frame_register_unwind(): "frame != NULL" assertion failure

[Andrew Cagney]

> Here's one possible "fix" for the assertion failure.  I've come to the
> conclusion that the real culprit is get_next_frame(), but, unfortunately,
> the rest of gdb is not yet ready for it to return the sentinel frame.
> (See the comment in the patch for more info.)  So, until it is, I propose
> that we use this hack...
> 
> 	* frame.c (create_sentinel_frame): Make static.  Add forward
> 	declaration.
> 	(frame_register_unwind): Add hack for converting NULL frames
> 	into a sentinel frame.

I think, if there is going to be a hac, then it should be in 
read_next_frame_reg().

However, can you please first investigate modifying the start of 
mips_init_extra_frame_info() where it does:

proc_desc = get_next_frame (fci) ? .....

to, when get_next_frame(fci) is null, call:

	find_proc_desc (get_frame_pc (fci), fci, 1);

that is the current, and not prev frame.

Andrew



> Index: frame.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/frame.c,v
> retrieving revision 1.66
> diff -u -p -r1.66 frame.c
> --- frame.c	2 Feb 2003 20:31:43 -0000	1.66
> +++ frame.c	13 Feb 2003 23:20:22 -0000
> @@ -39,6 +39,8 @@
>  #include "command.h"
>  #include "gdbcmd.h"
>  
> +static struct frame_info * create_sentinel_frame (struct regcache *regcache);
> +
>  /* Flag to indicate whether backtraces should stop at main.  */
>  
>  static int backtrace_below_main;
> @@ -180,6 +182,15 @@ frame_register_unwind (struct frame_info
>    gdb_assert (realnump != NULL);
>    /* gdb_assert (bufferp != NULL); */
>  
> +  /* Note: kevinb/2003-02-13: This is a hack.  The problem is that
> +     get_next_frame() can return NULL when it really ought to be
> +     returning the sentinel frame.  So, when we detect frame == NULL,
> +     just use the sentinel frame instead.  
> +     FIXME: Remove this hack once get_next_frame() has been fixed
> +     to never return NULL.  */
> +  if (frame == NULL)
> +    frame = create_sentinel_frame (current_regcache);
> +
>    /* NOTE: cagney/2002-11-27: A program trying to unwind a NULL frame
>       is broken.  There is always a frame.  If there, for some reason,
>       isn't, there is some pretty busted code as it should have
> @@ -429,7 +440,7 @@ frame_map_regnum_to_name (int regnum)
>  
>  /* Create a sentinel frame.  */
>  
> -struct frame_info *
> +static struct frame_info *
>  create_sentinel_frame (struct regcache *regcache)
>  {
>    struct frame_info *frame = FRAME_OBSTACK_ZALLOC (struct frame_info);
> 
>