Hey Diego, Thanks for looking into this. The build system has been an annoyance for awhile, but I'm struggling with the best way to deal with it.
On Wed, 2011-09-07 at 01:54 -0700, Diego Elio Pettenò wrote: > As I noted in my previous mail, the current build system is suffering > from a high level of recursion that makes the build (and the install) > very slow on modern multi-core systems using parallel make. > > Now, it isn't difficult to make the Automake files non-recursive, and > that would solve the problem. But there is a matter of choosing how to > do so, as there are different methods. > > My usual favourite method is to use subdir-objects to make sure that the > object files are built in the same directory hierarchy as the source > files, for the sake of order, but then build all the targets (libraries > and commands) in the top-level directory. > > Considering how many tools freeipmi consists of, this could be a bit > chaotic. Given the large number of tools, I'd definitely prefer not to do this. > The alternative would be to declare the tools in the subdirectories, but > that makes the Makefile.am syntax a bit more complex (i.e. to build > bmc-watchdog/bmc-watchdog you'd have something like > > bmc_watchdog_bmc_watchdog_SOURCES = This would probably be preferred. But before you go and do anything. Let me think about this more. There is still the tiny part of me that thinks, "It's not broke, no need to change." When I'm developing the code, it's still nice to be able to type 'make' in the subdir I'm in, instead of typing make at the top-level directory. But at the same time, I realize that I should probably give up that comfort for the sake of the distro maintainers out there. I've contemplated re-orging alot of the code too. Perhaps something in between would be good? Hypothetically re-org to something like: freeipmi/libraries freeipmi/tools freeipmi/man So the recursion would be 1 level deep? I sort of fear having some monster makefile in the top-level directory that includes almost everything in FreeIPMI. Al > Before trying to provide some build system patches I'd like some > comments on what is preferred to deal with it. > > Thanks, -- Albert Chu [email protected] Computer Scientist High Performance Systems Division Lawrence Livermore National Laboratory _______________________________________________ Freeipmi-devel mailing list [email protected] https://lists.gnu.org/mailman/listinfo/freeipmi-devel
