HP WDB 6.0 Release Notes

Support for Core File Debugging of MITR Applications

New Command for Searching a Pattern in the Memory Address Space

New Command to Unwind Beyond 10000 Frames

Updated Procedure to Process the Initialization File

Enhancement to the info shared Command

Enhancement to the info target Command”

Support for Automatic Debugging of Shared Libraries”

“Attaching a PA-RISC Debugger to a PA-RISC Process”

“Improvement in Performance of Memory Debugging for the string option

GDB shows ptype of the Fortran character-valued function as void.

WDB prints incorrect value while performing command-line calls through pointer or reference pointer of the member functions of the class.

The unpackcore command cannot handle the absolute path of the core file and fails with an error message that is not formatted properly.

GDB does not allow info leaks when another thread locks the mutex.

NaT bits are not displayed with info reg or $r15 from the signal frame in a core file.

GDB has issues while debugging MT applications linked with the liblwp library.

WDB misuses the DW_CHILDREN field of a DIE with type DW_TAG_structure_type.

The ptype -v command produces length of 0 or negative value.

In batch mode thread checking, the output directory cannot be specified in the rtcconfig file.

Certain command-line call issues exist for explicitly instantiated methods of template classes.

A mismatch occurs between the line number displayed for the breakpoint confirmation and the breakpoint arrival. This problem occurs only in case of FORTRAN programs, when the breakpoint is set on an internal procedure. For example, break internal_procedure. This problem is specific to Integrity systems.

A bad demangling in the stack trace occurs if the template is of type cast().

Heap profile lists biggest allocation last.

IA-64 debugger does not indicate dld.so mismatches between the core file and the system.

Issue with printing constant arrays during core file debugging.

GDB crashes while printing a C++ expression.

Problems in setting hardware watchpoints on local variables.

If the hardware watchpoint is used on an address of a local variable, the hardware watchpoint may trigger even if the variable goes out of scope due to the same memory location being reused by other stack frames. Similarly, a hardware watchpoint on a heap location may also trigger with other variables if the memory location gets deallocated and reallocated to other variables.

The Run Time Checking feature (Interactive and Batch Mode) of WDB cannot be used with applications that redefine or override the default system-supplied versions of the standard library routines (under libc.[sl|so] and libdld.[sl|so]).

Table 1 lists the dependent library routines for Run Time Checking using HP WDB.

The Run Time Checking (dynamic memory, libraries, and pthreads checking) in WDB is dependent on the semantics and the standard behavior of these library routines. Run Time Checking in WDB results in unexpected and unpredictable behavior when used with applications that substitute or redefine these library routines.

Before enabling the Run Time Checking feature in WDB, use the nm command to determine if your application or the dependent libraries in your application redefine or substitute these library routines.

Batch Mode RTC displays one of the following errors and causes the program to temporarily hang if the version of WDB and librtc.[sl|so] do not match, or if WDB is not available on the system:

/opt/langtools/bin/gdb: unrecognized option `-brtc'
Use `/opt/langtools/bin/gdb --help' for a complete list of options.

Or

execl failed. Cannot print RTC info: No such file or directory

This error does not occur under normal usage where WDB or librtc.[sl|so] is used from the default location at /opt/langtools/...

However, this error occurs if GDB_SERVER and/or LIBRTC_SERVER are set to a mismatched version of WDB or librtc.[sl|so] respectively.

If the application does not link in the libpthread library (lipbpthread.so|libpthread.s1) or the libpthread tracer library (libpthread_tr_so.l|libpthread_tr.l ), the following error message appears:

Tracing version of libpthread is not loaded: set thread_check before starting the program, link your program with libpthread_tr or set LD_LIBRARY_PATH to contain /opt/langtools/wdb/lib/hpux32

HP-WDB Debugger does not display backtrace under Backtrace at the point of event occurrence for thread conditions detected at thread-exit, if there is an implicit call to pthread_exit().

The find command does not search the entire memory.

The HP WDB 6.0 find command does not search the pattern in the entire memory address range starting from 0x0 to 0xff..ff. It displays the following warning message:

(gdb) find 010,0xffffffffffffff,"hello"
warning: Unable to access target memory at 0x8, halting search.
Pattern not found.

Performance Degradation in Memory Debugging

For 64-bit applications on HP-UX 11i v2 and later on Itanium systems, heap check performance degrades when the string option is set in the set heap-check command:

Incorrect usage of . and -> while printing the values for C++ applications

WDB allows users to interchangeably use "." and "->" while printing values for C++ applications. WDB incorrectly allows users to use "." with a pointer variable and "->" with a reference variable while printing the value for C++ applications.

Mismatch Between dld.so Libraries Causes Issues

A mismatch in the dld.so libraries available in the core file and in the system might cause issues in unwinding beyond some frames within the libc.so library. The workaround is to use the packcore command to obtain the core file and the involved load modules into the system on which the debugging occurs.

WDB does not print downcast classes

WDB does not correctly print the downcast classes, when the set print object command is set.

Debugging an attached process that is not compiled for debugging may generate warnings

On attaching the debugger to a program that is not compiled for debugging, the process may stop in a system call and the following warning message about the various registers is displayed:

No data warning: reading 'r3' register: No data warning: reading 'r4' register: No data
warning: reading 'r5' register: No data warning: reading 'r6' register: No data warning

This warning occurs when executing the step command, the backtrace command, or when attempting to view the register information. To avoid this warning message, use the finish command to execute the process until the system call returns. This warning message is generated when the process stops at a system call, and the registers cannot be read by the debugger. When the debugger calls the routine ttrace, it returns this warning.

Error attaching WDB to a process that is traced by tools using ttrace

HP WDB cannot attach to a process that is traced by tools that use ttrace, such as Caliper, adb, and tusc.

The debugger displays the following error message while attempting to attach to such a process:

Attaching to process <pid> failed.
Hint: Check if this process is already being traced by another gdb or other ttrace tools like caliper and tusc.
Hint: Check whether program is on an NFS-mounted file-system. If so,you will need to mount the file system with the "nointr" option with mount(1) or make a local copy of the program to resolve this problem.

For discussion/questions:

hpux-development-public@lists.hp.com

To subscribe, send “sub hpux-development-public” as the subject line to sympa@lists.hp.com

For announcements:

hpux-development-public-announce@lists.hp.com

To subscribe, send “sub hpux-development-public-announce” as the subject line to sympa@lists.hp.com