[parisc-linux] gcc-3.3 configuration

[parisc-linux] gcc-3.3 configuration

[Matthias Klose]

For Debian GNU/Linux gcc-3.3 is currently configured with

	--with-sjlj-exceptions

to allow a binary compatible upgrade from gcc-3.2 to gcc-3.3. The
other Debian platform, where sjlj based exceptions changed to dwarf2
based exceptions is m68k, therefore the CC.

As there are no hppa/m68k distributions released using gcc-3.2,
compatibility with other released distros is not an issue. OTOH, I see
Gentoo and probably other distros built for hppa, first starting with
gcc-3.3, and maybe configured the standard way.

Is it worth (or needed) to strive for binary compatibility of hppa
based distributions, based on gcc-3.3 and glibc-2.3.x?

If yes, does the following approach work: Build a libstdc++5-3.3,
conflicting with libstdc++5 (including libstdc++ with the same soname)
and doing binary uploads for hppa for all libstdc++5 dependent
packages? (The same could be done for m68k, but the compatibility
argument doesn't hold as Debian seems to be the only m68k based
distro).

	Matthias

Re: [parisc-linux] gcc-3.3 configuration

[Matthew Wilcox]

On Wed, May 21, 2003 at 01:42:53AM +0200, Matthias Klose wrote:
> For Debian GNU/Linux gcc-3.3 is currently configured with
> 
> 	--with-sjlj-exceptions
> 
> to allow a binary compatible upgrade from gcc-3.2 to gcc-3.3. The
> other Debian platform, where sjlj based exceptions changed to dwarf2
> based exceptions is m68k, therefore the CC.

*sigh*.  I'm very conflicted over this.  On one hand, Debian is the de-facto
standard for PA-RISC/Linux binaries.  If another distro cares about
binary compatibility, they should follow Debian.

On the other hand, dwarf exceptions are clearly superior to sjlj.
We've never _released_ a distro compiled with 3.2 so we can break the
binary ABI without major repurcussions.

I'm not sure it's my call to make; I can see arguments on both sides.

-- 
"It's not Hollywood.  War is real, war is primarily not about defeat or
victory, it is about death.  I've seen thousands and thousands of dead bodies.
Do you think I want to have an academic debate on this subject?" -- Robert Fisk

Re: [parisc-linux] gcc-3.3 configuration

[Matthew Wilcox]

On Wed, May 21, 2003 at 01:42:53AM +0200, Matthias Klose wrote:
> For Debian GNU/Linux gcc-3.3 is currently configured with
> 
> 	--with-sjlj-exceptions
> 
> to allow a binary compatible upgrade from gcc-3.2 to gcc-3.3. The
> other Debian platform, where sjlj based exceptions changed to dwarf2
> based exceptions is m68k, therefore the CC.

*sigh*.  I'm very conflicted over this.  On one hand, Debian is the de-facto
standard for PA-RISC/Linux binaries.  If another distro cares about
binary compatibility, they should follow Debian.

On the other hand, dwarf exceptions are clearly superior to sjlj.
We've never _released_ a distro compiled with 3.2 so we can break the
binary ABI without major repurcussions.

I'm not sure it's my call to make; I can see arguments on both sides.

-- 
"It's not Hollywood.  War is real, war is primarily not about defeat or
victory, it is about death.  I've seen thousands and thousands of dead bodies.
Do you think I want to have an academic debate on this subject?" -- Robert Fisk

Re: [parisc-linux] gcc-3.3 configuration

[Alan Cox]

On Mer, 2003-05-21 at 01:08, Matthew Wilcox wrote:
> I'm not sure it's my call to make; I can see arguments on both sides.

Thats at least one of the reasons. Reputation capital is a wonderous
thing. Accept reality, you are the Linus of parisc Linux like it or not
8)

Re: [parisc-linux] gcc-3.3 configuration

[John David Anglin]

> On Mer, 2003-05-21 at 01:08, Matthew Wilcox wrote:
> > I'm not sure it's my call to make; I can see arguments on both sides.
> 
> Thats at least one of the reasons. Reputation capital is a wonderous
> thing. Accept reality, you are the Linus of parisc Linux like it or not
> 8)

I agree.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)

Re: [parisc-linux] gcc-3.3 configuration

[John David Anglin]

> On Mer, 2003-05-21 at 01:08, Matthew Wilcox wrote:
> > I'm not sure it's my call to make; I can see arguments on both sides.
> 
> Thats at least one of the reasons. Reputation capital is a wonderous
> thing. Accept reality, you are the Linus of parisc Linux like it or not
> 8)

I agree.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)

Re: [parisc-linux] gcc-3.3 configuration

[Alan Cox]

On Mer, 2003-05-21 at 01:08, Matthew Wilcox wrote:
> I'm not sure it's my call to make; I can see arguments on both sides.

Thats at least one of the reasons. Reputation capital is a wonderous
thing. Accept reality, you are the Linus of parisc Linux like it or not
8)

Re: [parisc-linux] gcc-3.3 configuration

[Grant Grundler]

On Wed, May 21, 2003 at 01:08:54AM +0100, Matthew Wilcox wrote:
> I'm not sure it's my call to make; I can see arguments on both sides.

Having no clue about dwarf vs sjlj (or whatever), I have to trust
*someone* to make the right call. If it's clearly the right direction,
then figure out how to get there without breaking everything and
make use of your "reputation capital".

grant

Re: [parisc-linux] gcc-3.3 configuration

[Grant Grundler]

On Wed, May 21, 2003 at 01:08:54AM +0100, Matthew Wilcox wrote:
> I'm not sure it's my call to make; I can see arguments on both sides.

