Re: [Module::Build] mb advice in CPAN READMEs

[Marvin H. <ma...@re...>]

On Feb 13, 2006, at 6:37 PM, Marvin Humphrey wrote:

> On Feb 13, 2006, at 5:29 PM, David Golden wrote:
>> So unless you're doing those things I'd say give 'traditional' a  
>> try. If it works for you, then it's just that much easier for  
>> others.  (And one less potential bug report to worry about.)
>
> 'traditional' doesn't work in this case, because traditional  
> doesn't know about c_source, or .xs files in the lib tree.
>
> It's not that I'm doing something with MB that can't be done with  
> EUMM, it's just easier dealing with MB.

Let me elaborate:

Lingua::Stem::Snowball is the Perl interface to the C version of the  
Snowball stemming library.  These days, that C library comes in it's  
own tarball, which unpacks nicely into a dir called libstemmer_c.   
I've rewritten Lingua::Stem::Snowball to use the most recent version  
of the library; the revised module is much faster, offers UTF-8  
compatibility, doesn't segfault any more, and a bunch of other good  
stuff.  As of this morning, it's official: I'm taking over  
maintenance responsibility.

Unfortunately, the Makefile that comes with libstemmer_c isn't  
portable.  That's not a problem, if I use MB -- all I have to do is  
specify 'c_source => "libstemmer_c"' in the MB constructor, and put  
'#include "libstemmer.h"' in Snowball.xs.  MB ignores libstemmer_c/ 
Makefile; everything gets compiled in place, and it all Just Works.

As far as I can tell, EUMM has no equivalent behavior.  So far, I've  
explored four ways of attacking the problem using EUMM.

The first possibility is to override MY::postamble as per the the  
EUMM docs (which use sdbm as an example)...

     sub MY::postamble {
         return <<'MAKE_FRAG';
     $(MYEXTLIB): sdbm/Makefile
             cd sdbm && $(MAKE) all

     MAKE_FRAG
     }

... but that won't work, because the stock libstemmer_c/Makefile  
isn't portable.

The second way of doing things is to follow perlxstut example #4,  
which is godawful voodoo <http://perldoc.perl.org/ 
perlxstut.html#EXAMPLE-4>.  It involves writing a libstemmer_c/ 
Makefile.PL to generate a libstemmer_c/Makefile -- essentially, I  
have to duplicate the functionality of the existing libstemmer_c/ 
Makefile, rewriting it through a layer of indirection using EUMM's  
convoluted interface.  This Makefile.PL has to do things like  
override MY::top_targets (huh?), which spec Make targets like  
"pure_all" (huh?), "static" (huh?)...

The third way of doing things is used by the current  
Lingua::Stem::Snowball Makefile.PL, which was written a long time ago  
by people no longer involved with the project: use File::Find to grab  
all the C files and create a list of object files, supplying those to  
OBJECT.

Unfortunately, MakeMaker dumps all .o files into the current  
directory, and since libstemmer_c's C files are divided up into  
multiple directories, they contain stuff like...

    #include "../src_c/stem_UTF_8_english.h"

... which can't be found because of where MakeMaker puts the .o  
file.  So... it looks like what the original authors did was write a  
config.c file that #included every .h file, then tailored all the XS  
functions to use only the low-level API.  They essentially rewrote  
libstemmer.h/libstemmer.c from scratch.  Hopefully they didn't do  
this just because of MakeMaker, but perhaps because libstemmer.h  
wasn't available in its present form at the time.  Needless to say, I  
don't want to carry forward that tradition.

The fourth way was suggested by Tye McQueen (tye on PerlMonks, TYEMQ  
on CPAN).  He offered his Makefile.PL for Win32API::File as an  
example of how I might customize a Makefile.PL to force the compile  
commands into a desired form: <http://search.cpan.org/src/TYEMQ/ 
Win32API-File-0.09/Makefile.PL>. I stared at it a while stupidly,  
then gave up.

There's gotta be a way to do this with MakeMaker.  Maybe a way to add  
"-o $fullpath" to each compile command-string.  But here's where the  
difficulty of customizing MakeMaker comes in.  It's taken me a  
looonnnngggg time to even get to this point.  I've been through the  
EUMM/perlxs/perlxstut/etc docs time and again, and I think I've  
finally figured out what it would take to make that danged -o param  
appear: I have to override MY::c_o and perform a regex substitution  
on the value it normally returns.  That strikes me as kludgy and  
brittle.

Then there's MB....... which... just... works......

Marvin Humphrey
Rectangular Research
http://www.rectangular.com/