FW: dpkg support for Solaris
 

I don't think these were received by the mailing list so I'm sending again. *Sorry for any duplicates.
From: Andrew Stormont <andrew.stormont@nexenta.com>
Date: Thu, 10 Nov 2011 23:47:14 +0000
To: <debian-dpkg@lists.debian.org>
Subject: Re: dpkg support for Solaris

Hi,
I've also attached a patch which adds support for Solaris zones. *I imagine this won't be the sort of thing you can apply is-is like the other patch.
Thanks,Andy
From: Andrew Stormont <andrew.stormont@nexenta.com>
Date: Thu, 10 Nov 2011 14:03:18 +0000
To: <debian-dpkg@lists.debian.org>
Subject: dpkg support for Solaris

Hi,
Please could you apply the attached patch which fixes dpkg Support for Solaris.
Thanks,Andy

FW: dpkg support for Solaris
 

Hello,

On Mon, 14 Nov 2011, Andrew Stormont wrote:
> From: Andrew Stormont <andrew.stormont@nexenta.com>
> Date: Thu, 10 Nov 2011 14:03:18 +0000
> To: <debian-dpkg@lists.debian.org>
> Subject: dpkg support for Solaris
>
> Hi,
>
> Please could you apply the attached patch which fixes dpkg Support for
> Solaris.

It would help if you justified the changes... I'm not familiar enough with
Solaris/Nexenta to be able to judge the patch. So I would like some sort
of rationale... (if you sent proper commits generated with git
format-patch that would have been taken care of — provided that you write
good commit logs)

Let's skim through the patch for a quick review. Note however that I'm not
familiar with portability issues and some of my remarks might be
ill-advised. Feel free to point it out if you notice something odd.

> diff --git a/dselect/main.cc b/dselect/main.cc
> index 6508b1f..5b1c08b 100644
> --- a/dselect/main.cc
> +++ b/dselect/main.cc
> @@ -4,6 +4,7 @@
> *
> * Copyright © 1994,1995 Ian Jackson <ian@chiark.greenend.org.uk>
> * Copyright © 2000,2001 Wichert Akkerman <wakkerma@debian.org>
> + * Copyright © 2011 Nexenta Systems Inc. All rights reserved.

Please drop those "All rights reserved". It doesn't mix well with the
definition of the GPL...

> @@ -45,6 +46,7 @@
> #elif defined(HAVE_NCURSES_TERM_H)
> #include <ncurses/term.h>
> #else
> +#include <curses.h>
> #include <term.h>
> #endif

What does curses.h define that was missing?

> @@ -53,6 +55,10 @@
> #include <dpkg/dpkg-db.h>
> #include <dpkg/options.h>
>
> +#ifndef ERR
> +#define ERR (-1)
> +#endif
> +
> #include "dselect.h"
> #include "bindings.h"
> #include "pkglist.h"

On my Linux system ERR is defined by <curses.h>. Why is that needed on
Solaris when you already added curses.h?

> diff --git a/lib/dpkg/md5.c b/lib/dpkg/md5.c
> index 3da18c9..5e9f311 100644
> --- a/lib/dpkg/md5.c
> +++ b/lib/dpkg/md5.c
> @@ -15,6 +15,8 @@
> * MD5Context structure, pass it to MD5Init, call MD5Update as
> * needed on buffers full of bytes, and then call MD5Final, which
> * will fill a supplied 16-byte array with the digest.
> + *
> + * Copyright © 2011 Nexenta Systems Inc. All rights reserved.
> */

That file is in the public domain and it's best if we keep it that way, so
please accept the same and don't claim any copyright on it.

> @@ -41,7 +43,7 @@
> (cp)[1] = (value) >> 8;
> (cp)[0] = (value); } while (0)
>
> -static u_int8_t PADDING[MD5_BLOCK_LENGTH] = {
> +static uint8_t PADDING[MD5_BLOCK_LENGTH] = {
> 0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0

Hum, C99 is not a requirement to build dpkg. Some features are required
but those standard types are currently not part of it (see README and
doc/coding-style.txt). So maybe it's better to add the required typedefs
specifically for Solaris?

That said I don't really know why Guillem did not mandate C99 in its
entirety.

