 20, 2010 at 07:54:11PM -0700, Jerry DeLisle wrote:
> > Hi,
> >
> > This patch is simple. Test case attached.
> >
> > Regression tested on x86-64.
> >
> > OK for trunk?
>
> Is the file size guaranteed to be 0 for all
> non-seekable files on all operating systems
> supported by gfortran?  If someone is stupid
> enough to open /dev/null for append, then
> he should be prepared to suffer the consequence.

You will notice the patch does nothing that presumes anything about the file, 
seekable or not.

All that patch is doing is not seeking if the length is zero or less. The 
filesize function being used will return a value less then zero for a few error 
conditions but in this case I simply chose to ignore it.  So, if the file length 
happens to be zero, there is no need to seek.  If that happens to work for 
/dev/null on some systems, no harm done.

From the typical man page:

"Data written to a null or zero special file is discarded." So my thinking, it 
should not matter "how" the file is opened if it is going to be discarded 
anyway.  I suppose a user may find it useful during development of a program to 
write data to /dev/null, for example during trial runs where you would normally 
append, but for testing, do not want to keep growing the file.  Who knows, not 
my place to question why someone might want to do this.  It was suggested by one 
of our more valued testors.  That was enough for me.

OK for trunk.  :)

Jerry