I know there's been some discussion about the right dependencies of the *-sources packages, but at any rate, ncurses probably shouldn't be mandatory. I think there should either be no such dep ("the user is smart enough"), or the ncurses dep should be controlled by USE="ncurses". The USE way is the cleanest IMO, and could be perhaps further extended to USE="gtk qt". For example, if I want to use "make gconfig", I have to install libglade by hand. As for the infamous deps on lzop/xz-utils/gzip/bzip2, it's less of a deal, I guess, but again, it's natural in Gentoo to have them. Didn't somebody just forget to put USE flags to those packages? The flags "lzo", "lzma", "gzip" and "bzip2" are well and alive and even global. (No, besides these two sets of flags I don't see a space for any other. "Running" the *-sources package is about compiling the kernel, not running the produced kernel.) Reproducible: Always
To introduce such USE flags I guess we have to patch certain make options such that they bail out with an error that the USE flag needs to be enabled. Now we just assume these are present such that the full functionality can be used; but indeed, some users may not want all dependencies. As for the compression formats, genpatches is only available in xz and bringing them out in all four file formats would be tedious to do. I don't see anyone that has an issue with this dependency as it is quite minimal; granted, there is improvement possible here but it would require changes to the eclass as well. Patches to the eclass are welcome.
(In reply to comment #1) > To introduce such USE flags I guess we have to patch certain make options > such that they bail out with an error that the USE flag needs to be enabled. You're literal indeed. Wouldn't it be better to just extend the error messages that "make x/q/gconfig" print ("you don't have the necessary libs - remerge with USE=gtk")? The point is that maybe the make options will just happen to work for a user even without the USE flags. In which case - let them just use the make options freely. > As for the compression formats, genpatches is only available in xz and > bringing them out in all four file formats would be tedious to do. I don't > see anyone that has an issue with this dependency as it is quite minimal; > granted, there is improvement possible here but it would require changes to > the eclass as well. I didn't think of this. But xz-utils is required by @system on amd64/hardened and x86/default (and perhaps others as well). I suppose one of the reasons for that is that an increasing number of distfiles and patches come naturally in .xz. I suggest leaving the issue of the patch format to something beyond the scope of this bug, as that's where it seems to belong, to me. The compilability of the kernel, on the other hand, is specific to the *-sources packages and is dependent on the packagers' availability. > Patches to the eclass are welcome. Working on one.
Created attachment 348184 [details, diff] kernel-2.eclass.diff * binutils removed from DEPEND - no apparent need (?) * all RDEPENDs from top to gcc, just as DEPEND=sed, are "obvious" * the "modules" USE flag description in use.desc may need updating to something more general, like "Support installation of kernel modules." * I suppose portage merges REQUIRED_USE from eclasses and ebuilds (right?) - not using REQUIRED_USE+=... here * "initramfs" needs to be included in use.local.desc * the kernel sources have more: make *-pkg, make *docs - these would require new USE flags and deps, but I'll leave these for a later time
Created attachment 348186 [details, diff] kernel-2.eclass.diff Forgot the !build? condition in REQUIRED_USE.
Created attachment 348244 [details, diff] kernel-2.eclass.diff Moved the new ewarns into pkg_setup(). Could be overlooked: I dropped the RDEP on bc. Please re-add / let me know if there is a reason for it that I don't see.
Created attachment 348246 [details, diff] kernel-2.eclass.diff Beg your pardon, bc is necessary and so is the binutils version spec, for semi-archaic reasons.
Created attachment 349750 [details, diff] kernel-2.eclass.diff
Added the "development" local flag to pull in static code checking tools used by the root Makefile.
Will look at this by the upcoming 3.11 release; together with other eclass changes, to reflect the config / experimental discussion on gentoo-dev.
Dependencies have improved since this bug, I think they are adequate as this juncture
