[gentoo-dev] Arch-specific patches

[gentoo-dev] Arch-specific patches

[Aron Griffis]

[-- Attachment #1: Type: text/plain, Size: 2505 bytes --]

On a similar topic to the -fPIC question, I've realized today that 

    if use alpha; then
        epatch ${FILESDIR}/${P}-alpha.patch || die "epatch failed"
    fi

is really the wrong thing (and I'm not talking about the die part, which
I know is controversial).  I'm talking about the conditional application
of the patch.

I think this should really be:

    # Patch for 64-bit but apply everywhere
    epatch ${FILESDIR}/${P}-alpha.patch || die "epatch failed"

The reason is twofold:

    1. Devs presently break packages by bumping the version without
       reworking the patch.  But they don't KNOW they broke the package
       because they're working on a different architecture.
       Consequently, the arch teams get a bug report that just ends up
       being busy-work.

    2. We shouldn't be applying patches that are only "safe" for one
       architecture.  If possible (and I know this is a little more
       work), we should be writing patches that correct the problem
       without hurting other arches.  This practice would give us an
       easier time pushing patches upstream.

So what happens if a dev bumps the version on a package, and the patch
no longer applies?  Here's what I think:

    1. The dev should attempt to rework the patch, or determine if it
       can be removed.

    2. If it's too involved, then the dev should remove the arch or
       ~arch keyword from the ebuild, comment out the patch, and send
       email to the appropriate arch team.  That keeps users from
       attempting to update their systems only to have emerge fail in
       the middle for their arch.  Better to have the devs handle the
       problem than push it on to the users.

Does this make sense to everybody?  I'm pushing for this approach
because our current situation truly has a scalability problem.  We have
more devs available for working on packages than we have for working on
alternative architectures (where alternative means non-x86).  It's
really important that devs responsible for packages attempt to keep
their ebuilds cross-platform friendly and not rely on the arch teams to
come along and clean up afterward.

I'm not suggesting that anybody has been careless, but I am suggesting a
change in perspective on Gentoo's multi-arch support...

Thoughts?

Aron

-- 
Aron Griffis
Gentoo Linux Developer (alpha / ia64 / ruby / vim)
Key fingerprint = E3B6 8734 C2D6 B5E5 AE76  FB3A 26B1 C5E3 2010 4EB0


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Arch-specific patches

[Aron Griffis]

[-- Attachment #1: Type: text/plain, Size: 708 bytes --]

Brad House wrote:	[Wed Dec 10 2003, 09:00:12PM EST]
> So before you do this throughout, you must study the patch
> and make sure there is no way it can adversely affect
> other arches.

Right, that was the point of:

> >    2. We shouldn't be applying patches that are only "safe" for one
> >       architecture.  If possible (and I know this is a little more
> >       work), we should be writing patches that correct the problem
> >       without hurting other arches.  This practice would give us an
> >       easier time pushing patches upstream.

-- 
Aron Griffis
Gentoo Linux Developer (alpha / ia64 / ruby / vim)
Key fingerprint = E3B6 8734 C2D6 B5E5 AE76  FB3A 26B1 C5E3 2010 4EB0


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Arch-specific patches

[Seemant Kulleen]

[-- Attachment #1: Type: text/plain, Size: 1589 bytes --]

On Wed, 2003-12-10 at 17:49, Aron Griffis wrote:
> On a similar topic to the -fPIC question, I've realized today that 
> 
>     if use alpha; then
>         epatch ${FILESDIR}/${P}-alpha.patch || die "epatch failed"
>     fi
> 
> is really the wrong thing (and I'm not talking about the die part, which
> I know is controversial).  I'm talking about the conditional application
> of the patch.
> 
> I think this should really be:
> 
>     # Patch for 64-bit but apply everywhere
>     epatch ${FILESDIR}/${P}-alpha.patch || die "epatch failed"
> 
Hi Aron,

Here's my take on this (slightly off-topic).  The way Azarah designed
epatch to be used can be seen very specifically in the xfree tarballs. 
So, if we decide to implement a *STANDARD* gentoo naming scheme and
locations for all our patches then it would like:

xxx_arch_${P*}-description.patch

xxx = number -- determines order in which it is applied
${P*} can be either ${PN} if the patch applies well enough to all
versions, ${P} if specifically for one, or ${PF} if it's just for this
revision of the ebuild (though epatch doesn't know about that, so that's
thorny I guess).

