mirror of
https://github.com/Klipper4FlashForge/firmware.git
synced 2026-10-22 05:12:16 -05:00
Oleksandr ShyshatskyiandClaude Opus 5
dcd037c250
pkgs/ipk-install is POSIX sh because the PRINTER has no opkg and no ar -- measured, and its header says so. Neither is true in the build container, and the imitation is missing exactly the parts that decide what a payload should contain: it resolves no Depends, enforces no Conflicts, reads no Provides and handles no conffiles. Assembling a payload with it means arguing that the result matches what opkg would have done. pkg_buildopkg builds an x86-64 opkg from the same pinned tarball the mipsel one is built from, cached at work/.opkg-host, shaped as pkg_buildpython's twin so that "build a host tool from the pinned source" means one thing here. ipk-install keeps its job -- repairing a printer by hand -- and comes off the build path. Two things this build gets wrong if written the obvious way, both now checked rather than commented: --prefix must be $MODDIR. opkg bakes its state directory in at compile time (libopkg/Makefile.am: -DVARDIR="@localstatedir@"), so a host opkg built --prefix=/usr/local keeps its database somewhere the printer never reads and says nothing about it. The prefix is a literal string in the binary, so the question is put to the file. --disable-shared is not a size decision. The prefix is baked into the LIBRARY too, so a shared build produces a bin/opkg looking for libopkg.so.1 at a path that exists on the printer and not here. libarchive-dev is Debian's, deliberately: opkg's configure.ac requires libarchive, and this opkg is a build tool that never leaves the container. The same call is already made for the gcc that builds the build-python. Measured against the built feed: installs into an --offline-root with the database at $MODDIR/var/lib/opkg, and refuses the two chamber configs together (exit 255) on the Conflicts field ipk-install cannot read. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>