[3.0 regression] File naming confusion

[Neil Booth <neil at daikokuya dot demon dot co dot uk>]

Many moons ago, Benjamin Kosnik wrote:-

> I keep dropping the ball WRT the line-numbering issue. Here's a
> small example:

Here's a *really easy* testcase.  We MUST get this fixed for 3.0.

#line 500 "foo.c"
int
main ()
{
  return 0;
}

Compile with -g, and run gdb.  Put a breakpoint on main.  Run to see
the problem.

Clearly, the problem is the #line (or #pragma GCC system_header).

I'm 99% certain the bug is in c-lex.c in the cb_file_change function.
It goes like this [unimportant bits skipped]:-

  if (fc->reason == FC_ENTER)
    {
	[...]
	  push_srcloc (fc->to.filename, 1);
	  debug_start_source_file (fc->to.filename);
	[...]
    }
  else if (fc->reason == FC_LEAVE)
    {
	[...]
	  pop_srcloc ();
	  debug_end_source_file (input_file_stack->line);
	[...]
    }

This handles the cases of FC_ENTER (which is when entering the main
file or a #included file), FC_LEAVE (when leaving a header).  There is
one more case for which we do nothing special: FC_RENAME.  This is
used for #line and the #pragma.

It is clear to me that debug_* and maybe *_srcloc are needed for
FC_RENAME, but I'm not sure what.

I'm going to try a few things now.

Neil.