Marpa's Sudden Portability Adventure
I've made bold claims for the portability of Marpa. [ Marpa is a general BNF parser generator -- it parses from any grammar that you can write in BNF. If the grammar is of one of the kinds currently in practical use (yacc, LR(k), LALR, LL, recursive descent, etc.), this parsing is in linear time. ]
The boldness of my claims evinced no ambition to test them under fire. But, when my main development box (a 2-year Dell laptop running Ubuntu) suddenly died, my only other choice for a development platform was a MacBook G4 running Mac OS Tiger. So test my portability claims is what I had to do.
Not that my brags about portability had been without foundation. I upload development versions for cpantesters regularly. (By the way, to all of you at cpantesters: a big "thank you".) As far as the configuration, build and runtime environments went, I had a lot of reason to be confident.
The problem is that I am now working on the XS version of Marpa. When I was working on the Pure Perl version of Marpa, my maintenance and development environments were not separate from the configuration, build and runtime environments. Development was also in Pure Perl, with no additional tools required. But for the XS version of Marpa, maintenance and development require GNU autotools, as well as Don Knuth's TeX and Cweb.
Let me emphasize that you can, now and in the future, configure, build and run Marpa::XS on boxes without GNU autotools or TeX. It's just that you can't properly work on or patch the source code without them. Also, it is worth emphasizing that, in the open source spirit, Marpa::XS ships with all the files necessary for maintainance and development. So you always have the "source" in the fullest sense, and all the tools you need to develop are available free with source. But part of making Marpa::XS as portable as possible is ensuring that it will configure, build and run happily in environments which don't have all the development tools it requires.
Which brings me back to my crisis. I could be confident about configuring, building and running my development version of Marpa::XS on a wide variety of platforms. The cpantesters had proved this for me. Less certain was whether my development environment could be recreated.
The sudden switch involved a sudden change of chips (Intel to PPC), and operating systems (the latest Ubuntu to a Mac OS that is two major releases out of date). More dramatic changes are possible (both are POSIX systems), but for a port suddenly forced on me, this was more than change enough, thank you.
I hoped everything I needed was on MacPorts. That was assuming that vintage 2005 systems, running a Mac OS version two major releases obsolete, were a priority with the helpful folks at MacPorts.
In the event, I was back developing with 24 hours. The only change I had to make to my codebase was to three lines in Build.PL, which specified module versions. Most of the 24 hours was taken up with compiles. The Macbook G4 was never considered a fire-breather, even back in 2005 when it was new.
I blog about Perl, with a focus on parsing and Marpa, my parsing algorithm based on Jay Earley's.
I suppose you still wouldn't support perl 5.8?
@Vyacheslav: Sorry. I think I understand the concern. I've worked in production environments where installing new software anywhere on a box is just simply out of the question. If the machine is mission critical, and it shipped with Perl 5.8, then 5.8 is all there is going to be.
But I've no plans to backport Marpa to 5.8. My reasoning is this. With perlbrew and local::lib, most environments can bring in Perl 5.10. There may be good reasons why perl 5.10 is out of the question. But exactly those same reasons will usually mean Marpa (which is alpha) is also not a good idea.