[user] segfault w/ error 4

[U.Mutlu]

U.Mutlu wrote, On 09/11/2012 07:52 AM:
> Ian Lance Taylor wrote, On 09/11/2012 07:34 AM:
>> On Mon, Sep 10, 2012 at 10:06 PM, U.Mutlu <for-gmane@mutluit.com> wrote:
>>
>>> recently a server daemon program suddenly started repeatedly crashing.
>>> syslog says:
>>>   segfault at 7f6d003d5453 ip 00007f6d49c9dd6d sp 00007f6d469c1b60 error 4 in
>>> libstdc++.so.6.0.17[7f6d49bff000+e8000]
>>>
>>> What does "error 4" mean? Is this documented somewhere?
>>>
>>> Is this an application error or rather an error in libstdc++.so ?
>>>
>>> The error is not always repeatable, ie. happens occasionally, reason unknown
>>> yet.
>>>
>>> What's the easiest method to locate the bug in this server daemon pgm?
>>
>> That message does not come from libstdc++.  You neglected to mention
>> anything about the system, but that looks like the log from an x86
>> Linux kernel when a program crashes.  For the Linux kernel, error 4
>> means that this is a user-mode access.
>>
>> You can get this kind of thing if you pass a NULL pointer to a
>> libstdc++ routine that is not expecting one, so this may be a bug in
>> your program.  Or it may be a bug in libstdc++.

There was in a function a string comparisions whereby a const std::string
was compared against a const ptr to char ("if (s == psz) ...").
Not sure if that could be the reason (IMHO yes), as it indeed looked ugly & dangerous
I now have replaced the char ptr by an ordinary std::string.
Will wait & observe what happens... :-)

> System is debian 7 (wheezy) w/ latest updates from the debian repository.
> The error happens only with the 64-bit version of the system library on multicore systems (amd64).
> On 32bit-multicore-systems the error doesn't happen. That's the weird thing.
> Any suggestions for locating and fixing the bug are welcome as the program crashes every half hour or so :-(