Re: [PATCH, gfortran] Fix initialization of arrays of strings

[Feng Wang <wf_cs at yahoo dot com>]

--- Steve Kargl <sgk@troutmask.apl.washington.edu>写道:

> On Tue, Jul 12, 2005 at 01:13:52AM +0800, Feng Wang wrote:
> > --- Steve Kargl <sgk@troutmask.apl.washington.edu>????:
> > 
> >> On Mon, Jul 11, 2005 at 02:40:01PM +0800, Feng Wang wrote:
> >>> I think we should fix this in array.c
> >>> (resolve_character_array_constructor).
> >>> This is earlier than add_init_expr_to_sym. And we should also add
> >>> it to the gfc_current_ns->cl_list to void memory leak. The list
> >>> will be deallocated with the namespace.
> >>> 
> >> 
> >> I don't understand what you want to do.  I just watched the
> >> compiler go through resolve_character_array_constructor
> >> and gfc_set_constant_character_len and it appears to do want
> > 
> > Yes, it go through resolve_character_array_constructor.
> > 1. I think it is resolve_character_array_constructor that does
> >    not allocate character length struct compiling this test and
> >    cause ICE in add_init_expr_to_sym. 
> 
> module z
>    integer :: i
>    character(6) :: a(2) = (/ ('main  ' , i = 1, 2) /)  ! ICE.
>    character(6) :: b(2) = (/ 'abcd  ' , 'efghij' /)    ! Compiles fine.
> end module
> 
> In the case of b(2), everthing works as expected.  That is, memory is
> allocated during the resolve_character_array_constructor call.  In the
> a(2) case we get an ICE because the implied-do-loop is not expanded
> in resolve_character_array_constructor, only memory for 1 copy of 
> 'main  ' is allocated.  The implied-do-loop is expanded in 
> add_init_expr_to_sym and additional memory is allocated to propagate
> a copy of 'main  ' to all elements of a(2).

All the 'main  ' memory is allocated during gfc_set_constant_character_len
call. I mean that resolve_character_array_constructor does not call
gfc_get_charlen to set ts.cl if the constractor's expr is not of EXPR_CONSTANT
(a(2)). This cause an ICE when invoking gfc_free_expr (init->ts.cl->length) in
add_init_expr_to_sym. The memory I said is the memory of (gfc_charlen).

> 
> > 2. When invoking gfc_get_charlen, we should add its result to
> >    gfc_current_ns->cl_list.  In the gfc_get_charlen, we get some
> >    memory.  Add it to the list, it will be freed when its namespace
> >    is deallocated.
> 
> This is not important for my patch and this ICE.  Shouldn't gfc_get_charlen
> add its result to gfc_current_ns->cl_list instead of the calling routine?
> 
I also think this would be better.

Best Regards,
Feng Wang

--
Creative Compiler Research Group,
National University of Defense Technology, China.


	

	
		
___________________________________________________________ 
雅虎免费G邮箱－中国第一绝无垃圾邮件骚扰超大邮箱
http://cn.mail.yahoo.com/?id=77071