On Wed, Oct 13, 2021 at 2:39 PM Vegard Nossum <vegard.nossum@xxxxxxxxxx> wrote: > > > On 10/13/21 3:34 AM, Masahiro Yamada wrote: > > On Wed, Oct 13, 2021 at 2:01 AM Vegard Nossum <vegard.nossum@xxxxxxxxxx> wrote: > >> > >> If a given compression algorithm for the kernel image is not usable on > >> the host system, there is no point prompting for it. > >> > >> We can use the kconfig preprocessing feature to check if the command is > >> available or not. I've chosen to test this using "which", which exits > >> with success if the given command exists in PATH (or it is an absolute > >> path), which mimics exactly how it would be found in the kernel's > >> Makefiles. > >> > >> This uses the make variables that are set in Makefile and/or the > >> command line, so you can do e.g. > >> > >> make KGZIP=pigz menuconfig > >> > >> and it will test for the correct program. > >> > >> I am intentionally adding these dependencies to e.g. KERNEL_LZ4, as > >> opposed to HAVE_KERNEL_LZ4, since the latter are "select"-ed > >> unconditionally by the architectures that use them, so they are not > >> suitable for depending on anything else. > >> > >> I've put RFC in the subject as maybe there are downsides to this that > >> I'm not aware of. > >> > >> Signed-off-by: Vegard Nossum <vegard.nossum@xxxxxxxxxx> > >> --- > > > > > > I think we should keep the host-tools dependency open in general. > > You can easily install necessary packages. > > So just to be clear, you object to the patch because it hides the > possibility of using a different compression algorithm from the user and > doesn't give them a chance to know that it exists when it's not already > installed? Yes, I object to it. > I don't really think this is much different from any other choice block > in the config where the visibility of the choices have dependencies on > other config options. Kconfig checks some compiler features, otherwise we have no way to avoid build errors. > In my case, the reason for doing this patch was that I was seeing build > failures during randomized testing (satrandconfig) due to missing > programs, and these build failures appear only at the end of potentially > very time-consuming builds. I do not know Oracle's case, but I think other CI systems install all needed packages for randconfig. Similar discussions happened a couple of times in the past. We keep the tools dependency open since it is just a matter of "apt install" or "dnf install" basically. "and equally importantly, your build servers will actually do a better job of covering the different build options." (Linus Torvalds) [1] [1]: https://lore.kernel.org/all/CAHk-=wjjiYjCp61gdAMpDOsUBU-A2hFFKJoVx5VAC7yV4K6WYg@xxxxxxxxxxxxxx/ > Maybe we can introduce a new option similar to COMPILE_TEST (or > ADVANCED_OPTIONS) so that the options are showed by default, even when > the host program is not available? I do not think it is worth it. > +config UNAVAILABLE_COMPRESSORS > + bool "Prompt for compressors that are not available" > + help > + Note: Enabling this option can lead to build failures if > + the chosen compressor is not available on the host machine. > > config KERNEL_GZIP > bool "Gzip" > depends on HAVE_KERNEL_GZIP > + depends on UNAVAILABLE_COMPRESSORS || $(success,which $(KGZIP)) > help > The old and tried gzip compression. It provides a good balance > between compression ratio and decompression speed. > > > Vegard -- Best Regards Masahiro Yamada