However, I agree 384% with you about making patches that can be applied
across all architectures without harming or otherwise affecting the
non-this ones.

Ciao,
-- 
Seemant Kulleen
Developer and Project Co-ordinator,
Gentoo Linux					http://dev.gentoo.org/~seemant

Public Key: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x3458780E
Key fingerprint = 23A9 7CB5 9BBB 4F8D 549B 6593 EDA2 65D8 3458 780E


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Arch-specific patches

[Donnie Berkholz]

[-- Attachment #1: Type: text/plain, Size: 1025 bytes --]

On Wed, 2003-12-10 at 21:55, Seemant Kulleen wrote:
> Here's my take on this (slightly off-topic).  The way Azarah designed
> epatch to be used can be seen very specifically in the xfree tarballs. 
> So, if we decide to implement a *STANDARD* gentoo naming scheme and
> locations for all our patches then it would like:
> 
> xxx_arch_${P*}-description.patch
> 
> xxx = number -- determines order in which it is applied
> ${P*} can be either ${PN} if the patch applies well enough to all
> versions, ${P} if specifically for one, or ${PF} if it's just for this
> revision of the ebuild (though epatch doesn't know about that, so that's
> thorny I guess).
> 
> However, I agree 384% with you about making patches that can be applied
> across all architectures without harming or otherwise affecting the
> non-this ones.

If we really want to enforce this, we should deprecate and then remove
the _arch_ section of the syntax for bulk patching so anyone using it is
forced to have clean multi-arch patches.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Arch-specific patches

[Aron Griffis]

[-- Attachment #1: Type: text/plain, Size: 1648 bytes --]

Hi Seemant,

Seemant Kulleen wrote:	[Wed Dec 10 2003, 09:55:36PM EST]
> Here's my take on this (slightly off-topic).  The way Azarah designed
> epatch to be used can be seen very specifically in the xfree tarballs. 
> So, if we decide to implement a *STANDARD* gentoo naming scheme and
> locations for all our patches then it would like:

Having just finished some work on the xfree patches for ia64, I'm
familiar with the scheme.  In fact, it's part of my motivation for
writing the original email.

The xfree ebuilds have a few of _ia64_ patches.  For example,
xfree-4.3.99.901 has the following

    5105_ia64_4.2.99.901-ati-radeon-pagesize.patch
    5150_ia64_4.3.0-radeon-preint10.patch
    5350_ia64_4.2.99.901-hp-nv-memory-barrier.patch
    7000_ia64_4.1.0-hp-vgaHW-memory-barrier2.patch

I have read through each one of these and determined that every one
could be changed from "ia64" to "all".  That would greatly decrease the
burden on the ia64 team by making the xfree team verify the patches
still apply when the version is bumped.

Naturally there should still be a comment at the top of the patch to
label it as an ia64-fixer so that if there is a real problem with it,
the xfree team knows (1) who to contact, (2) which arch to disable for
the ebuild, that kind of thing.

> However, I agree 384% with you about making patches that can be
> applied across all architectures without harming or otherwise
> affecting the non-this ones.

Good! :-)

Aron

-- 
Aron Griffis
Gentoo Linux Developer (alpha / ia64 / ruby / vim)
Key fingerprint = E3B6 8734 C2D6 B5E5 AE76  FB3A 26B1 C5E3 2010 4EB0


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-dev] Arch-specific patches

[Aron Griffis]

[-- Attachment #1: Type: text/plain, Size: 484 bytes --]

Donnie Berkholz wrote:	[Wed Dec 10 2003, 10:34:57PM EST]
> If we really want to enforce this, we should deprecate and then remove
> the _arch_ section of the syntax for bulk patching so anyone using it is
> forced to have clean multi-arch patches.

I'd rather call this a recommendation rather than jumping directly to
enforcement.

-- 
Aron Griffis
Gentoo Linux Developer (alpha / ia64 / ruby / vim)
Key fingerprint = E3B6 8734 C2D6 B5E5 AE76  FB3A 26B1 C5E3 2010 4EB0


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]