Intel® Software Network

When trying to configure Eclipse (3.3.2, CDT 4.0.3) to use idb (intel64) version 10.1.35, I get a segmentation fault. Tracing what happens, it seems the Eclipse plugin for idb (that I copied from the ia64 distribution) starts up idb with '-gdb -i mi' and initiates the session with the following two gdb/mi commands:

(gdb)
-gdb-set confirm off
^done
(gdb)
-gdb-set width 0
~"\n"
~"Intel(R) Debugger for applications running on Intel(R) 64, Version 10.1-35  caught signal \"Segmentation fault\" (11).\n"
~"This is an unexpected condition and may indicate the presence of a defect.\n"
~"If you wish to report this, please include the stack trace that follows.\n"
~"/lib/libc.so.6 [0x2b3be0fba7d0]\n"
~"/usr/lib/libstdc++.so.5(_ZNSi4syncEv+0x57) [0x2b3be0d0b097]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_ZN9FilterBuf4syncEv+0x5a) [0xca4b26]\n"
~"/usr/lib/libstdc++.so.5(_ZNSi4syncEv+0x5a) [0x2b3be0d0b09a]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_ZN9FilterBuf4syncEv+0x5a) [0xca4b26]\n"
~"/usr/lib/libstdc++.so.5(_ZNSi4syncEv+0x5a) [0x2b3be0d0b09a]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_ZN9FilterBuf4syncEv+0x5a) [0xca4b26]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_ZN10FilterBaseD2Ev+0x63) [0xca6ccf]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_ZN11FilteredIOSD0Ev+0xf3) [0xca6ea3]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_ZThn208_N11FilteredIOSD0Ev+0xd) [0xca704f]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e [0xc8a5c3]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_Z23parseStringAsExpressionPKcb+0x5d) [0xc8a6fb]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_Z10MiCXXparsev+0x3bc0) [0xda409e]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_ZN7Session19parseCommandFromCinENS_4ModeE+0x14f) [0xc7b0a3]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e [0xcafbf0]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_Z15ProcessCommandsv+0x47) [0xcaf46f]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_Z7idbMainiPPKcS1_+0x159) [0xcaee29]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(main+0x3c) [0xb87a98]\n"
~"/lib/libc.so.6(__libc_start_main+0xf4) [0x2b3be0fa6b44]\n"
~"/opt/intel/idbe/10.1.015/bin/idb-e(_ZStlsISt11char_traitsIcEERSt13basic_ostreamIcT_ES5_c+0x5a) [0x7ba02a]\n"

When I run idb on the command line, I can reproduce the error. It seems the MI interface is not correctly implemented.

Your sample shows an gdb prompt while using idb it should show the (idb) prompt. How did you call idb? How did you build the application? Wich gcc and/or Intel Compiler icc? What is the OS?

Thanks&regards,

Hubert.

The debuggee is built with icc and ifortran (both v 10.1) with '-debug full' as option, under Ubuntu 7.10 amd64 with an Intel processor.

Well, the manner in which I call idb is written in my post, but here it is again:

$ idb -gdb -i mi <executable>

idb is clearly running, I can see it in the list of processes.

This starts idb in gdb and MI interpreter compatible mode and the prompt says (gdb), I thought the idea is that IDEs with debugger parsers should be 'fooled' by the prompt and accept the output from idb as if it was from gdb.

The crash report is clearly from the Intel debugger...

If I start it without debuggee, I get this:

$ idb -gdb -i mi
~"Intel(R) Debugger for applications running on Intel(R) 64, Version 10.1-35 , Build 20080310\n"
(gdb)

Your answer to my post (statement that idb should always have (idb) as prompt) suggests that you have never run or tested idb with these options, if I may counter attack...

Other MI commands work, but the command reported here causes a segmentation fault.

However, this may well be related to my OS distribution. I've tried it on an older CentOS (Redhat) and it seems to work, so I guess the answer is that when it is written "unsupported", then it really is. Pitty though, I think there are a few other Ubuntu users out there...

Thanks for the follow-up.

I checked with one of our debugger architects and we were able to confirm a few things.

First, sorry for the misstatement on the (idb) vs. (gdb) prompt. You are absolutely correct that the MI interpreter mode is one of the few modes in which IDB does not show it's typical prompt. Sorry for the confusion.

IDB 10.1.035 does as you observed not officially support Ubuntu 7.10. We were able to do some basic internal installation smoke testing, but the validation cut-off date was too early to allow full validation with Ubuntu 7.10.

We were able to reproduce yor problem on some older IDB builds and we were also able to briefly verify that it doesn't seem to show up any more in our latest internal pre-beta build for the next major release.

We will do some broader testing. The development team asked for the exact debugger build version and I will forward the banner information "~"Intel(R) Debugger for applications running on Intel(R) 64, Version 10.1-35 , Build 20080310\n"
you provided to them.

I am afraid I can't promise a quick fix or workaround at this time, but we will do our best to isolate this problem with Ubuntu 7.10 and try to find such workaround.

Thanks, Rob

  

P.S. : I may end up having additional questions for you since we can currently not reproduce with the 10.1.035 IDB builds we tried. Now that we have the exact build number we may be lucky though.

 

Could I ask you to provide the exact version info with build number for the C++ and Fortran compilers you use as well. In case we have a dependency this might be important.

Thanks.

Another update:

On x86_64 Ubuntu 7.10 with the 10.1.035 IDB release we can reproduce ths issue. We can trace it to some of the library functions for C++ streams. This problem does currently not occur with other Linux distributions. It looks as though there is an issue with the specific C++ libraries Ubuntu 7.10 uses. Whether it is a bug specific to one package in the Ubuntu distribution or whether it is a feature/behavior change in this library functions in general that causes the problem is unclear at the moment.

We will keep investigating.

 

I have not been able to check this page for a while, thanks for following up. You are correct, I'm running the 64-bit version of Ubuntu 7.10, but I plan to upgrade to 8.04 shortly.

Here is the build info:
Intel(R) C Compiler for applications running on Intel(R) 64, Version 10.1    Build 20080312 Package ID: l_cc_p_10.1.015
Intel(R) Fortran Compiler for applications running on Intel(R) 64, Version 10.1    Build 20080312 Package ID: l_fc_p_10.1.015

In any case, we found that there were other functions missing in idb that works fine in gdb. In particular we have trouble getting idb to print variables that are passed to fortran sub-routines.
File 1:
call foo(TooFoo)
...
File 2:
subroutine foo(footoo)

idb just returns a blank when doing 'print footoo' or 'print TooFoo' when debugger prompt is inside subroutine foo; whereas gdb at least returns an address, and then preceding the variable name with &, we get the value.