On Thu, 08 Nov 2001 11:03:43 -0500, Pragnesh Sampat <[EMAIL PROTECTED]> wrote: >Out of curiosity, (please pardon me for jumping in like this since I >am still catching up with kbuild issues), but has Cons >(http://www.gnu.org/software/cons) been considered for linux kernel >build? It does a better job at dependencies (using MD5 checksums) and >has nicer support for generating as many object trees as required from >single source tree (once for each cpu, for example).
I looked at cons and other make replacements but rejected them. One of the requirements for kbuild is to use standard GNU tools, I did not think we could get a kernel build accepted if it required extra software (ignoring the "why do we need Python for kbuild" crowd ;). So I use a wrapper around make, all the wrapper code is supplied by kbuild. None of the make replacements could cope with multiple source trees. kbuild must support compilation of drivers, architectures etc. that are not yet in Linus's tree. Separate source, single compile mechanism. None of the replacements could cope with the seriously weird kernel requirements. The kernel dependency tree is very weakly defined, we do not attempt to track config or header dependencies in the makefiles but we expect a single config change to automatically recompile only the affected files. Files are compiled and linked differently for kernel and modules, changing a single config can have major side effects. _______________________________________________ kbuild-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/kbuild-devel