> diff --git a/lib/dpkg/md5.h b/lib/dpkg/md5.h
> index 731a2af..5f3a869 100644
> --- a/lib/dpkg/md5.h
> +++ b/lib/dpkg/md5.h
> @@ -10,27 +10,33 @@
> * This code has been tested against that, and is equivalent,
> * except that you don't need to include two pages of legalese
> * with every copy.
> + *
> + * Copyright © 2011 Nexenta Systems Inc. All rights reserved.
> */
>
> #ifndef _MD5_H_
> #define _MD5_H_
>
> +#include <config.h>
> +

An header file should not include <config.h> (even if that particular
header file is not a public one).

> +#ifdef HAVE_SYS_CDEFS
> #include <sys/cdefs.h>
> +#endif

So this test should probably be changed into something else. Not sure
what though... this header is provided by glibc but is not glibc specific
apparently.

If we can't find anything better, we could go with this I guess:
#if !defined(__sun)
#include <sys/cdefs.h>
#endif

> @@ -31,6 +33,7 @@
> # define OSHurd
> #elif defined(__sun)
> # define OSsunos
> +# undef HAVE_KVM_H
> #elif defined(OPENBSD) || defined(__OpenBSD__)
> # define OSOpenBSD
> #elif defined(hpux)

Why? Does kvm.h exist on Solaris and is it something totally unrelated?

> +#if defined(OSsunos)
> + sprintf(buf, "/proc/%d/status", pid);
> +#else
> sprintf(buf, "/proc/%d/stat", pid);
> +#endif

Does /proc/<pid>/status on Solaris follow the same structure than the
/proc/<pid>/stat on Linux, that is having the executable name
between parenthesis somewhere in the string?

Cheers,
--
Raphaël Hertzog ◈ Debian Developer

Pre-order a copy of the Debian Administrator's Handbook and help
liberate it: http://debian-handbook.info/go/ulule-rh/


--
To UNSUBSCRIBE, email to debian-dpkg-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20111115142246.GC17899@rivendell.home.ouaza.com">h ttp://lists.debian.org/20111115142246.GC17899@rivendell.home.ouaza.com

FW: dpkg support for Solaris
 

Hi!

On Mon, 2011-11-14 at 15:45:54 +0000, Andrew Stormont wrote:
> From: Andrew Stormont <andrew.stormont@nexenta.com>
> Date: Thu, 10 Nov 2011 23:47:14 +0000
> To: <debian-dpkg@lists.debian.org>
> Subject: Re: dpkg support for Solaris
>
> I've also attached a patch which adds support for Solaris zones. I imagine
> this won't be the sort of thing you can apply is-is like the other patch.

So, AFAIUI Solaris zones are a bit like BSD jails, and the intent I
gather from the patch is that it's not unpacking when inside a zone,
because the data has been unpacked elsewhere (in the global zone)?

If this is the case then I think it's a bad idea, similar to user
installable packages, because the databases and the unpacked data
will get out of sync with each other.


> From: Andrew Stormont <andrew.stormont@nexenta.com>
> Date: Thu, 10 Nov 2011 14:03:18 +0000
> To: <debian-dpkg@lists.debian.org>
> Subject: dpkg support for Solaris

> Please could you apply the attached patch which fixes dpkg Support for
> Solaris.

I've pushed fixes for several of these issues.

> diff --git a/dselect/main.cc b/dselect/main.cc
> index 6508b1f..5b1c08b 100644
> --- a/dselect/main.cc
> +++ b/dselect/main.cc
> @@ -53,6 +55,10 @@
> #include <dpkg/dpkg-db.h>
> #include <dpkg/options.h>
>
> +#ifndef ERR
> +#define ERR (-1)
> +#endif
> +
> #include "dselect.h"
> #include "bindings.h"
> #include "pkglist.h"

This should not be needed, given that Solaris should be conforming to
either the SVr4 or XPG4 curses, which define that the return value for
error for several functions is ERR.

