Re: whither specs?

[Theodore Papadopoulo <Theodore dot Papadopoulo at sophia dot inria dot fr>]

rth@redhat.com said:
>> The driver should be doing this itself, IMO.
> There are cons to DT_RPATH that in my opinion outweigh any benefit.
> This has been dicussed before; search the archives. 

Right, I remember this discussion. But yet it seems to really be 
useful for some people all the more that since all the gcc librairies 
are dynamic by default, if you want to test multiple compilers, you 
have to specify each time not only a path but also a LD_LIBRARY_PATH 
or LD_RUN_PATH.

Why can't we provide this feature optionnally with a --enable-xxx parameter a 
configure time (by default off). This would allow:

	- for the people that would like to have the feature to get 
          it, and it is an often demanded feature. See how many people
          achieve it by hacking the spec file.

        - for some (large?) scale testing of the pros & cons of having
          this feature.

I do not question your reasons, but as far as my remembering goes, the cons 
were about changing the paths for the libraries after a while. Sometimes people
know they will not do that. At the very least, it looks like some of us are 
doing so without finding the problems that had been raised at the 
time... I do not know how representative we are, if this works better 
on some systems than on some other but I suspect a 
few people would give
it a try. So maybe it could be tried as an experimental feature for a while.
It is not such a difficult or invasive patch to make (actually to 
implement this I first had to make a small cleanup of the lib_spec 
functions, which I still intend to submit by itself...).

	Theo.


--------------------------------------------------------------------
Theodore Papadopoulo
Email: Theodore.Papadopoulo@sophia.inria.fr Tel: (33) 04 92 38 76 01
 --------------------------------------------------------------------