This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

FW: configure-build ?


(try again, without the attachment, and as plain text)


> From: jay.krell@cornell.edu
> To: gcc@gcc.gnu.org
> Subject: configure-build ?
> Date: Mon, 15 Sep 2008 04:05:25 +0000
>
>
>
> Toplevel makefile has configure-host and configure-target.
> Is there a reason why no configure-build? It might be nice.
> My build wrapper applies some patches after configuration:
>
> for a in ["build", "host", "target"]:
> #
> # configure-build does not exist
> #
> if a != "build":
> Run(Obj, "make configure-" + a + " " + Environ)
> PatchOutLibIConv(Obj)
> PatchOutOptimizer(Target, Obj)
> Run(Obj, "make all-" + a + " " + Environ)
>
> Run(Obj, "make all " + Environ)
>
> if True:
> print("installing " + Host + "/" + Target)
> Date()
> Run(Obj, "make install " + ExtraInstall + Environ)
>
> I can try to provide a patch..much later.
>
> Also, configure -no-libiconv might be nice.
> I've gotten gcc from folks for Solaris that didn't work for me due to the libiconv.so dependency.
> I'm a big stick in the mud wrt non 7-bit text files. :) My editor doesn't understand Unicode.
>
> Maybe configure -no-opt?
> One measurement I did, removing -O2 sped up building the compiler by half.
> (This was with 4.3.x, not current svn.)
> Currently I'm more concerned with getting something that builds and works, code quality
> is not a huge concern. I also don't like debugging optimized code.
>
> Perhaps leaving -no-opt as an easy option is bad, it'd lead to people using it more.
> My function PatchOutOptimizer is a *little* tricky, just that it visits so many variables
> in so many files. (was attached but bounced).
>
> - Jay


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]