[patch, libfortran] PR40508 Memory Leak

[Jerry DeLisle <jvdelisle at verizon dot net>]

Hi all,

First to explain the problem.  Internal units have a gfc_unit structure that is 
allocated in memory at the beginning of an IO statement (read/write) and freed 
at the end.  This means that there is no unit structure retained from one 
statement to the next and therefore saving format data pointed to in that 
structure is moot.

The problem begins with bothering to save the format to begin with and ends with 
not freeing it within free_internal_unit.  This means that the following patch 
is actually the best way to go for now.

This hints at a further optimization we could do.  Why allocate an internal unit 
structure every time we do an I/O?  There may be a way to reuse an internal unit 
structure for repeated units.  I will save that for a later date.

I also have a confession to make. In format.c you will see where I added a bool 
save_format flag.  This was the start of disabling saving formats for specific 
cases that could be troublesome, internal units being one of those.  I failed to 
finish that part of the patch when I submitted it before.  Regardless, I am 
working on these related items as part of the CP2K bug we found.  When I get 
that all tested, I will submit a full patch.

In the meantime, OK to commit the following to trunk in the interim?

Regression tested on x86-64-linux-gnu.

Regards,

Jerry


2009-06-21  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

	PR libfortran/40508
	* io/format.c: Don't save parsed format data for internal units.

Index: format.c
===================================================================
--- format.c	(revision 148448)
+++ format.c	(working copy)
@@ -1218,7 +1218,8 @@ parse_format (st_parameter_dt *dtp)
       free_format_hash_table (dtp->u.p.current_unit);
       return;
     }
-  save_parsed_format (dtp);
+  if (!is_internal_unit (dtp))
+    save_parsed_format (dtp);
 }