Hi Jo, On Thu, 2016-08-18 at 15:13 +0200, Jo-Philipp Wich wrote: > We really need to find a way to combine that, that subtarget churn is > taking precious resources. > > Any subtarget change requires manual intervention on the builders, > manual intervention on the rsync mirrors and between 30 to 50GB of > additional disk space, not taking into account the additional CPU time > needed to build 99% identical code.
I understand that and introduction of a new subtarget was just the last resort for me. Initially I had an impression that I would set ramdisk as a feature of a particular profile but that didn't work for me - ramdisk was enabled that way for all profiles. > Can't we just make initramfs optional and require people to build it > themselves if they need that? Indeed I may leave initramfs selectable in menuconfig and depending on its state one or another set of images (for different devices) will be built. But what I don't really like in that approach - there will be implicit dependency. What I mean if user wants to build for device that uses SD-card image - that's straight-forward, default settings will bring it to you. But if you'd like to build for device that is unlucky to accept SD-card image you may just guess OpenWRT/Lede doesn't support that device. Because again - I tried to distinguish initramfs/non-initramfs builds per-profile but that didn't work for me unfortunately. > I mean who uses initramfs builds in > production anyway? This sounds more like a debug sort of thing to me. Well, really? "git grep ramdisk" shows (except ARC): * apm821xx * ar71xx/mikrotik * rt3050 * ramips/rt305x So ramdisk is used still here and there and we need to take care of it anyways. BTW I expected that kind of discussion on the mailing list when send my last patch-set. Instead entire series got silently applied :) -Alexey _______________________________________________ Lede-dev mailing list Lede-dev@lists.infradead.org http://lists.infradead.org/mailman/listinfo/lede-dev
