As far as I can tell from the svn diff ( http://cctbx.svn.sourceforge.net/viewvc/cctbx?view=revision&revision=15762) the only difference that passing --enable-boost-threads made was to include the "-pthread" flag, and when not enabled the 'BOOST_DISABLE_THREADS' option was passed instead to the compiler. I do not know if passing the 'BOOST_DISABLE_THREADS' option could cause a conflict when linking to a program compiled with the -pthread option. Neither of these options are passed to the compiler in the current cctbx. Boost.Threads support was removed from the cctbx because it was only experimental, and was superseded by better alternatives. Do you still observe the same problem if you use a more recent version of the cctbx, either svn or one of the nightly builds (http://cci.lbl.gov/builds/ - there isn't really much distinction between "released" and "unreleased" builds)? Cheers, Richard On 17 October 2012 16:56, Jeffrey Van Voorst wrote:

My question is of a slightly different nature. Specifically, what type of "magic" is enabled by --boost-enable-threads in the latest release of cctbx that allows the libiotbx_pdb.so library to be linked correctly to a simple program when that linking includes the "-pthread" flag? By simple, I mean a program that has two lines in the main function -- reading the PDB file and constructing the hierarchy.

My goal is to try to understand the dependencies and why the simple program would have a memory corruption when libiotbx_pdb.so was not compiled using --boost-enable-threads.

--Jeff

On 10/17/12 5:28 PM, Richard Gildea wrote:

...

Hi Jeff,

Boost.Threads support in the cctbx was removed back in August ( http://phenix-online.org/**pipermail/cctbxbb/2012-August/**000468.htmlhttp://phenix-online.org/pipermail/cctbxbb/2012-August/000468.html, http://cctbx.svn.sourceforge.**net/viewvc/cctbx?view=** revision&revision=15762http://cctbx.svn.sourceforge.net/viewvc/cctbx?view=revision&revision=15762). If you are using a recent copy of the sources then the libtbx configure command should raise an error if you pass the now defunct --enable-boost-threads option.

Cheers,

Richard

______________________________**_________________ cctbxbb mailing list [email protected] http://phenix-online.org/**mailman/listinfo/cctbxbbhttp://phenix-online.org/mailman/listinfo/cctbxbb

Actually, Richard pointed out to me privately that as part of removing the support of Boost.Thread, I also remove the option -DBOOST_DISABLE_THREADS. Without it, some part of Boost supports threading, whatever that means. I did that on purpose, just to see if something would break, although I forgot to mention that in the commit message. It would again be very useful to know the full stacktrace to see whether that could be the problem. Luc

On Oct 18, 2012, at 11:21 AM, Jeffrey Van Voorst wrote:

Ok, let me look into this a bit tomorrow. The issue is free() which gets an invalid address from a shared_ptr for root_data in iotbx_pdb when it is time for the whole pdb hierarchy to disappear. My main issue is I am not sure how to go about debugging shared_ptrs nor what architecture dependent flags or other voodoo (if any) is specified by -pthread.

Could you try adding -DBOOST_DISABLE_THREADS to compile your C++ program? Luc

Hi Jeffrey,

My assumption is: if I compile cctbx with -DBOOST_DISABLE_THREADS and link to the libraries from programs that were not compiled with that flag, I will run into issues of having differing implementations of smart_ptrs.

That was my assumption too indeed.

Am I correct in inferring that the latest SVN version of cctbx will configure (by default) to not use -pthreads and not use -DBOOST_DISABLE_THREADS?

Yes, indeed.

I would like to have just one copy of cctbx built, and be able to use the libiotbx_pdb.so library in both single and multithreaded code. [...] However, for the purposes of my package it is easiest for me to compile cctbx once.

Imho you should be able to achieve that by compiling the both of your program and the cctbx without BOOST_DISABLE_THREADS. The only influence of that flags is whether to disallow using atomic increment and decrement of the reference count used by shared ptr. Atomic in the sense of thread-safe, just to be crystal clear. Iiuc, without -D BOOST_DISABLE_THREADS (unless you define some other macros but I am sure you don't) for Intel processors, shared ptr should end up using increment and decrement written in Intel assembly language that use the Intel specific instruction "lock" to make those operations atomic. That code works however many threads are involved, the only issue being the performance degradation compared to non-atomic increments or decrements. But in practice, for what we do, this should be anecdotical.

If you could point me to a good SCons tutorial or a good example script that has documentation, that would be very useful.

Our SConscript's are the stuff of nightmares, agreed! I am not sure a generic SCons tutorial would be an useful considering how idiosyncratic our SConscript's are. There are two places to start, that you may already know, the SCons documentation and the SCons manpage. Eventually I reckon you just ask around on this forum every time you scratch your head for more than a few minutes! There are more than a handful of SCons expert split in 2 time zones, thus providing nearly a 24-hour service ;-) Best wishes, Luc

It looks like the change Luc checked in yesterday regarding the determination of the boost version has partially broken a build from scratch. If you type the command: libtbx.refresh before issuing make then I think it should work. Cheers, Richard On 18 October 2012 10:01, Jeffrey Van Voorst wrote:

I am trying to use the set of commands to get cctbx from SVN ( http://cctbx.sourceforge.net/**current/installation.html#** using-the-cctbx-svn-repositoryhttp://cctbx.sourceforge.net/current/installation.html#using-the-cctbx-svn-r... **). However, either I am unable to follow the commands or something is not configured properly.

I did the following steps: svn export https://cctbx.svn.sourceforge.**net/svnroot/cctbx/trunk/** libtbx/development/cctbx_svn_**getting_started.cshhttps://cctbx.svn.sourceforge.net/svnroot/cctbx/trunk/libtbx/development/cct... /bin/csh cctbx_svn_getting_started.csh mkdir cctbx_build mv sources cctbx_sources cd cctbx_build python ../cctbx_sources/cctbx_**project/libtbx/configure.py iotbx source setpaths.sh make

Make fails almost immediately. I have listed the output from configure and the error from the make command on a gist (https://gist.github.com/** 3912960 https://gist.github.com/3912960).

I just checked: if I configure using mmtbx instead of iotbx then building the package works.

--Jeff

______________________________**_________________ cctbxbb mailing list [email protected] http://phenix-online.org/**mailman/listinfo/cctbxbbhttp://phenix-online.org/mailman/listinfo/cctbxbb

Ok, this is fixed sufficiently well on my end. Thanks for your help and comments. --Jeff