Programming Forums

Programming Forums (http://www.programmingforums.org/forumindex.php)
-   C++ (http://www.programmingforums.org/forum15.html)
-   -   Code Segfaults when compiled with optimizations, doesn't in arbitrary situations (http://www.programmingforums.org/showthread.php?t=15154)

Jessehk Feb 9th, 2008 8:50 PM
Code Segfaults when compiled with optimizations, doesn't in arbitrary situations
 
I don't expect people to look through the complicated code that is the problem. I'm mostly looking for some debugging advice because I'm obviously doing something wrong, but at a loss of where to look.

When I compile my code with debugging on, there are no memory errors (checked with valgrind) and everything runs perfectly.

When I compile a "release" mode which includes -O3 and -DNDEBUG, the code segfaults almost immediately. However, when I insert a single output statement in one area of code,

ie,
:
  1. std::cout << "Fine" << std::endl;

and recompile (with the same optimizations), everything runs perfectly once again.
Does anyone recognize any symptoms here or have any advice on what class of errors might be causing this bizarre behavior?

Thanks in advance. :)

grumpy Feb 9th, 2008 9:03 PM
Re: Code Segfaults when compiled with optimizations, doesn't in arbitrary situations
 
There are many possible causes but, if I had to guess, you're probably falling off the end of an array and clobbering a variable that happens to be located at the end of that array. Some other code then comes along and uses that variable (eg as an index into another array) and dies horribly.

Valgrind is not guaranteed to find such things, and changing optimisation settings just changes what is getting clobbered by changing how data/code/etc are laid out in memory. An output statement (particularly one that doesn't output any variables) can appear to fix such problems, as it (depending on how the compiler works) has a small side-effect of changing how things are laid out in memory.

Jessehk Feb 9th, 2008 10:00 PM
Re: Code Segfaults when compiled with optimizations, doesn't in arbitrary situations
 
After about an hour of review, I couldn't find anything obvious. However, there was one spot in my code where I was using a boost::shared_ptr where a pointer of any kind wasn't necessary. I got rid of it, and the code runs fine. I'm not exactly sure why, but everything seems to be fine now. Thanks grumpy.

Jessehk Feb 10th, 2008 10:33 AM
Re: Code Segfaults when compiled with optimizations, doesn't in arbitrary situations
 
Just a bit of an update. It turns out that the fact that the code was working before was just a fluke.
It was just a lucky coincidence like before.

The real culprit was Boost.Foreach.
For some reason, something like this:
:
  1. const char LETTERS[] = { 'A', 'B' };
  2.  
  3. BOOST_FOREACH( char letter, LETTERS ) {
  4.     // ...
  5. }

was not iterating the correct number of times and that was the result of the segfault. I'm trying to figure out if it was a bug in my code (which ATM seems unlikely) or a bug with Boost.Foreach and arrays.

In any case, your suspicion that the code was falling off the end of an array was correct.

Sane Feb 10th, 2008 12:23 PM
Re: Code Segfaults when compiled with optimizations, doesn't in arbitrary situations
 
Boost Foreach needs to know how long the sequence is. Currently there's no way of telling where your LETTERS array ends.

It can iterate over C-style strings. However, your { 'A', 'B' } is not null-terminated. So the following should also work, since it now knows how large your array is:

:
const char LETTERS[] = { 'A', 'B', '\0' };

BOOST_FOREACH( char letter, LETTERS ) {
    // ...
}

This should also work:

:
std::string LETTERS( "AB" );

BOOST_FOREACH( char letter, LETTERS ) {
    // ...
}

Jessehk Feb 10th, 2008 12:30 PM
Re: Code Segfaults when compiled with optimizations, doesn't in arbitrary situations
 
I believe you're wrong, Sane. :)

:
  1. #include <iostream>
  2.  
  3. #include <boost/foreach.hpp>
  4.  
  5. int main() {
  6.     const int numbs[] = { 1, 2, 3, 4, 5 };
  7.  
  8.     BOOST_FOREACH ( int n, numbs ) {
  9.         std::cout << n << std::endl;
  10.     }
  11. }

:
$ valgrind ./example
==11578== Memcheck, a memory error detector.
==11578== Copyright (C) 2002-2007, and GNU GPL'd, by Julian Seward et al.
==11578== Using LibVEX rev 1804, a library for dynamic binary translation.
==11578== Copyright (C) 2004-2007, and GNU GPL'd, by OpenWorks LLP.
==11578== Using valgrind-3.3.0, a dynamic binary instrumentation framework.
==11578== Copyright (C) 2000-2007, and GNU GPL'd, by Julian Seward et al.
==11578== For more details, rerun with: -v
==11578==
1
2
3
4
5
==11578==
==11578== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 19 from 1)
==11578== malloc/free: in use at exit: 0 bytes in 0 blocks.
==11578== malloc/free: 0 allocs, 0 frees, 0 bytes allocated.
==11578== For counts of detected errors, rerun with: -v
==11578== All heap blocks were freed -- no leaks are possible.

It is able to do this was some automagic templates of some kind.

I am also unable to reproduce the original bug when I test the specific conditions in a seperate file and compilation. I'm guessing the bug (if it exists) only occurs when very specific conditions are true.

Sane Feb 10th, 2008 12:33 PM
Re: Code Segfaults when compiled with optimizations, doesn't in arbitrary situations
 
They say numbers are different. But for C-style strings, it has to know how long it is by the null-terminator.

http://www.boost.org/regression-logs...l/foreach.html

Jessehk Feb 10th, 2008 12:52 PM
Re: Code Segfaults when compiled with optimizations, doesn't in arbitrary situations
 
Oops! Thanks for the correction. I guess that explains it.


All times are GMT -5. The time now is 3:50 PM.

Powered by vBulletin® Version 3.7.0, Copyright ©2000 - 2008, Jelsoft Enterprises Ltd.
Copyright ©2007 DaniWeb® LLC