Infamous PR fortran/19925

[Steve Kargl]

Well, I spent some time last week thinking about PR
fortran/19925 and have decided that the best option
may be to punt.  Consider the somewhat far fetched 
but legal Fortran, 

   program test
     implicit none
     integer j
     integer, parameter :: n = 1000000
     integer, parameter :: i(n) = (/ (2*j, j = 1, n) /)
     real :: x(i(5))
     do j = 1, i(5)
        x(j) =  sin(i(j) * 1.23)
     end do
     print *, x
   end program test

Most possible solutions that I could fathom would inhibit
the optimizer from possibly optimizing the above loop.  

So, I've come up with the following patch.  It does not
fix the compile time problem, the memory consumption 
problem during compilation, nor the size of an object file.
It simply gives a user a method (or more appropriately a
gun) to shoot their feet if they do something like the above.
The above code will not compile with gfortran.  With my 
patch, one can do

gfortran -fmax-array-construct=1000000 test.f90

If the gfortran maintainers prefer a better solution, then I'll
certainly won't be offended.

2008-09-13  Steven G. Kargl  <kargls@comcast.net>

	PR fortran/19925
	* trans-array.c (gfc_trans_array_constructor_value): Replace an
	internal_error with gfc_error_now.
	* array.c: Remove GFC_MAX_AC_EXPAND
	(gfc_expand_constructor): Use gfc_option.flag_max_array_constructor.
	* gfortran.h: Add flag_max_array_constructor
	* lang.opt: Add -fmax-array-constructor option.
	* invoke.texi: Document option.
	* options.c (gfc_init_options): Set
	gfc_option.flag_max_array_constructor
	(gfc_handle_option): Process option.


-- 
Steve
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pr19925.diff
Type: text/x-diff
Size: 5881 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20080914/96691bb1/attachment.bin>