[Boost-build] Finding the Boost root directory in sandbox libraries and within a Boost tree

Subject: [Boost-build] Finding the Boost root directory in sandbox libraries and within a Boost tree
From: Edward Diener (eldiener_at_[hidden])
Date: 2010-11-28 21:06:46

In my sandbox library I want to have a way that finds the Boost root 
directory of an installation for my jamfiles, both when the end-user 
uses my jamfiles from within the sandbox directory and also when the 
end-user moves my sandbox library under a Boost tree.

I have been given two methods to do this, one by John Maddock and one by 
Daniel James, for a sandbox library, but I do not see how either method 
will work properly after the end-user moves my library under a Boost tree.

In the method suggested by John Maddock, mentioned in the sandbox 
readme.txt file, the lines:

import modules ;
local boost-path = [ modules.peek : BOOST ] ;

are put in a jamfile and boost-path points to the Boost root. But this 
requires the end-use to either set the BOOST environment variable to the 
boost root directory or specify

--boost=path

on the command line where path is the root of a Boost tree. It also 
appears to require the boost-build.jam in the sandbox directory. I can 
not see how this method is supposed to work once an end-user moves my 
sandbox library under a Boost tree since clearly the boost-build.jam is 
different. I don't mind instructing the end-user that they need to add 
--boost=path to their bjam invocations for my sandbox library wherever 
it is placed, since this is much more flexible than setting an 
environment variable, but given that boost-build.jam will be different 
in a Boost tree, I can not see how this will work there.

In the method suggested by Daniel James I should use the jam file setup 
in the sandbox 'example' library, with its project-root.jam and its 
various jamfiles at the project root level and at the test and doc 
levels. But this requires the BOOST_ROOT environment variable to be set, 
which is something I doubt an end-user is usually going to want to do 
when they move the library under a Boost root since it hardcodes a Boost 
root location. If the end-user then wants to use a sandbox library for a 
different Boost tree, the BOOST_ROOT environment would then need to be 
reset. Otherwise this method seems fine so it seems the best choice of 
the two.

Is there any easier way that a sandbox directory can find the Boost root 
without either of these machinations, so that when the end-user tries to 
use the jamfiles of a sandbox library from within its sandbox directory, 
or moves it under a Boost tree, it sets a bjam local variable to the 
Boost tree root correctly in either case ? Perhaps I am asking for too 
much for a sandbox library. Currently for my sandbox library I am asking 
the end-user to move the library under a Boost tree in order to use it 
and hardcoding a local variable to the Boost root of that tree based on 
that, but I would rather do otherwise especially as it should be 
possible to try my sandbox library without having to move it under a 
Boost tree. I guess I am looking for a combination of Daniel James' 
method and the ease of use in John Maddock's method where a command-line 
option points to the root of the Boost tree since I feel that requiring 
the end-user to set/change environment variables seems onerous.