On Jun 15, 2015, at 1:27, Simon J. Gerraty <s...@juniper.net> wrote:

> Konstantin Belousov <kostik...@gmail.com> wrote:
>> I see the same problem on the up-to-date stable/10 host, trying to build
> I'm building HEAD on stable/10, I just updated the tree and did a clean
> tree buildworld ok.
>> HEAD.  This is completely unacceptable, we have documented and always
>> supported just
>>      make buildworld buildkernel
>> on latest stable to get buildable and bootable HEAD.
> I think that proposition became shaking as soon as the current options
> model was added to the build - any new option will break building
> unless you use the share/mk from head.
> Hmm ok I've introduced the concept of options being set during sys.mk
> which makes the above issue slightly worse.
> Craig as already committed a fix for src/Makefile,
> hmm I thought it had logic to switch to the tree's share/mk
> in which case the change to src/Makefile should be all that's needed.
>> There is yet another issue with the build system, I have
>>      INSTALL+=-CS
>> in make.conf for around 15 years.  Apparently it is broken now.
> Ah! make.conf is getting included earlier.
> So that {local,src}.sys.mk can be included earlier so that any of them
> can provide pointer to external toolchain before sys.mk sets CC etc.
> That sucks, though that implies you are getting sys.mk from src/ at
> least after the build gets going.
>> The installkernel target results in
>> ph/vjc (install)
>> --- _kmodinstall ---
>> CS -o root -g wheel -m 555   ng_vjc.ko 
>> /usr/home/kostik/build/bsd/DEV/p/boot/kernel
>> CS: not found
>> etc.

This is going to break a whole lot of stuff — especially because I’ve 
discovered people either try and be overly clever with make, or like 
.include’ing things more than once >:(.

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail

Reply via email to