Differences between revisions 39 and 40
Revision 39 as of 2013-09-18 21:56:47
Size: 7728
Editor: DougEvans
Comment:
Revision 40 as of 2013-11-01 04:45:18
Size: 8378
Editor: DougEvans
Comment:
Deletions are marked like this. Additions are marked like this.
Line 108: Line 108:

== GDB is printing "optimized out" when I try to print a variable ==

  . This commonly happens when debugging optimized code.
  The compiler is far more interested in generating optimal code than in generating code for which GDB can reconstruct a variable's value at an arbitrary point in the program.
  And that's A Good Thing for high levels of optimization of course.
  When this happens the compiler may generate debug information for a variable, but will leave out telling GDB how to fetch the variable's value (because it's either impossible or too difficult).
  GDB recognizes this and then prints such variables as "optimized out".

FAQ

Contents

  1. Where is the main GDB project page?
  2. Is there online documentation available?
  3. How do I disable the "Type <return> to continue, or q <return> to quit" pagination prompt in GDB?
  4. How do I redirect output from GDB to a file?
  5. How can I get GDB to remember command history across sessions?
  6. GDB reports a nonsensical return value from an inferior function call. What's going on?
  7. How do I load/unload a shared library in GDB?
  8. How to show the current instruction when single-stepping instructions?
  9. GDB reports "Cannot find user-level thread for LWP 23957: generic error", how do I fix this?
  10. GDB does not see any threads besides the one in which crash occurred; or SIGTRAP kills my program when I set a breakpoint.
  11. GDB 6.8 doesn't compile with GCC x.y because of -Werror, what should I do?
  12. Why doesn't anybody on the GDB IRC channel answer my question about my small distro's GDB, or my custom GDB?
  13. I've run into a bug in GDB while using XCode. Can you help?
  14. How do I print an STL container?
  15. When connecting to gdbserver I get "Remote register badly formatted", "g packet reply to long" etc.
  16. When I try "break 1" or "list" I get: "No symbol table is loaded."
  17. Ending of the string is truncated with "...", is there a way to get the whole string?
  18. How to trace every function entry and return?
  19. Getting an internal error or other error while attaching to processes on GNU/Linux
  20. Is there a way to set commands to be run on a segfault?
  21. Is there a way to step into the last call (foo) without stepping into the inner calls (bar, baz)? "foo (bar (), baz ());"
  22. GDB is printing "optimized out" when I try to print a variable

1. Where is the main GDB project page?

2. Is there online documentation available?

3. How do I disable the "Type <return> to continue, or q <return> to quit" pagination prompt in GDB?

  • Either with set height 0 or set pagination off

4. How do I redirect output from GDB to a file?

  • See help set logging

5. How can I get GDB to remember command history across sessions?

  • (gdb) set history save on
    (gdb) set history filename ~/.gdb-history
    (gdb) set history size 1000
  • Note: This may become the default, but for older GDBs do the above.

6. GDB reports a nonsensical return value from an inferior function call. What's going on?

  • GDB doesn't know the return type nor the type of the arguments for that function call, because there's no debug information available for it. Either provide debuginfo for the program or library which contains the function, or cast the function to a function pointer of the appropriate signature.

    For example, to call fabs, which takes a double and returns a double, use:

    (gdb) print ((double (*) (double)) fabs) ( -1.25 )

7. How do I load/unload a shared library in GDB?

  • GDB doesn't manipulate shared libraries. This is done by the operating system's dynamic linker running in the inferior. GDB just obtains the list of shared libraries from it, and works with that.
  • If one wants to manually load a shared library into the inferior one can manually call dlopen and dlsym, or the equivalent in your system, if there is one. However you have to know whether the inferior is stopped at a safe point to do so; often that's the case, but caveat emptor.

  • (gdb) set $dlopen = (void*(*)(char *, int)) dlopen
    (gdb) set $my_lib = $dlopen ("/tmp/mylib.so", 1) // 1 == RTLD_LAZY
    (gdb) set $dlsym = (void*(*)(void*, char *)) dlsym
    (gdb) set $foo = $dlsym ($my_lib, "foo") // int foo () { return 42; }
    (gdb) p $foo ()
    $1 = 42
    (gdb) set $dlclose = (int(*)(void*)) dlclose
    (gdb) call $dlclose ($my_lib)
  • See LoadingCodeIntoActiveSession for a more complete example.

8. How to show the current instruction when single-stepping instructions?

  • (gdb) display/i $pc

9. GDB reports "Cannot find user-level thread for LWP 23957: generic error", how do I fix this?

  • There are several common causes:
    • You have a mismatch between libthread_db.so.1 and libpthread.so.0 (this most often happens when you have multiple installations of glibc, or when you debug a program on remote target, and host and target have different glibc versions).

