[vox-tech] a "Bus Error" C++ issue, platform, compiler or STL?

[Bill Broadley]

On 08/24/2010 07:17 PM, Hai Yi wrote:
> Thanks Jeff and Bill.
> Jeff, how do you know that the value,0xba7bfff, is corrupted? Because
> it's only 3.5 bytes? Could it be possible a value with zero stripped?
> Also, what's the mapping of the argument list b/w the code and that in
> the stack?

You are the one with the strange compiler and over 10 year old OS.  I'd 
suggest debugging a known working code and check out the mapping 
yourself.  Debug a working program, look at the stack before and faster 
function calls, track the function parameters, dereference some pointers.

Bus errors, segmentation faults, core dumps and the like are usually 
memory errors, I'd check all of the following:
* Return values, especially on any allocation
* Make sure memory is free once, and exactly once.  (double frees are
   common)
* Make sure you are using STL allocation and resize functions correctly,
   at least some of the time they depend on the user doing the memory
   allocation.
* Learn to recognize pointers, look at a few 1000, you didn't mention
   if the OS was 64 bit or 32 bit.

Keep in mind corruptions can propagate, memory protection is on the page 
boundaries.  So you can have quite a few corruptions before you cross a 
boundary.  You might have trashed the stack and not notice
till later.

Again I'd suggest running it in a different environment, try a newer g++ 
with -Wall -pedantic it might well just point to something that you are 
doing is illegal, ill advised, or undefined.




>
> Thanks again!
> Hai
>
>
> On Tue, Aug 24, 2010 at 7:05 AM, Hai Yi<yihai2004 at gmail.com>  wrote:
>> ---------- Forwarded message ----------
>> From: Bill Broadley<bill at broadley.org>
>> Date: Tue, Aug 24, 2010 at 3:44 AM
>> Subject: Re: [vox-tech] a "Bus Error" C++ issue, platform, compiler or STL?
>> To: vox-tech at lists.lugod.org
>>
>>
>> On 08/23/2010 05:43 PM, Hai Yi wrote:
>>> Hello All:
>>>
>>> I'm on a C++ project and have been strugling with an "Bus Error" issue
>>> for a couple of weeks. I used dbx to debug the code and here is the
>>> stack:
>>> ...
>>> ficc_trade_leg::set_trailer1(const string&    arg)
>>> df_string_tuple::df_string_tuple(0x1ae5180, 0xffbea53c, 0xfd578ac0,
>>> 0x0, 0x18, 0x0)
>>> basic_string,allocator>::basic_string,allocator>(0x1ae518c,
>>> 0xfd578ac0, 0xffbea540, 0xff3de7a8, 0x0, 0x5)
>>> string_ref,allocator>::references(0xba7bfff, 0x0, 0x8, 0xfb9c1e54, 0x0, 0x0)
>>>
>>>
>>> set_trailer1 is creating an instance of df_string_tuple, whose
>>> constructor looks like
>>>
>>> df_string_tuple::df_string_tuple(const df_tag&    Tag, const string&
>>> Value): df_tuple(Tag), Value_(Value)
>>> {}
>>>
>>> It seems to me that debugger nailed down to the 2nd parameter of the
>>> above constructor. Isn't it more likely an issue of Solaris patch, C++
>>> compiler or STL lib rather than a coding issue?
>>
>> Most of the bugs I've seen with STL code have been of the variety of
>> assuming that when you resize something, say a vector, that the
>> programmer assumes STL is doing the memory allocation for the new array,
>> and the STL lib of course is assuming the programmer is doing the memory
>> allocation.  Resulting in usually a segmentation fault.  It's been a
>> very long time since I've tried anything on SunOS or the embarrassing
>> excuse of a compiler that sun ships.  Most of the open source I played
>> with back when 5.8 was somewhat new said something along the lines of:
>> * Do not use solaris tar it's buggy
>> * Do not use solaris make it's buggy
>> * Do not use solaris compiler it's buggy (here's binaries to bootstrap
>>    gcc).
>>
>> Not sure if the solaris bus error is the equivalent of the linux segfault.
>>
>>> My platform is SunOS 5.8, not sure about the STL lib or CC compiler.
>>
>> If possible I'd recommend trying the same code in a standard linux
>> environment.  The newer gcc compilers seem to have improvement C++ error
>> messages quite a bit.
>>
>> I suspect if you asked someone could set you up with a linux account if
>> need be.
>> _______________________________________________
>> vox-tech mailing list
>> vox-tech at lists.lugod.org
>> http://lists.lugod.org/mailman/listinfo/vox-tech
>>
> _______________________________________________
> vox-tech mailing list
> vox-tech at lists.lugod.org
> http://lists.lugod.org/mailman/listinfo/vox-tech