Hello. I'm writing to you as to maintainer of FHS multiarch proposal. (http://www.linuxbase.org/futures/ideas/multiarch/) I'm working on Debian cross-toolchain packages. Recentry, there was a short discussion on debian-devel mailing list (http://lists.debian.org/debian-devel/2005/07/msg00517.html) about placement of files of cross-compilation environment. In that discussion, Goswin von Brederlow <brederlo@informatik.uni-tuebingen.de> suggested me to get in contact with FHS multiarch maintainer with the following issue. Cross-compilation setup is not the same as multiarch setup: multiarch is about running different architecture binaries on the system (and building binaries intended to run on the same system), while cross-compilation is about building binaries for foriegn architectures, that can't run on the build system. However, both multiarch and cross-compilation setups require of installation of libraries and headers for different architectures on the same host. So it seems reasonable to use common policies about placing those files. Cross-compilation setups are in wide use for many years, and there is a de-facto standard that libs are placed into ${prefix}/${target}/lib, and headers are placed into ${prefix}/${target}/include. This convention is coded in binutils and gcc packages, as well as in other toolchains. This differs from what you propose for multiarch (${prefix}/lib/${target} ans ${prefix}/include/${target}). Your argument is that you wish to prevent /usr pollution. But it seems that this 'pollution' argument is not strong enough to change the way how cross-compilation setups worked for years. *) The 'pollution' is actually tiny - number of 'arch directories' that will appear under /usr will not be larger than number of architectures that multiarch system supports - which is probably very small. *) Having things placed under common prefix makes it easier to create chroots, target system images and similar things. *) Many existing cross-compilation tools work with current setup, and quite a lot developers expect things to be this way. What do you thing about the above? Isn't it better to make multiarch proposal consistent with de-facto standard of cross-compilation setup? WBR, Nikita Youshchenko.
Attachment:
pgpghq8s3LBob.pgp
Description: PGP signature