[Drbd-dev] Re: [DRBD-cvs] r1598 - trunk/user

[Drbd-dev] Re: [DRBD-cvs] r1598 - trunk/user

[Lars Marowsky-Bree]

On 2004-10-19T09:32:24, svn@svn.drbd.org wrote:

> Author: phil
> Date: 2004-10-19 09:32:21 +0200 (Tue, 19 Oct 2004)
> New Revision: 1598
> 
> Modified:
>    trunk/user/Makefile
>    trunk/user/drbdmeta.c
> Log:
> [patch by LGE] 
>  * removed the dependency on glib, and use the kernel's
>    includes instead. 

Do not do that. Please. Including kernel-headers from user-space is an
absolute nightmare, it keeps breaking sporadically and in particular on
non-i386 archs. The kernel headers just aren't meant to be used in
user-space (and are documented to not be supported for this), and all
the build issues I've had with OCFS, Lustre, ... have been caused by
this.

In particular these headers

> +#include <asm/byteorder.h>	/* for the __cpu_to_le64 etc. functions  */
> +#include <linux/bitops.h>	/* for the hweight functions  */
> +#include <linux/types.h>	/* for the __u32/64 type defs */

_are known to break_ on ia64/ppc64/s390x. I'd be very grateful if you
could avoid this.


Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX AG - A Novell company

Re: [Drbd-dev] Re: [DRBD-cvs] r1598 - trunk/user

[Lars Ellenberg]

/ 2004-10-19 11:13:07 +0200
\ Lars Marowsky-Bree:
> On 2004-10-19T09:32:24, svn@svn.drbd.org wrote:
> 
> > Author: phil
> > Date: 2004-10-19 09:32:21 +0200 (Tue, 19 Oct 2004)
> > New Revision: 1598
> > 
> > Modified:
> >    trunk/user/Makefile
> >    trunk/user/drbdmeta.c
> > Log:
> > [patch by LGE] 
> >  * removed the dependency on glib, and use the kernel's
> >    includes instead. 
> 
> Do not do that. Please. Including kernel-headers from user-space is an
> absolute nightmare, it keeps breaking sporadically and in particular on
> non-i386 archs. The kernel headers just aren't meant to be used in
> user-space (and are documented to not be supported for this), and all
> the build issues I've had with OCFS, Lustre, ... have been caused by
> this.
> 
> In particular these headers
> 
> > +#include <asm/byteorder.h>	/* for the __cpu_to_le64 etc. functions  */
> > +#include <linux/bitops.h>	/* for the hweight functions  */
> > +#include <linux/types.h>	/* for the __u32/64 type defs */
> 
> _are known to break_ on ia64/ppc64/s390x. I'd be very grateful if you
> could avoid this.

hm.

I just don't want to depend on glib-devel just for some endianness issues.
rather copy waht is neccessary into our own headers.
which is not too much, anyways.

#  linux/bitops.h
we only care for the hweight functions, which are plain c.
I don't see where they could break on any arch.

# linux/types.h
we only care for the __u32 __u64 type defines.
ok, we can drop that and just use sys/types.h,
%s/__u\(.*\)/uint\1_t/

# asm/byteorder.h
ok, we can drop that, and use utils.h instead.
we'd need to define *64 variants of swap and *_to_* functions, though.

konsensfähig?

	lge

Re: [Drbd-dev] Re: [DRBD-cvs] r1598 - trunk/user

[Lars Ellenberg]

/ 2004-10-19 12:19:18 +0200
\ Lars Ellenberg:
> > In particular these headers
> > 
> > > +#include <asm/byteorder.h>	/* for the __cpu_to_le64 etc. functions  */
> > > +#include <linux/bitops.h>	/* for the hweight functions  */
> > > +#include <linux/types.h>	/* for the __u32/64 type defs */
> > 
> > _are known to break_ on ia64/ppc64/s390x. I'd be very grateful if you
> > could avoid this.
> 
> hm.
> 
> I just don't want to depend on glib-devel just for some endianness issues.
> rather copy waht is neccessary into our own headers.
> which is not too much, anyways.
> 
> #  linux/bitops.h
> we only care for the hweight functions, which are plain c.
> I don't see where they could break on any arch.
> 
> # linux/types.h
> we only care for the __u32 __u64 type defines.
> ok, we can drop that and just use sys/types.h,
> %s/__u\(.*\)/uint\1_t/
> 
> # asm/byteorder.h
> ok, we can drop that, and use utils.h instead.

nonsense.
that is a nonstandard header, too.
well, then I'd just copy_n_paste together our own endian header?

> konsensfähig?

	lge

Re: [Drbd-dev] Re: [DRBD-cvs] r1598 - trunk/user

[Philipp Reisner]

> nonsense.
> that is a nonstandard header, too.
> well, then I'd just copy_n_paste together our own endian header?
>

good idea!

-Philipp
-- 
: Dipl-Ing Philipp Reisner                      Tel +43-1-8178292-50 :
: LINBIT Information Technologies GmbH          Fax +43-1-8178292-82 :
: Schönbrunnerstr 244, 1120 Vienna, Austria    http://www.linbit.com :