Pietro Abate homepage

The other day I wrote about my experience to set up the build system for a simple library. However, since parmap includes only two simple stubs to syscall I didn't have the chance to talk how to convince ocamlbuild to build stubs that depend on an external dynamic library.

I'm sure, facing this problem you had a look at this example : http://brion.inria.fr/gallium/index.php/Ocamlbuild_example_with_C_stubs

Despite providing all elements, the page is a bit scars of details and explanations (at least for me...).

So, suppose you want to build ocaml stubs for a C library called toto . Good practices tell us to put our stubs in a file that is called toto_stubs.c and then add the .o file in a clib file (libtoto_stubs.clib ) that ocamlbuild is going to use to build out dynamic library.

So far so good. Now we need to add a tag, say use_toto that we will use to compile our binaries and libraries. Our _tags file will look like :

 <libtoto_stubs.*>: use_toto

Here I use only one tag. In the example of the ocamlbuild they use two tags, one to compile, one to link.

At this point we need to explain to ocamlbuild how to build our library. First we add a compile rule where we say that whenever we compile a c object that use toto, then we must also add its include directory.

Second we add a link flag to add to the dynamic library all the information it needs to load other external libraries. This is important as we don't want to add any other flags anywhere else. When we use -ltoto_stubs we want all other dependencies automatically satisfied by the linker. Note the libtoto.so referred by -ltoto is the C library for which we are writing these bindings and that sits in /usr/lib/.

At the end we add a flag that whenever we try to build out ocaml module ( say toto.cma ), we want to add all necessary information to load at run time its dynamic component.

Using ocamlobjdump we can easily check if the cma contains indeed this information. The output should look something like this :

In order to generate cmxa and a objects we need to specify few other flags and dependencies like :

As always, if I'm missing something, please drop a note in the comment page.

Today I struggled once again to build a simple library with ocamlbuild, so once for all I decided to write something about it. I'm sure next time, googling for an answer I'll find this post and shake my head in despair :)

The library in question is parmap written by Roberto Di Cosmo to speed up computations on modern multi processors computers. We want to build everything: cma, cmxa and cmxs. Moreover, we want to build a shared library that contains stubs for a couple of bindings to C functions. On top of that, I want to use configure to make sure that the platform I'm using supports a specific syscall.

We start by copying ocaml.m4, the more or less standard autoconf macros for OCaml ( source here ). I prefer to copy this file in my source tree as I don't want to impose to download it from a website to compile my library. This file ends up in the directory m4. Then to use it, I need to invoke aclocal as follows aclocal -I m4. This will take care of making the m4 macros known to the configure script.

Next step is the configure script. I need check for the standard ocaml utils, if extlib is known by ocamlfind and if the function sched_setaffinity is available on the system. This is the standard autoconf way to detect if a specific function is available. Together with the AC_CONFIG_HEADERS([config.h]) call, this will define the variable HAVE_DECL_SCHED_SETAFFINITY to 1 in the file config.h that can be later used in the C source code.

Once this is done I've to generate the config .h.in with autoreconf and then run autoconf to generate my configure script. If this go through, you should be able to run your configure script as usual. So far so good. Now it time to convince ocamlbuild to play nice with us.

First we need to write a small myocamlbuild file to set few dependencies and compilation flags. The first rule we add is a dependency to convince ocamlbuild to copy the config.h file in the _build directory. Without this rule, ocamlbuild is unable to figure out the dependency just by looking at the C source file (I've found a [bug http://caml.inria.fr/mantis/view.php?id=5107] on the inria bts about this problem).

The second rule adds a couple of compilations options to compile our C code. I think this can be done on a file basis, but since these options make no harm, I use to compile all my C objects.

The third rule is there to make aware ocamlbuild that is I want to compile a library that uses libparmap, I must specify that libparmap is a dynamically linked library. The forth rule is an analogues rule for native libraries.

The fifth and sixth rule are one to compile the .a and then to link it. This is need to correctly generate cmxs objects.

But of course this is only part of the ocamlbuild configuration. We also need to specify how to build these libraries and what to include. To build the stubs library we specify a .clib file in which we tell to ocamlbuild the C objects it has to link together.

This is accomplish with by adding the file libparmap_stubs.clib :

To build parmap.cm{x,}a and parmap.cmxs I need to other files, respectively :

and at last we write the _tags file to correctly associate different flags to each component.

This should build all you goodies in one go :

Relevant files:

• http://gitorious.org/parmap/parmap/blobs/pipes/myocamlbuild.ml
• http://gitorious.org/parmap/parmap/blobs/pipes/configure.ac
• http://gitorious.org/parmap/parmap/blobs/pipes/Makefile.in
• http://gitorious.org/parmap/parmap/blobs/pipes/_tags

All the source code is in Git parmap - in the pipes branch. If I'm doing something wrong, or not in the standard way, please tell !!!

The results of the MISC competition 2011 were announced during the Workshop LoCoCo in Perugia Monday 12th of September.

The winners are :

• for the track paranoid and trendy the solver aspuncud, a solver based on an experimental version of clasp with iterative unsatisfiable core elimination by Oliver Matheis.
• and for the track user the solver gj-user-solver, a refactored SAT4J with a custom Pseudo Boolean Optimisation algorithm by Graham Jenson.

The novelty of this year was the introduction of a new aggregation function to the user track that can be used to maximize or minimize the numerical value or a cudf property.

All competing solvers were able to provide an answer the large majority of problems. Performances were used to break a few ties this year, and we can clearly see that solvers are getting faster and faster. Now most solvers are able to answer a 'common' installation problem under 3 seconds (see for example the solver aspuncud in the sub category dudf-real of the paranoid track) providing at the same time an optimal solution w.r.t. the chosen optimization criteria.

We received a positive feedback from all participants despite a small glitch in our ranking function. We definitely looking forward to establish MISC as an international solver competition for the next few years.

I've just released the new version of ocaml-buddy, my ocaml bindings to the buddy BDD c library. Thanks to a fruitful interaction with Jimmy Thomson, I've fixed a couple of memory leaks and cleaned up the code a bit.

grab it when is still hot : https://github.com/abate/ocaml-buddy/tree/0.5

comments a testers are welcome. A debian package is on the way