Hi, > -----Original Message----- > From: [email protected] [mailto:[email protected]] > Sent: Samstag, 22. Februar 2020 16:29 > To: [email protected]; [email protected] > Subject: Re: [OpenWrt-Devel] Some questions - help needed > > On 20. 2. 22. 9:48 PM, [email protected] wrote: > > to make a variable "per-device", you have to add it to DEVICE_VARS, e.g. > > DEVICE_VARS += DEVICE_TYPE > > I've tried it, but it did not work. DEVICE_VARS is used in include/image.mk > but > DEVICE_TYPE is used in include/target.mk, I guess that is the cause.
The different files shouldn't be the problem, but if you have added DEVICE_TYPE to DEVICE_VARS, every device will need to have it set. Those which do not have a DEVICE_TYPE of their own will just take over the DEVICE_TYPE of the previous device (!), so setting it for one device will effectively also set it for all the others after it. The default DEVICE_TYPE in target.mk then will most probably only work for those devices before the first per-device definition. So, if you chose to have DEVICE_TYPE per-device, you will have to add it to Device/Default definitions in all targets using it, as this will then provide a real per-device default value in contrast to the situation just described. Note that this just describes how DEVICE_VARS works, I have not had a look into what DEVICE_TYPE actually does. Best Adrian > > It seems "DEVICE_TYPE:=nas" does two things: > 1. add more packages to DEFAULT_PACKAGES > 2. add hdparm in busybox > The first one might be done in a device-specific way, but I'm not sure about > the > second one... > > > I've once tried to solve this for the same problem but with kmod-i2c-core > > IIRC, > > and then quit at some point because I couldn't find the initial source of > > the > problem. > > I found it: i2c-gpio-custom. If I remove the package directory, all i2c > packages > are nested now. Maybe scripts/package-metadata.pl has some bugs? But it's not > the > main topic here. > > I replaced all "DEPENDS:=kmod-i2c-core" with "DEPENDS:=+kmod-i2c-core", and > removed > unnecessary kmod-i2c-core in DEVICE_PACKAGES. And build test (on ramips) was > successful. > So I think I can provide a patch for it (though it needs more build tests). > > But what I'm wondering is, how far will we go? Can we add "+" to all kernel > packages (unless it leads to a dependency error)? If "select" alone is enough, > then why do we also use "depends on"? _______________________________________________ openwrt-devel mailing list [email protected] https://lists.openwrt.org/mailman/listinfo/openwrt-devel
