From: Beman Dawes (beman_at_[hidden])
Date: 2000-12-17 11:01:23
IIRC, the key reason we originally decided not supply library build scripts
for various platforms was because we thought adding a new library meant
modifying build scripts and makefiles for platforms the developer knew
nothing about and had no way of testing.
Could this be solved by adding a level of indirection?
If the names (and relative paths) of the .ccp files to be built were
supplied in a text file in an agreed upon boost format, don't most
platforms have some kind of native scripting language capable of doing the
build? Even Windows batch files can handle this.
Thus when a new boost library is added, only the single text file has to be
modified. The scripts (or makefiles or whatever) don't have to be
modified. A developer can test on his or her usual platform and compiler
without worrying about other platforms or compilers.
Some integration of the library build process and the regression test
process would be enough to detect specific platform or compiler problems.
While in theory such an approach would mean maintaining an ever larger
number of build scripts, I'm not sure that is a practical problem. Many of
the practical problems are already being addressed now that we run
regression tests for a number of platform/compiler combinations.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk