This is the mail archive of the
gdb-patches@sources.redhat.com
mailing list for the GDB project.
Re: [rfa/testsuite] gdb1250.exp, new test script
- From: Daniel Jacobowitz <drow at mvista dot com>
- To: gdb-patches at sources dot redhat dot com
- Date: Tue, 15 Jul 2003 12:19:38 -0400
- Subject: Re: [rfa/testsuite] gdb1250.exp, new test script
- References: <200307151613.h6FGDvnQ011927@duracef.shout.net>
On Tue, Jul 15, 2003 at 12:13:57PM -0400, Michael Elizabeth Chastain wrote:
> Rats, I said I would do this yesterday. Here it is: a new test script
> for PR gdb/1250.
>
> http://sources.redhat.com/gdb/bugs/1250
>
> This is the bug where gdb loses its marbles when backtracing through a
> function which calls a noreturn function such as 'abort'. The calling
> function has no epilog after the call to 'abort', so when gdb looks at
> the frame for that function, gdb sees the first instruction of the
> *next* function and uses information for that function instead.
>
> This happens in gdb.base/corefile.exp but I think it is nice to have a
> specific test for it.
>
> I tested this on native i686-pc-linux-gnu with gdb HEAD, gcc 2.95.3 and
> v3, dwarf-2 and stabs+.
>
> It PASSed with gcc 2.95.3 because gcc 2.95.3 does not optimize away the
> epilog. I think that this is okay. The gdb user really just wants to
> put breakpoints on things like 'abort' and 'exit' and have it work, and
> if it works because the compiler is simple, that is okay.
>
> It KFAILed with all the gcc 3.3's that I used.
>
> I would like to commit this to HEAD, wait a few days or a week, and then
> commit it to gdb_6_0-branch.
>
> OK to commit?
You misspelled exiting as "exitting". Other than that, it looks fine
(with ChangeLog); please check it in on HEAD. I'm not sure about the
branch.
>
> Michael C
>
> ===
>
> # Copyright 2003 Free Software Foundation, Inc.
>
> # This program is free software; you can redistribute it and/or modify
> # it under the terms of the GNU General Public License as published by
> # the Free Software Foundation; either version 2 of the License, or
> # (at your option) any later version.
> #
> # This program is distributed in the hope that it will be useful,
> # but WITHOUT ANY WARRANTY; without even the implied warranty of
> # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> # GNU General Public License for more details.
> #
> # You should have received a copy of the GNU General Public License
> # along with this program; if not, write to the Free Software
> # Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
>
> # Tests for PR gdb/1250.
> # 2003-07-15 Michael Chastain <mec@shout.net>
>
> # This file is part of the gdb testsuite.
>
> if $tracelevel then {
> strace $tracelevel
> }
>
> #
> # test running programs
> #
> set prms_id 0
> set bug_id 0
>
> set testfile "gdb1250"
> set srcfile ${testfile}.c
> set binfile ${objdir}/${subdir}/${testfile}
>
> if { [gdb_compile "${srcdir}/${subdir}/${srcfile}" "${binfile}" executable {debug}] != "" } {
> gdb_suppress_entire_file "Testcase compile failed, so all tests in this file will automatically fail."
> }
>
> gdb_exit
> gdb_start
> gdb_reinitialize_dir $srcdir/$subdir
> gdb_load ${binfile}
>
> if ![runto abort] then {
> perror "couldn't run to breakpoint"
> continue
> }
>
> # See http://sources.redhat.com/gdb/bugs/1250
> #
> # In a nutshell: the function 'beta' ends with a call to 'abort', which
> # is a noreturn function. So the last instruction of 'beta' is a call
> # to 'abort'. When gdb looks for information about the caller of
> # 'beta', it looks at the instruction after the call to 'abort' -- which
> # is the first instruction of 'alpha'! So gdb uses the wrong frame
> # information. It thinks that the test program is in 'alpha' and that
> # the prologue "push %ebp / mov %esp,%ebp" has not been executed yet,
> # and grabs the wrong values.
> #
> # By the nature of the bug, it could pass if the C compiler is not smart
> # enough to implement 'abort' as a noreturn function. This is okay.
> # The real point is that users often put breakpoints on noreturn
> # functions such as 'abort' or some kind of exitting function, and those
> # breakpoints should work.
>
> gdb_test_multiple "backtrace" "backtrace from abort" {
> -re "#0.*abort.*\r\n#1.*beta.*\r\n#2.*alpha.*\r\n#3.*main.*\r\n$gdb_prompt $" {
> pass "backtrace from abort"
> }
> -re "#0.*abort.*\r\n#1.*beta.*\r\n$gdb_prompt $" {
> # This happens with gdb HEAD as of 2003-07-13, with gcc 3.3,
> # binutils 2.14, either -gdwarf-2 or -gstabs+, on native
> # i686-pc-linux-gnu.
> #
> # gdb gets 'abort' and 'beta' right and then goes into the
> # weeds.
> kfail "gdb/1250" "backtrace from abort"
> }
> }
>
> ===
>
> /* Test program for stack trace through noreturn function.
>
> Copyright 2003 Free Software Foundation, Inc.
>
> This file is part of the gdb testsuite.
>
> This program is free software; you can redistribute it and/or modify
> it under the terms of the GNU General Public License as published by
> the Free Software Foundation; either version 2 of the License, or
> (at your option) any later version.
>
> This program is distributed in the hope that it will be useful,
> but WITHOUT ANY WARRANTY; without even the implied warranty of
> MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> GNU General Public License for more details.
>
> You should have received a copy of the GNU General Public License
> along with this program; if not, write to the Free Software
> Foundation, Inc., 59 Temple Place - Suite 330,
> Boston, MA 02111-1307, USA.
>
> This file was written by Michael Elizabeth Chastain (mec@shout.net). */
>
> #include <stdlib.h>
>
> int global = 0;
>
> void gamma (int *parray)
> {
> return;
> }
>
> void beta ()
> {
> int array [4];
> array [0] = global++;
> array [1] = global++;
> array [2] = global++;
> array [3] = global++;
> gamma (array);
> abort ();
> }
>
> int alpha ()
> {
> global++;
> beta ();
> return 0;
> }
>
> int main ()
> {
> int i;
> global++;
> i = alpha ();
> return i;
> }
>
--
Daniel Jacobowitz
MontaVista Software Debian GNU/Linux Developer