cvs commit: ports/java/jdk16 Makefile

[Greg Lewis]

On Wed, Jun 09, 2010 at 12:47:57AM +0300, Ion-Mihai Tetcu wrote:
> On Tue, 8 Jun 2010 17:28:08 -0400
> "Philip M. Gollucci" <pgollucci at p6m7g8.com> wrote:
> > On 6/8/2010 4:34 PM, Rob Farmer wrote:
> > > On Tue, Jun 8, 2010 at 1:17 PM, Ion-Mihai Tetcu
> > > <itetcu at freebsd.org> wrote:
> > >> itetcu      2010-06-08 20:17:37 UTC
> > >>
> > >>  FreeBSD ports repository
> > >>
> > >>  Modified files:
> > >>    java/jdk16           Makefile
> > >>  Log:
> > >>  Mark BROKEN, I'm tired to see QAT loosing time trying to build it.
> > > 
> > > Could you move this inside an .if defined(PACKAGE_BUILDING) or
> > > something? It seems a little extreme (a lot of stuff has a jdk
> > > dependency)
> 
> I know, believe me:
>  > ./QAT_getFailledDepends.sh -d java/jdk16
> +--------------------------------------------+
> | No. of failed ports becasue of java/jdk16: |
> +--------------------------------------------+
> |                                        145 |
> +--------------------------------------------+
> 
> > and I just built this yesterday on amd64 current with no issues.
> > > 
> > Yeah I've built it too and recently (as much as I despise java)
> > http://people.freebsd.org/~pgollucci/FreeBSD/logs/9-CURRENT-amd64-rubygems18/jdk-1.6.0.3p4_15.log
> > 
> > nothing against glewis, he's doing great work.
> 
> It's been failing each time on QAT since 15.02, and for others since
> January. I don't see what it has to do with package building.
> 
> I guess it would nor err like this after: 
>  r200039/_FreeBSD_version 900003 for HEAD  and  r203299/800502
> (QAT's base OS is a 8.0-RELEASE-p2)
> 
> However it's failing for me on an other 8-STABLE tinderbox also, which
> has a much newer base OS  kernel, but in a different way.
> 
> I will invetigate more tomorrow, time permitting.

Please revert your commit.  Its building fine for myself and many other
users.  I cannot reproduce the QAT failure and there are no PRs about it.

QAT complains thusly:

../../../../src/solaris/hpi/native_threads/src/interrupt_md.c:115: error:
static declaration of 'sigignore' follows non-static declaration
/usr/include/signal.h:103: error: previous declaration of 'sigignore' was here

The build uses the following logic to detect the presence of sigignore:

ifeq ($(OS_VENDOR),FreeBSD)
OS_RELDATE := $(shell sysctl kern.osreldate | cut -d' ' -f2)
HAS_SIGIGNORE := $(shell if [ $(OS_RELDATE) -ge 800502 -a $(OS_RELDATE) -lt 900000 -o $(OS_RELDATE) -ge 900003 ]; then echo "1"; fi)
ifeq ($(HAS_SIGIGNORE),1)
override HAVE_SIGIGNORE          = true
endif
endif

I suspect that works for 99.9% of users but may fail for jailed 
applications like QAT where syctl outputs a kern.osreldate that doesn't
match the userland headers.  I'm open to suggestions on different logic,
but I don't feel the narrow failure mode justifies your change.  If you
don't want QAT to build the port then modify QAT to be able to exclude
ports rather than hurting the vast majority of users for whom this port
does build.

FWIW the port builds without problem on my 8.0-RELEASE-p2 amd64 box.

-- 
Greg Lewis                          Email   : glewis at eyesbeyond.com
Eyes Beyond                         Web     : http://www.eyesbeyond.com
Information Technology              FreeBSD : glewis at FreeBSD.org