> diff --git a/utils/start-stop-daemon.c b/utils/start-stop-daemon.c
> index 03a7b57..6f62a42 100644
> --- a/utils/start-stop-daemon.c
> +++ b/utils/start-stop-daemon.c
> @@ -31,6 +33,7 @@
> # define OSHurd
> #elif defined(__sun)
> # define OSsunos
> +# undef HAVE_KVM_H
> #elif defined(OPENBSD) || defined(__OpenBSD__)
> # define OSOpenBSD
> #elif defined(hpux)
> @@ -1039,7 +1042,7 @@ pid_is_exec(pid_t pid, const char *name)
> }
> #endif
>
> -#if defined(OSLinux)
> +#if defined(OSLinux) || defined(OSsunos)
> static bool
> pid_is_user(pid_t pid, uid_t uid)
> {
> @@ -1095,7 +1098,7 @@ pid_is_user(pid_t pid, uid_t uid)
> }
> #endif
>
> -#if defined(OSLinux)
> +#if defined(OSLinux) || defined(OSsunos)
> static bool
> pid_is_cmd(pid_t pid, const char *name)
> {
> @@ -1103,7 +1106,11 @@ pid_is_cmd(pid_t pid, const char *name)
> FILE *f;
> int c;
>
> +#if defined(OSsunos)
> + sprintf(buf, "/proc/%d/status", pid);
> +#else
> sprintf(buf, "/proc/%d/stat", pid);
> +#endif
> f = fopen(buf, "r");
> if (!f)
> return false;

These are not right, the Solaris proc(4) layout is not documented to be
compatible with the one on Linux.

I've some local changes fixing the kvm support for at least FreeBSD
and Mac OS X, which should help for Solaris too, but will polish and
push only after having dealt with the multiarch merge.

thanks,
guillem


--
To UNSUBSCRIBE, email to debian-dpkg-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20111116035020.GA16088@gaara.hadrons.org">http://lists.debian.org/20111116035020.GA16088@gaara.hadrons.org

FW: dpkg support for Solaris
 

On Tue, 2011-11-15 at 15:22:46 +0100, Raphael Hertzog wrote:
> On Mon, 14 Nov 2011, Andrew Stormont wrote:
> > diff --git a/lib/dpkg/md5.c b/lib/dpkg/md5.c
> > index 3da18c9..5e9f311 100644
> > --- a/lib/dpkg/md5.c
> > +++ b/lib/dpkg/md5.c
> > @@ -15,6 +15,8 @@
> > * MD5Context structure, pass it to MD5Init, call MD5Update as
> > * needed on buffers full of bytes, and then call MD5Final, which
> > * will fill a supplied 16-byte array with the digest.
> > + *
> > + * Copyright © 2011 Nexenta Systems Inc. All rights reserved.
> > */
>
> That file is in the public domain and it's best if we keep it that way, so
> please accept the same and don't claim any copyright on it.

This does not matter any more given the pushed changes, but in
addition I don't think these changes are copyrightable, as they are
just a symbol rename (at least according to the GNU maintainers doc).

> Hum, C99 is not a requirement to build dpkg. Some features are required
> but those standard types are currently not part of it (see README and
> doc/coding-style.txt). So maybe it's better to add the required typedefs
> specifically for Solaris?

They are assumed to be present, and checked by dpkg-compiler.m4. Those
are not on the doc, because they don't really need compiler support,
and can be easily mapped to other types by configure.

> That said I don't really know why Guillem did not mandate C99 in its
> entirety.

Because C99 is not yet fully implemented by many compilers (not even gcc).

> > +#ifdef HAVE_SYS_CDEFS
> > #include <sys/cdefs.h>
> > +#endif
>
> So this test should probably be changed into something else. Not sure
> what though... this header is provided by glibc but is not glibc specific
> apparently.
>
> If we can't find anything better, we could go with this I guess:
> #if !defined(__sun)
> #include <sys/cdefs.h>
> #endif

It's a BSDism and it's not needed.

> > @@ -31,6 +33,7 @@
> > # define OSHurd
> > #elif defined(__sun)
> > # define OSsunos
> > +# undef HAVE_KVM_H
> > #elif defined(OPENBSD) || defined(__OpenBSD__)
> > # define OSOpenBSD
> > #elif defined(hpux)
>
> Why? Does kvm.h exist on Solaris and is it something totally unrelated?

kvm implementations vary slightly from system to system, given that
they expose kernel internal structures to user-land.

regards,
guillem


--
To UNSUBSCRIBE, email to debian-dpkg-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20111116040050.GB16088@gaara.hadrons.org">http://lists.debian.org/20111116040050.GB16088@gaara.hadrons.org