[patch, libgfortran] Fix PR20257 End of record occurs when writing large arrays

[Brooks Moses]

FX Coudert wrote:
>>We can reserve a few numbers for internal use - say, 2**31-1024 to 2**31-1
>>(the maximum positive GFC_INTEGER_4).  Attempts to open or otherwise use
>>these should be an error, and inquire(exist) could return false, like
>>negative numbers right now.
> 
> It might be a stupid suggestion (I don't have access to the code right 
> now): why not use negative numbers?

This seems like a very parallel case to user-defined derived-type I/O 
procedures in Fortran 2003 (Section 9.5.3.7.2).  When such a procedure 
is called for an internal file, "the UNIT argument shall have a 
processor-dependent negative value".  (Note that this is the argument 
that's passed by the processor to a user-defined subroutine, not the 
other way around.)  So, once we implement that, we'll have to be doing 
something involving identifying internal files by a negative unit number 
anyway; might as well go ahead and do it now and make things consistent.

...

Also, on a slightly related note (but I wrote this all before I realized 
that the above case existed, so I might as well send it), we'll 
eventually be connecting negative unit numbers to external files as 
well.  There's an addition to the Fortran syntax for the standard after 
f2003 that would allow for automatic allocation of unit numbers, and in 
particular uses negative unit numbers for this allocation (to avoid any 
possible conflicts with unit numbers in existing code).  More info in 
this thread on comp.lang.fortran, and in the actual J3 work item:

http://groups.google.com/group/comp.lang.fortran/browse_frm/thread/5c2aebdfc608c23d/
http://j3-fortran.org/doc/year/06/06-138r2.txt

How this affects the present case, in my opinion, is that gfortran will 
eventually need to support negative unit numbers whether it currently 
does or not, so there's no harm in crossing that bridge now.  Also, it 
appears to me that there should be no future problem with blocking out 
chunks of negative unit numbers for internal use, as they will only be 
automatically assigned rather than user-assigned so we don't have to 
worry about conflicts with user code even in future Fortran versions.

- Brooks