Re: [Boost-build] Why does library naming not include architecture details

Subject: Re: [Boost-build] Why does library naming not include architecture details
From: Vladimir Prus (ghost_at_[hidden])
Date: 2010-08-13 02:45:44

On Wednesday 11 August 2010 18:06:10 Ian Emmons wrote:

> I also would like to see the address model and architecture added to the library names.  This would greatly simplify my process to build Boost libraries (since I could use a single stage directory) and my site-config.jam file (since it would not need a ton of conditionals).

There's a tradeoff here. Adding new things to name simplify life for folks who build
for many targets in the same place. But it seems to complicate matters if you build
for one target at a time, and you need to specify names of libraries somewhere --
e.g. in install scripts. So far, "add architecture to name" folks were not pushing
too hard.

> 
> A related gripe:  The current system of splitting the address model and architecture as two independent features is awkward, because these are not orthogonal attributes of a binary image.  When you look at Windows and Linux, it is tempting to think of them as orthogonal, but when you include IA64, Macintosh, and the host of smart phone architectures and operating systems, it becomes clear that many architectures have only one address model, and Macintosh (at least) incorporates several architectures and address models at once.  A better way for Boost.Build to represent the architecture and address model combinations would be to imitate the architecture identifiers used by GCC.

Well, GCC actually uses *target* identifiers -- which fuse architecture, OS, and "manufacture"
(which is something of no particular meaning). Also, in addition to that, you have
-m32/-m64 options. So, it's not clear this classification is much better.

Thanks,

--
Vladimir Prus
http://vladimir_prus.blogspot.com
Boost.Build: http://boost.org/boost-build2