Having no clue about dwarf vs sjlj (or whatever), I have to trust
*someone* to make the right call. If it's clearly the right direction,
then figure out how to get there without breaking everything and
make use of your "reputation capital".

grant

Re: [parisc-linux] gcc-3.3 configuration

[Randolph Chung]

> If yes, does the following approach work: Build a libstdc++5-3.3,
> conflicting with libstdc++5 (including libstdc++ with the same soname)
> and doing binary uploads for hppa for all libstdc++5 dependent
> packages? (The same could be done for m68k, but the compatibility
> argument doesn't hold as Debian seems to be the only m68k based
> distro).

Aside from the evilness of doing binNMUs of this magnitude, I doubt a
"transition" that doesn't change the SONAME will work. As soon as the
new libstdc++ is installed, every c++ app on the box will instantly
break. This means if anything happens e.g. to apt during the update, the
system will get into a very nasty state. I don't think this is worth it.

If we need to change the SONAME, then there's no binary compatibility
anyway...

As willy mentioned in his other email, if other distros want binary
compatibility, they should follow what Debian is doing (on hppa), since
Debian was there first :-)

my 2 cents,
randolph
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

Re: [parisc-linux] gcc-3.3 configuration

[Randolph Chung]

> If yes, does the following approach work: Build a libstdc++5-3.3,
> conflicting with libstdc++5 (including libstdc++ with the same soname)
> and doing binary uploads for hppa for all libstdc++5 dependent
> packages? (The same could be done for m68k, but the compatibility
> argument doesn't hold as Debian seems to be the only m68k based
> distro).

Aside from the evilness of doing binNMUs of this magnitude, I doubt a
"transition" that doesn't change the SONAME will work. As soon as the
new libstdc++ is installed, every c++ app on the box will instantly
break. This means if anything happens e.g. to apt during the update, the
system will get into a very nasty state. I don't think this is worth it.

If we need to change the SONAME, then there's no binary compatibility
anyway...

As willy mentioned in his other email, if other distros want binary
compatibility, they should follow what Debian is doing (on hppa), since
Debian was there first :-)

my 2 cents,
randolph
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

Re: [parisc-linux] gcc-3.3 configuration

[Carlos O'Donell]

> Aside from the evilness of doing binNMUs of this magnitude, I doubt a
> "transition" that doesn't change the SONAME will work. As soon as the
> new libstdc++ is installed, every c++ app on the box will instantly
> break. This means if anything happens e.g. to apt during the update, the
> system will get into a very nasty state. I don't think this is worth it.
> 
> If we need to change the SONAME, then there's no binary compatibility
> anyway...
> 
> As willy mentioned in his other email, if other distros want binary
> compatibility, they should follow what Debian is doing (on hppa), since
> Debian was there first :-)

*Plants flag here*

c.

Re: [parisc-linux] gcc-3.3 configuration

[Carlos O'Donell]

> Aside from the evilness of doing binNMUs of this magnitude, I doubt a
> "transition" that doesn't change the SONAME will work. As soon as the
> new libstdc++ is installed, every c++ app on the box will instantly
> break. This means if anything happens e.g. to apt during the update, the
> system will get into a very nasty state. I don't think this is worth it.
> 
> If we need to change the SONAME, then there's no binary compatibility
> anyway...
> 
> As willy mentioned in his other email, if other distros want binary
> compatibility, they should follow what Debian is doing (on hppa), since
> Debian was there first :-)

*Plants flag here*

c.

Re: [parisc-linux] gcc-3.3 configuration

[Nicholas Leippe]

On Tuesday 20 May 2003 11:23 pm, Randolph Chung wrote:
> > If yes, does the following approach work: Build a libstdc++5-3.3,
> > conflicting with libstdc++5 (including libstdc++ with the same soname)
> > and doing binary uploads for hppa for all libstdc++5 dependent
> > packages? (The same could be done for m68k, but the compatibility
> > argument doesn't hold as Debian seems to be the only m68k based
> > distro).
>
> Aside from the evilness of doing binNMUs of this magnitude, I doubt a
> "transition" that doesn't change the SONAME will work. As soon as the
> new libstdc++ is installed, every c++ app on the box will instantly
> break. This means if anything happens e.g. to apt during the update, the
> system will get into a very nasty state. I don't think this is worth it.
>

How many critical apps are actually C++?  Is the apt suite itself?
As long as /sbin/* is C, I don't see this as a showstopper for migrating to 
using a superior mechanism, which imo, should be a big goal--the sooner it's  
taken care of the less problematic it will be...

just a thought...

Nick

[parisc-linux] gcc-3.3 configuration

[Matthias Klose]

For Debian GNU/Linux gcc-3.3 is currently configured with

	--with-sjlj-exceptions

to allow a binary compatible upgrade from gcc-3.2 to gcc-3.3. The
other Debian platform, where sjlj based exceptions changed to dwarf2
based exceptions is m68k, therefore the CC.

As there are no hppa/m68k distributions released using gcc-3.2,
compatibility with other released distros is not an issue. OTOH, I see
Gentoo and probably other distros built for hppa, first starting with
gcc-3.3, and maybe configured the standard way.

Is it worth (or needed) to strive for binary compatibility of hppa
based distributions, based on gcc-3.3 and glibc-2.3.x?

If yes, does the following approach work: Build a libstdc++5-3.3,
conflicting with libstdc++5 (including libstdc++ with the same soname)
and doing binary uploads for hppa for all libstdc++5 dependent
packages? (The same could be done for m68k, but the compatibility
argument doesn't hold as Debian seems to be the only m68k based
distro).

	Matthias