preprocessor/6084: cpp Segmentation Fault

[Gerwin, Joshua A]

> -----Original Message-----
> From: Phil Edwards [mailto:phil@jaj.com]
> Sent: Friday, April 12, 2002 1:27 PM
> To: joshg@hf.intel.com; gcc-gnats@gcc.gnu.org; 
> rnesius@ichips.intel.com;
> tuan.nguyen@intel.com; gcc-bugs@gcc.gnu.org; 
> neil@daikokuya.demon.co.uk
> Subject: Re: preprocessor/6084: cpp Segmentation Fault
> 
> 
> I haven't been able to reproduce this using today's 3.1 
> sources.  (More at
> the end.)
> 
My hunch is that this is either a 3.0 problem, perhaps fixed in your version
of 3.1, or that the libc.so we're linking during compile is significantly
different.  Note that I could produce a segfault in strlen with minimal
options and no source at all, simply by having both -dD and -DM present.

%cat empty.c
/* no source */
%which cpp
/path/to/gcc/3.0.3-64/bin/cpp
%cpp -dM -dD empty.c
cpp: Internal error: Segmentation Fault (program cpp0)
Please submit a full bug report.
See <URL:http://www.gnu.org/software/gcc/bugs.html> for instructions.

Running gcc, I observe

(gdb) file /path/to/gcc/3.0.3-64/lib/gcc-lib/sparcv9-sun-solaris2/3.0.3/cpp0
Load new symbol table from
"/path/to/gcc/3.0.3-64/lib/gcc-lib/sparcv9-sun-solaris2/3.0.3/cpp0"? (y or
n) y

Reading symbols from
/path/to/gcc/3.0.3-64/lib/gcc-lib/sparcv9-sun-solaris2/3.0.3/cpp0...done.
(gdb) run -dM -dD empty.c
Starting program:
/path/to/gcc/3.0.3-64/lib/gcc-lib/sparcv9-sun-solaris2/3.0.3/cpp0 -dM -dD
empty.c

Program received signal SIGSEGV, Segmentation fault.
0xffffffff7f44035c in strlen () from /usr/lib/64/libc.so.1
(gdb) bt
#0  0xffffffff7f44035c in strlen () from /usr/lib/64/libc.so.1
#1  0xffffffff7f48d6d4 in _doprnt () from /usr/lib/64/libc.so.1
#2  0xffffffff7f48f180 in fprintf () from /usr/lib/64/libc.so.1
#3  0x100003d28 in maybe_print_line ()
#4  0x100003e1c in cb_define ()
#5  0x100004904 in do_define ()
#6  0x100004778 in run_directive ()
#7  0x100011d20 in init_builtins ()
#8  0x100012344 in cpp_start_read ()
#9  0x100003680 in do_preprocessing ()
#10 0x10000349c in main ()
(gdb) 

Also noteworthy, reversing the parameters fixes the segfault!

(gdb) run -dD -dM
The program being debugged has been started already.
Start it from the beginning? (y or n) y
Starting program:
/usr/intel/pkgs/gcc/3.0.3-64/lib/gcc-lib/sparcv9-sun-solaris2/3.0.3/cpp0 -dD
-dM
^D
#define __USER_LABEL_PREFIX__
#define __HAVE_BUILTIN_SETJMP__ 1
#define __REGISTER_PREFIX__
#define __VERSION__ "3.0.3"

Program exited normally.
(gdb) run -dM -dD
Starting program:
/usr/intel/pkgs/gcc/3.0.3-64/lib/gcc-lib/sparcv9-sun-solaris2/3.0.3/cpp0 -dM
-dD
^D
Program received signal SIGSEGV, Segmentation fault.
0xffffffff7f44035c in strlen () from /usr/lib/64/libc.so.1

Here's the maybe_print_line function and related symbols, for handy
reference:

static void
maybe_print_line (line)
     unsigned int line;
{
  /* End the previous line of text (probably only needed until we get
     multi-line tokens fixed).  */
  if (print.printed)
    {
      putc ('\n', print.outf);
      print.lineno++;
      print.printed = 0;
    }

  if (options->no_line_commands)
    {
      print.lineno = line;
      return;
    }

  /* print.lineno is zero if this is the first token of the file.  We
     handle this specially, so that a first line of "# 1 "foo.c" in
     file foo.i outputs just the foo.c line, and not a foo.i line.  */
  if (line >= print.lineno && line < print.lineno + 8 && print.lineno)
    {
      while (line > print.lineno)
        {
          putc ('\n', print.outf);
          print.lineno++;
        }
    }
  else
    {
      print.lineno = line;
      print_line ("");
    }
}

static void
print_line (special_flags)
  const char *special_flags;
{
  /* End any previous line of text.  */
  if (print.printed)
    putc ('\n', print.outf);
  print.printed = 0;

  fprintf (print.outf, "# %u \"%s\"%s%s\n",
           print.lineno, print.last_fname, special_flags,
print.syshdr_flags);
}

(gdb) whatis print
type = struct printer
(gdb) ptype struct printer
type = struct printer {
    FILE *outf;
    char *last_fname;
    char *syshdr_flags;
    unsigned int lineno;
    unsigned char printed;
}

This is about all the source diving I can do for it today.

Josh Gerwin, System Programmer/Analyst
joshua.a.gerwin@intel.com
Intel Corp.
NorthWest Engineering Computing