    • You are using 64-bit debugger to debug 32-bit program, and your kernel has a 32-bit ptrace emulation bug. FIXME: add reference to specific kernel fix.

    This has also been known to happen when one of DOSEMU's signal handlers is invoked from DPMI context, where the $gs register has a value different from what GDB and/or libthread_db.so.0 expect; SamuelBronson found running the program under gdbserver to alleviate the problem, at least with version 6.8.50.20090620-cvs on i386.

10. GDB does not see any threads besides the one in which crash occurred; or SIGTRAP kills my program when I set a breakpoint.

  • This frequently happen on Linux, especially on embedded targets. There are two common causes:
    • you are using glibc, and you have stripped libpthread.so.0

    • mismatch between libpthread.so.0 and libthread_db.so.1

    GDB itself does not know how to decode "thread control blocks" maintained by glibc and considered to be glibc private implementation detail. It uses libthread_db.so.1 (part of glibc) to help it do so. Therefore, libthread_db.so.1 and libpthread.so.0 must match in version and compilation flags. In addition, libthread_db.so.1 requires certain non-global symbols to be present in libpthread.so.0. Solution: use strip --strip-debug libpthread.so.0 instead of strip libpthread.so.0.

11. GDB 6.8 doesn't compile with GCC x.y because of -Werror, what should I do?

  • GDB 6.8 was released with -Werror configured in by default. You can disable it by configuring with:
  • $ /path/to/gdb-6.8/configure --disable-werror

12. Why doesn't anybody on the GDB IRC channel answer my question about my small distro's GDB, or my custom GDB?

  • We can help with FSF's original GDB, and also with the GDBs shipped by the biggest distros. if you are lucky, somebody might know some peculiarities about some small distro's GDB.

13. I've run into a bug in GDB while using XCode. Can you help?

  • Apple has their own fork of GDB, and they are no longer pulling changes from the FSF GDB. Any bugs in this fork should be reported to Apple; there's nothing we can do about them. The FSF GDB has recently been ported to MacOS; however, it is believed that it will not work in XCode.

14. How do I print an STL container?

15. When connecting to gdbserver I get "Remote register badly formatted", "g packet reply to long" etc.

  • Your gdb 32bit/64bit architecture setting may not match that of gdbserver. Try to load the executable by file first. You may also check set/show architecture.

16. When I try "break 1" or "list" I get: "No symbol table is loaded."

  • Compile the debugged program with gcc option -g.

17. Ending of the string is truncated with "...", is there a way to get the whole string?

  • Use set print elements -1.

18. How to trace every function entry and return?

  • With recompilation try itrace.c. Without recompilation using SystemTap try a SystemTap script. Without recompilation using GDB try trace.pl GDB commands generator. Otherwise just for breakpoint on each function entry you can use GDB rbreak. With recompilation one can also put breakpoint on __cyg_profile_func_enter and __cyg_profile_func_exit as done in itrace.c.

19. Getting an internal error or other error while attaching to processes on GNU/Linux

  • Try setenforce 0 (SELinux) or echo 0 >/proc/sys/kernel/yama/ptrace_scope (ptrace scope) to disable system security protections.

20. Is there a way to set commands to be run on a segfault?

  • Without GDB you can setup (with Linux kernel) /proc/sys/kernel/core_pattern. See man 5 core. It is used by bugreporting tools like ABRT or Apport.

21. Is there a way to step into the last call (foo) without stepping into the inner calls (bar, baz)? "foo (bar (), baz ());"

  • Such native feature is missing in GDB (DW_TAG_GNU_call_site should make it possible now).

  • There are several workarounds though:
  • Write a macro using record, next, reverse-step, reverse-finish, step, record stop if the inferior functions do not run for too long.

  • Get ignorefunc.cmd and use ignorefunc bar and ignorefunc baz before doing step.

  • Use the skip command. Do skip bar and skip baz before doing step.

22. GDB is printing "optimized out" when I try to print a variable

  • This commonly happens when debugging optimized code. The compiler is far more interested in generating optimal code than in generating code for which GDB can reconstruct a variable's value at an arbitrary point in the program. And that's A Good Thing for high levels of optimization of course. When this happens the compiler may generate debug information for a variable, but will leave out telling GDB how to fetch the variable's value (because it's either impossible or too difficult). GDB recognizes this and then prints such variables as "optimized out".

None: FAQ (last edited 2014-04-09 17:20:45 by brobecke)

All content (C) 2008 Free Software Foundation. For terms of use, redistribution, and modification, please see the WikiLicense page.