Re: [boost] [EXTERNAL] [1.61.0] Master branch is closed

Subject: Re: [boost] [EXTERNAL] [1.61.0] Master branch is closed
From: Belcourt, Kenneth (kbelco_at_[hidden])
Date: 2016-04-23 19:29:34

> On Apr 23, 2016, at 1:31 PM, Robert Ramey <ramey_at_[hidden]> wrote:
> 
> On 4/23/16 12:02 PM, Belcourt, Kenneth wrote:
>> 
>> Hi Robert,
>> 
>> I noticed that all the Boost.MPI and graph parallel tests that use serialization are broken on develop and master with an error similar to this:
>> 
>> In file included from libs/graph_parallel/src/mpi_process_group.cpp:14:
>> In file included from ./boost/graph/distributed/mpi_process_group.hpp:30:
>> In file included from ./boost/mpi.hpp:32:
>> In file included from ./boost/mpi/skeleton_and_content.hpp:32:
>> ./boost/mpi/detail/ignore_iprimitive.hpp:40:21: error: no template named 'array' in namespace 'boost::serialization'; did you mean simply 'array'?
>>    void load_array(serialization::array<T> &, unsigned int )
>>                    ^~~~~~~~~~~~~~~~~~~~
>>                    array
>> ./boost/array.hpp:61:11: note: 'array' declared here
>>    class array {
>>          ^
>> fatal error: too many errors emitted, stopping now [-ferror-limit=]
>> 1 warning and 20 errors generated.
>> 
>>    "g++"  -ftemplate-depth-128 -O3 -Wall -gdwarf-2 -fexceptions -Wno-inline -arch x86_64  -DBOOST_ALL_NO_LIB=1 -DBOOST_GRAPH_NO_LIB=1 -DNDEBUG  -I"." -I"/Users/kbelco/Projects/local/openmpi-1.8.4/include" -I"libs/graph_parallel/src" -c -o "bin.v2/libs/graph_parallel/build/darwin-4.2.1/release/link-static/threading-multi/mpi_process_group.o" "libs/graph_parallel/src/mpi_process_group.cpp”
>> 
>> It’d be nice if serialization worked with MPI for the release, though I’m not sure it’s a show stopper.
> 
> This is unfortunate.  It seems that no one is watching the boost mpi tests related to serialization.

Guilty.

>  Does Boost MPI have a current maintainer?  Does it need one?

Well, perhaps CMT in the future but, for now, I’ll try to bring it up to date.

> 
> I'll take a look at this when we get 1.61 out the door

Thanks, ping me if I can help test or make changes.

> It turns out that writing a custom archive is not as easy as it's cracked up to be.  I think that there's some coupling between the interface and implementation which means that once in a while I've made changes somewhere which broke existing archives.  That is, the archive API is not rigorously defined.  As it happens, this illustrates another drum I've been beating on lately.  The one which says waiting until the library is done to write the documentation makes for an ill designed library.  But that's another thread.

Sounds good.

Noel