RPSL automation

Nick Hilliard nick at inex.ie
Mon Jun 8 00:05:00 UTC 2009


On 22/05/2009 19:13, Nick Hilliard wrote:
> I've committed another monster patch on the cruft-cleanout branch as
> r226, leaving 48.8kloc. At this stage, about 57% of the codebase has
> been removed with no loss of functionality.
>
> The code (i.e. peval, RtConfig, rpslcheck and librpsl.a) now compiles
> and runs successfully on linux (ubuntu 9.04)/x86, freebsd 7.2/amd64 and
> MacOSX 10.5.7. If anyone has the time / inclination, maybe you could try
> compiling it on your favourite system to see if it compiles / works. In
> particular, I haven't checked to see if it compiles or runs on solaris.

Some more updates on this branch (currently at r247):

- codebase now stands at 41.2kloc

- lots more unused files have been deleted, and a number of unreferenced 
class members removed from various places

- a couple of files and some code have been moved from one place to another 
in the repository, where it made sense to do so

- the code now compiles without warnings on GCC4.2 and GCC4.3 (r247)

- wiki page added (http://irrtoolset.isc.org/wiki/CruftCleanout) which 
includes a relatively complete XCode 3.1 project file for Mac users (you'll 
still need to run autoconf and configure manually).

Currently, the code compiles and runs on mac os x and ubuntu.  FreeBSD and 
Solaris are not working just now; both are having yylineno problems, and 
Solaris does not like strsep() (added in r229, but will probably replace 
this with strtok()).

Next on the agenda is to do something sensible with the directory 
structure.  As it stands, libCore has been gutted (25kloc to 2400), so it 
makes no sense for this to be a standalone library any longer, particularly 
given that it requires its own (badly implemented) autoconf setup.

librpsl depends on some of the remaining code in libCore, so it would make 
sense to amalgamate these two libs into one.

The remaining libs (in gnug++, irrutil, irr, normalform and re2dfa) should 
be amalgamated into a single library (libirrtoolset anyone?), which will be 
dependent on librpsl.  This will cause lots of files to move around the 
place, but i think it would be worth it.  Besides, this is svn, not cvs.

I can't see a whole load of reason to have three separate but very similar 
configure scripts, two of which have some really silly bugs.  Would anyone 
object if I reduce these to a single configure script?  This will allow 
freebsd and solaris to be fixed properly by putting in proper lex / bison 
configuration detection.

Nick



More information about the irrtoolset mailing list