On Sun, Oct 21, 2012 at 9:24 AM, Petr Štetiar <yn...@true.cz> wrote: > Ok, sorry. I've meant "mtd" package, but wrote "mtd-utils". Anyway, the > conditions metioned in my previous email still apply. Without that "mtd" > package it's not possible to use jffs2 /overlay. I noticed that many of the system scripts use mtd, so to reflect this in the package metadata I tried to make mtd a dependency of base-files. This would mean that if you select base-files, you also get mtd, achieving the same effect as making mtd mandatory.
Unfortunately, that broke the build, because mtd.c includes headers directly from the kernel tree instead of sticking to the exported headers only. I suspect this was introduced by this change: commit 6b08cfb731a86c548c93ae426a668f6336f4be17 Author: jogo <jogo@3c298f89-4303-0410-b956-a3cf2f4a3e73> Date: Mon Oct 8 09:59:47 2012 +0000 package: mtd: use in-kernel bcm63xx image tag header Use the image tag header from the kernel instead of an external header file. openwrt exports kernel headers into build_dir/linux-X/linux-X.Y.Z/user_headers, but the bcm63xx header (bcm963xx_tag.h) is not present there. I don't know what the criteria for exporting a header is. Could be one of several things: 1) bcm963xx_tag.h might be missing something in the source code to get it exported. 2) mach (arch/X/include/asm/mach-X/*.h) headers, which is where bcm963xx_tag.h lives, might be treated specially. After all, the kernel probably doesn't want to expose a machine-dependent interface to userspace. 3) Some other reason I haven't considered because I have no real clue how any of this works. I also don't know why the dependency of base-files on mtd surfaced this problem. It must have changed the build order for some packages. I don't think just reverting the change is the right move either - we probably *do* want to share this header to avoid duplication. There must exist device interfaces that are conditioned on a specific machine, maybe this header just needs to be moved to a platform-independent include dir? Hope this helps :) If anyone knows some of the philosophy behind this, I would be happy to hack away at some patches, but as it stands, I don't have much to go on. Catalin _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel