Solved 10.3 build error

[Crivens]

When trying to make buildworld on my main machine (amd64 running releng-10.2, I run into a build error.

/usr/src/lib/libc/posix1e/acl_support_nfs4.c:51:8: error: use of undeclared identifier 'ACL_ENTRY_INHERITED'
     { ACL_ENTRY_INHERITED, "inherited", 'I' },

Looking at the internet has not given me much clue for this as has searching here. It surely is only some small thing, but I can't see it.
I'm somewhat unenthusiastic to update the kernel without a built userland in reach.

 

[kpa]

Well that's odd. I'm sure I compiled 10.3 pre-releases on 10.2 and everything went fine. The macro in question exists only 10.3:

http://svnweb.freebsd.org/base/releng/10.3/sys/sys/acl.h?revision=296373&view=co

Compare to:

http://svnweb.freebsd.org/base/releng/10.2/sys/sys/acl.h?revision=285830&view=co

However, the world build should use only headers from the sources if everything is working correctly. Try cleaning out /usr/obj completely and try again.

 

[Crivens]

Just did that, also I tried it with ccache and without...
The current state of affairs, should it be of relevance, is

Revision: 298354

 

[kpa]

I found the commit that added/changed the macro, it was done on stable/10 before releng/10.3 was forked.

http://svnweb.freebsd.org/base?view=revision&revision=290893

The comment doesn't seem to apply to the changes on the acl and nfsv4 related files so whatever the problem is we also have a sloppy commit that doesn't document everything that was changed.

 

[Crivens]

Strange things... I ended up distrusting svn telling me no changes were present. So I removed the complete source tree and checked it out again, and then it builds. I still have the old one around on a backup, but I am not so hell-bent to get to the bottom of this to resurrect that one for comparison.

 

[Crivens]

Sorry, but I have to re-open this.
It turns out the svn checkout is not the issue.
My /etc/make.conf looks like this:

WITH="GALLIUM"  # Needed for Radeon cards, but does no harm with Intel devices.
OVERRIDE_LINUX_BASE_PORT=c6
OVERRIDE_LINUX_NONBASE_PORTS=c6

OPTIONS_UNSET+=PULSEAUDIO
OPTIONS_UNSET+=HAL
OPTIONS_UNSET+=GVFS

WITH_CCACHE_BUILD=yes

#begin ccache settings
#.if ${.CURDIR:M*/ports/devel/ccache}
#  NO_CCACHE=yes
#.else
#  CC=/usr/local/libexec/ccache/world/cc
#  CXX=/usr/local/libexec/ccache/world/c++
#.endif
#end ccache settings

When I remove the comments which currently block ccache from being used, the build breaks.
And I have cleared the cache, yes.

 

[Crivens]

So - I re-installed ccache not setting the CLANGLINK and LLVMLINK options, removed my entries in make.comf and set them according to the install instruction. Copy-paste. It looked the same, so I was surprised when "make buildworld" immediately balked. Turns out that formatting your make.conf to make conditional scopes from ".if" and so on is a bad idea - I added some whitespace to make clear which .if controlled what. Maybe that was it, I removed it. Right now, the build is humming away and filling the cache. I'll file this under "tales from the twilight zone". Have a nice weekend.