Hi Ludo, >>> * gnu/packages/bash.scm (bash-minimal): Assume getcwd works correctly >>> when cross compiling. >> No tabs please. Otherwise LGTM, thanks! > I spoke too fast. On master (Bash 4.3), this is unnecessary AFAICS: > > --8<---------------cut here---------------start------------->8--- > $ ./pre-inst-env guix build bash-minimal --target=mips64el-linux-gnu > --no-grafts --no-build-hook > /gnu/store/1v6l54xzmzb19hdg5bizvzfz995lh1cp-bash-minimal-4.3.42-doc > /gnu/store/gawbc9mj2i37ycym06rbybi1k4kn8zfq-bash-minimal-4.3.42-include > /gnu/store/vic4zf9gpdzrcvj8kimb9cs3049ld60d-bash-minimal-4.3.42 > $ git describe > v0.11.0-1743-gfe9bdb5 > --8<---------------cut here---------------end--------------->8---
Building bash was never a problem, it worked also for me without the patch. The problem is that, when cross-compiling, bash "configure" scrip cannot test whether system getcwd works properly or not, and thus decides to use its own version of it, which does not correctly work inside bind mounts (and therefore fails in the chroot, but works outside it). Thus, unpatched bash will fail when used to build another derivation. The problem was first reported here: https://lists.gnu.org/archive/html/guix-devel/2013-10/msg00063.html Although in that case Mark worked it around by having /tmp in a different partition, which incidentally causes bash-provided getcwd to work correctly. Maybe I can put a more detailed explanation on bash.scm; I packed the patch in a hurry in case Efraim needed it for its bootstrapping also. My changes make bash "configure" assume the system getcwd works correctly (which in our case we know it does as dependencies are controlled). > Is this addressing a problem you had with Bash 4.4 (on core-updates)? No, it happens with bash-4.3 in master (and probably with most other recent versions as well). Thanks! Carlos
