Hello, I intend to fully drop all remaining support for multilib from rebootstrap. If you agree with this, you may stop reading now and skip replying.
Back in Brest, we were discussing a number of toolchain matters in various BoFs. Some of those sessions touched on multilib and my impression back then was that nobody loved multilib. While we prefer cross toolchains, some of the multilibs target non-release architectures (e.g. amd64 -> x32) and such support is supposed to be kept at least for now. There seemed rough consensus on phasing out multilib usage where feasible. Debian packages should no longer use multilib options such as -m32, -m64 or -mx32 during build and use cross toolchains instead. I've sent patches to some packages such as #1107125, #1107120, #1107127 and #1107128. Getting rid of multilib in library packages such as glibc, ncurses, readline or zlib1g is a different story saved for later. In any case, the trend seems clear and the need for multilib is becoming exceptional. As such, I want to stop considering it part of the architecture cross bootstrap. This basically amounts to turning on the "nobiarch" build profile unconditionally, but it also means ripping out the remaining support for multilib and no longer testing (nor fixing) it. The outcome on my end would be a noticeable simplification of the tooling. The cross bootstrap presently includes a manual order for 40 out of 150 builds. Three of those would readily go away together with a pile of workarounds. Note that cross-toolchain-base and cross-toolchain-base-ports also perform a cross bootstrap. I am not in a position to drop their support for multilib, but their support will likely suffer from me no longer testing multilib support. Even though the cross toolchains provide multilib support, it is difficult to use in cross builds, because there is no gcc-multilib-for-host package and depending on gcc-multilib readily turns your package cross-b-d-uninstallable. Please let me know if you have any objections or concerns. I suggest centralizing any followup discussions to the [email protected] mailing list only. Helmut
