Hi All,
I noticed that a lot of packages call make passing both "all" and "install"
target at the same time. This is not right because we cannot sure that a rule
like "install: all" is present in the package's makifile. To fix the problem,
we
need to do two different make calls, the first with "all" target and the second
with "install".
After this, most packages should me parallel-make-aware.
(I will send the patches asap).
I Have another RFC:
1) define PKGMAKE:=$(MAKE) -j$(CONFIG_JOBS)
2) define PKGMAKEJ1:=$(MAKE) -j1
My idea is to parallelize also the OpenWRT makefiles... what do you think about
this?
best regards,
luigi
Il lunedì 19 ottobre 2009 12:39:20 Luigi 'Comio' Mantellini ha scritto:
> Hi All,
>
> I noticed some packages (like ncurses, ppp, ...) that don't like the
> parallel make (with -jX X>1).
> My proposal is to define:
>
> MAKEJ1=$(MAKE)
> MAKE:=$(MAKE) -j$(CONFIG_JOBS)
>
> and use MAKEJ1 for these packages until we don't fix the issue.
>
> is it ok?
>
> ciao
>
> luigi
>
> Il domenica 18 ottobre 2009 18:39:05 Felix Fietkau ha scritto:
> > Luigi Mantellini wrote:
> > > Hi Felix
> > >
> > > On Sun, Oct 18, 2009 at 4:22 PM, Felix Fietkau <n...@openwrt.org> wrote:
> > >> Though you prefer this option, I believe using external toolchain
> > >> support in OpenWrt will always be the exception, not the rule. We can
> > >> easily add toolchains like CodeSourcery's or ELDK's to our own build
> > >> infrastructure, as their sources are open as well.
> > >> Even external toolchain support itself does not guarantee any form of
> > >> consistency, since you could have accidentally used the wrong
> > >> toolchain and the system might not notice. Because of that, I also
> > >> consider the dependency on external toolchains to be bogus.
> > >
> > > Ok. I agree. I know that "external" is not equivalent to say
> > > "reliable".
> > >
> > >> Here's how I would ensure consistency:
> > >> - Change OpenWrt toolchain identification string to include the arch
> > >> - On each distcc build host, compile a simple test program such as
> > >> 'int main(int argc, char **argv){ return 0; }' to assembler and grep
> > >> for the .ident string and compare it against the same thing on the
> > >> local machine.
> > >
> > > ok good idea.
> > >
> > > Anyway, I include my last patch version. I just tested compiling a
> > > kernel and busybox of an arm target using the codesourcery
> > > toolchain... it seems to work fine but I interrupted the compilation
> > > (today is sunday... and I need to see also the sun).
> > > We can consider this patch as start point, if you agree.
> >
> > Yep, sounds good. By the way, I'm currently working on integrating a gcc
> > version with CodeSourcery patches into OpenWrt. Maybe that way you also
> > won't have to use external toolchains anymore. I've stripped the diff of
> > the latest codesourcery toolchain against upstream gcc 4.3.3 down to
> > around 2M in size and merged it with our other patches. I'm currently
> > checking whether it compiles...
> >
> > - Felix
> > _______________________________________________
> > openwrt-devel mailing list
> > openwrt-devel@lists.openwrt.org
> > https://lists.openwrt.org/mailman/listinfo/openwrt-devel
>
--
Luigi 'Comio' Mantellini
R&D - Software
Industrie Dial Face S.p.A.
Via Canzo, 4
20068 Peschiera Borromeo (MI), Italy
Tel.: +39 02 5167 2813
Fax: +39 02 5167 2459
Email: luigi.mantell...@idf-hit.com
_______________________________________________
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel
