Re: [boost] Build system/GCC questions

From: Thomas Matelich (sosedada_at_[hidden])
Date: 2001-06-04 17:55:29

David Abrahams wrote:

> I am trying to understand how best to support shared libraries in the build
> system. Several people have mentioned to me that under GCC, DLL source files
> should be compiled with -fPIC. Is there any reason not to compile all
> targets with -fPIC under GCC? Does it generate less-efficient code? If it's
> important to be able to build some targets without -fPIC, it's clear to me
> that we need a way of expressing constraints that a target type places on
> its build configuration. Also, I assume this constraint is transitive, at
> least on any targets that are /statically/ linked to the shared library.
> Please enlighten.

I can't give you a canonical answer, but I'll share what I know.  A shared
library must be built with PIC object files.  Anything can be built with PIC
code, but linking will be slower for static entities than if you didn't have
PIC.  I can't remember if PIC code runs slower, but I don't think so.

Regarding your transitive assumption, things are peculiar.  On Linux (ELF
format), a shared library can link in a static library even if the static
library was not built PIC.  On HP-UX (not ELF), anything linked into a shared
library must be build PIC.  So, it is tempting to build everything PIC because
of platforms like HP-UX.

I would have expected Jam to have already figured this stuff out.  If you're in
the mood to look at a different build system, you could get tmake from
ftp.trolltech.com/pub/freebies/tmake.  They have a hierarchical system of rules
regarding building apps and libs on many platform-compiler combinations.
Alternatively, qmake (c++ implementation) is bundled into the qt 2.30 beta and
adds support for Mac.

I was going to suggest boost look at qmake, but I don't know if they go as far
with things like debug and release.  Otherwise, their .pro file format is quite
nice.

--
Thomas O Matelich
Senior Software Designer
Zetec, Inc.
sosedada_at_[hidden]
tmatelich_at_[hidden]