ldconfig when PACKAGE_BUILDING=YES (and linux ports)

[Alexander Leidinger]

Quoting Boris Samorodov <bsam at ipt.ru> (from Sat, 04 Aug 2007 02:09:47 +0400):

[CCing emulation at ...]

> On Fri, 03 Aug 2007 23:58:47 +0200 Pav Lucistnik wrote:
>> Boris Samorodov píše v so 04. 08. 2007 v 01:30 +0400:
>
>> > Seems that running ldconfig while building a package at package
>> > cluster (i.e. when PACKAGE_BUILDING is defined) is quite useless. [1]
>> >
>> > To be more specific I'm interested at linux ports. ATM we run linux
>> > ldconfig (using linuxulator) _at package building_. Hence to create a
>> > package for FC6 port we should change compat.linux.osrelease (which I
>> > don't like and try to avoid). If the "ldconfig" stage may be skipped
>> > when PACKAGE_BUILDING is defined then things get way too easier both
>> > for default kernel linux.osrelease and default linux_base port change.
>
>> I don't follow - what is the problem?
>
> An FC6 port can't be build (and more specific -- linux-fc6 ldconfig
> doesn't run) with current default compat.linux.osrelease=2.4.2. So
> this sysctl should be changed to 2.6.16 for package building sake.
> When the default compat.linux.osrelease will be switched to 2.6.16 we
> will get the other way round problem if we try to build and FC4 port.
>
> I don't like the status quo and want to find a way to siplify it.

It's not only a ldconfig problem, it's a generic problem. The gtk  
ports run plugin detection programs (gtk-query-immodules-2.0-32 and  
gdk-pixbuf-query-loaders-32) at installation time.

Do you think it is a problem when the non-default linux port is not  
available as a package? Currently I don't think it is a big problem  
(you can check the value of the sysctl and IGNORE if it is not ok).

When we switch the default, it will be a problem for those releases  
which we still support but which have not the "good" default value for  
the linux emulation (AFAIK pointyhat is running -current with some  
jails for RELENG_x builds). One workaround would be that portmgr sets  
the right value in the jail for the package build for the  
corresponding release. This would be the cleanest solution, as all  
linux ports are then build in the right environment and we don't have  
to add magic code to every linux port (or bsd.port.mk).

Kris, your opinion?

Bye,
Alexander.

-- 
Howe's Law:
	Everyone has a scheme that will not work.

http://www.Leidinger.net    Alexander @ Leidinger.net: PGP ID = B0063FE7
http://www.FreeBSD.org       netchild @ FreeBSD.org  : PGP ID = 72077137