From: John Maddock (boost.regex_at_[hidden])
Date: 2003-08-29 06:13:19
I'm still playing with your revised install code, and I think that by and
large I'm getting to like it :-)
Still a few more points though:
I'm pretty happy with the "stage" target as an alternative to install, but I
wish it was called something else - "binaries" is still my favourite, but
you've already listed the other options, and as ever finding a good name is
I think I can live with the headers going in "versioned" sub-directories,
one of the things I'm keen on is that an installed boost can be discovered
by an autoconf script, and I think this is the case here.
I'm still not happy about the libraries going in $prefix/lib/something/ for
the following reasons:
1) The "something" doesn't add anything, this is already an architecture
specific directory, so putting files in "x86-linux" for example adds
nothing. Some libraries do seem to install themselves in
$prefix/lib/libname/ (for example kde seems to do that, but I haven't dug
into the sources to verify that it's a kde thing,and not just the way Red
Hat have packaged it). I don't think this adds anything in our case either,
since all our libraries have a "boost_" prefix anyway.
2) The sub-directory name can't be discovered by autoconf as far as I can
see - as it stands the name is very much a "Jam'ism", and doesn't relate to
uname or anything else as far as I can see.
3) The libraries won't be in the users LD_LIBRARY_PATH: clearly there are
workarounds for this - like starting boost-dependent programs via a shell
script that sets the necessary variables, but this is another hurdle for
already hassled programmers to overcome (and an added complication for
installation scripts - the user wouldn't be able to just use automake
without jumping through hoops).
p.s. If you just want me to shut up about this, just say so... ! :-)
Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk