[gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Mikle Kolyada]

Signed-off-by: Mikle Kolyada <zlogene@gentoo.org>
---
As discussed with mgorny at #gentoo-qa
 glep-0081.rst | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/glep-0081.rst b/glep-0081.rst
index 082e705..e5a4db2 100644
--- a/glep-0081.rst
+++ b/glep-0081.rst
@@ -107,6 +107,9 @@ a build time dependency (``DEPEND``) must be used.  If the user is
 needed at install and/or run time, a run time dependency (``RDEPEND``)
 must be used.
 
+Both ``acct-group`` and ``acct-user`` eclasses have the ``KEYWORDS``
+variable predefined, keywords must not be altered locally via ebuilds.
+
 
 Maintaining users/groups
 ------------------------
-- 
2.21.0

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Brian Evans]

[-- Attachment #1.1: Type: text/plain, Size: 1028 bytes --]

On 8/7/2019 11:20 AM, Mikle Kolyada wrote:
> Signed-off-by: Mikle Kolyada <zlogene@gentoo.org>
> ---
> As discussed with mgorny at #gentoo-qa
>  glep-0081.rst | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/glep-0081.rst b/glep-0081.rst
> index 082e705..e5a4db2 100644
> --- a/glep-0081.rst
> +++ b/glep-0081.rst
> @@ -107,6 +107,9 @@ a build time dependency (``DEPEND``) must be used.  If the user is
>  needed at install and/or run time, a run time dependency (``RDEPEND``)
>  must be used.
>  
> +Both ``acct-group`` and ``acct-user`` eclasses have the ``KEYWORDS``
> +variable predefined, keywords must not be altered locally via ebuilds.
> +
>  
>  Maintaining users/groups
>  ------------------------
> 

I object to this as I feel they can incorrect such as on prefix.

Also, this goes against the established practice of committing directly
to stable.  These are exactly the same as virtuals as "they install
nothing" and "just run a script" (to verify dependencies).

Brian


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 834 bytes --]

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Mikle Kolyada]

[-- Attachment #1.1: Type: text/plain, Size: 1358 bytes --]


On 07.08.2019 18:32, Brian Evans wrote:
> On 8/7/2019 11:20 AM, Mikle Kolyada wrote:
>> Signed-off-by: Mikle Kolyada <zlogene@gentoo.org>
>> ---
>> As discussed with mgorny at #gentoo-qa
>>  glep-0081.rst | 3 +++
>>  1 file changed, 3 insertions(+)
>>
>> diff --git a/glep-0081.rst b/glep-0081.rst
>> index 082e705..e5a4db2 100644
>> --- a/glep-0081.rst
>> +++ b/glep-0081.rst
>> @@ -107,6 +107,9 @@ a build time dependency (``DEPEND``) must be used.  If the user is
>>  needed at install and/or run time, a run time dependency (``RDEPEND``)
>>  must be used.
>>  
>> +Both ``acct-group`` and ``acct-user`` eclasses have the ``KEYWORDS``
>> +variable predefined, keywords must not be altered locally via ebuilds.
>> +
>>  
>>  Maintaining users/groups
>>  ------------------------
>>
> I object to this as I feel they can incorrect such as on prefix.
>
> Also, this goes against the established practice of committing directly
> to stable.  These are exactly the same as virtuals as "they install
> nothing" and "just run a script" (to verify dependencies).
>
> Brian


They are not exactly the same as virtuals,  ~arch keywords in virtuals
make sense because they may have dependency providers that are also in
~arch status. These eclasses pull nothing, therefore keywords altering
here meaningless by definition.



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Michał Górny]

[-- Attachment #1: Type: text/plain, Size: 1275 bytes --]

On Wed, 2019-08-07 at 11:32 -0400, Brian Evans wrote:
> On 8/7/2019 11:20 AM, Mikle Kolyada wrote:
> > Signed-off-by: Mikle Kolyada <zlogene@gentoo.org>
> > ---
> > As discussed with mgorny at #gentoo-qa
> >  glep-0081.rst | 3 +++
> >  1 file changed, 3 insertions(+)
> > 
> > diff --git a/glep-0081.rst b/glep-0081.rst
> > index 082e705..e5a4db2 100644
> > --- a/glep-0081.rst
> > +++ b/glep-0081.rst
> > @@ -107,6 +107,9 @@ a build time dependency (``DEPEND``) must be used.  If the user is
> >  needed at install and/or run time, a run time dependency (``RDEPEND``)
> >  must be used.
> >  
> > +Both ``acct-group`` and ``acct-user`` eclasses have the ``KEYWORDS``
> > +variable predefined, keywords must not be altered locally via ebuilds.
> > +
> >  
> >  Maintaining users/groups
> >  ------------------------
> > 
> 
> I object to this as I feel they can incorrect such as on prefix.
> 
> Also, this goes against the established practice of committing directly
> to stable.  These are exactly the same as virtuals as "they install
> nothing" and "just run a script" (to verify dependencies).
> 

We don't do stabilization for virtuals.  We set KEYWORDS to be the union
of KEYWORDS of providers.

-- 
Best regards,
Michał Górny


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 618 bytes --]

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Michał Górny]

[-- Attachment #1: Type: text/plain, Size: 932 bytes --]

On Wed, 2019-08-07 at 18:20 +0300, Mikle Kolyada wrote:
> Signed-off-by: Mikle Kolyada <zlogene@gentoo.org>
> ---
> As discussed with mgorny at #gentoo-qa
>  glep-0081.rst | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/glep-0081.rst b/glep-0081.rst
> index 082e705..e5a4db2 100644
> --- a/glep-0081.rst
> +++ b/glep-0081.rst
> @@ -107,6 +107,9 @@ a build time dependency (``DEPEND``) must be used.  If the user is
>  needed at install and/or run time, a run time dependency (``RDEPEND``)
>  must be used.
>  
> +Both ``acct-group`` and ``acct-user`` eclasses have the ``KEYWORDS``
> +variable predefined, keywords must not be altered locally via ebuilds.
> +
>  
>  Maintaining users/groups
>  ------------------------

I've told you I don't believe this belongs in the GLEP.  It's
implementation detail, and if at all, it should go through tree policy.

-- 
Best regards,
Michał Górny


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 618 bytes --]

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Ulrich Mueller]

[-- Attachment #1: Type: text/plain, Size: 833 bytes --]

>>>>> On Wed, 07 Aug 2019, Mikle Kolyada wrote:
 
> +Both ``acct-group`` and ``acct-user`` eclasses have the ``KEYWORDS``
> +variable predefined, keywords must not be altered locally via ebuilds.
> +
 
GLEP 81 doesn't specify anything about the implementation. In fact,
it neither requires that it is implemented via eclasses, nor does it
specify any ebuild or eclass variables. So, I'd say that keywording
is an implementation detail and doesn't belong in the GLEP.

Plus, the eclasses explicitly allow KEYWORDS to be overridden by the
ebuild:

: ${KEYWORDS:=alpha amd64 arm arm64 hppa ia64 m68k ~mips ppc ppc64 ~riscv s390 sh sparc x86 ~ppc-aix ~x64-cygwin ~amd64-fbsd ~x86-fbsd ~amd64-linux ~x86-linux ~ppc-macos ~x64-macos ~x86-macos ~m68k-mint ~sparc-solaris ~sparc64-solaris ~x64-solaris ~x86-solaris}

Ulrich

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Kent Fredric]

[-- Attachment #1: Type: text/plain, Size: 1808 bytes --]

On Wed, 7 Aug 2019 11:32:43 -0400
Brian Evans <grknight@gentoo.org> wrote:

> I object to this as I feel they can incorrect such as on prefix.
> 
> Also, this goes against the established practice of committing directly
> to stable.  These are exactly the same as virtuals as "they install
> nothing" and "just run a script" (to verify dependencies).

I'll have to disagree: Installing/Upgrading virtuals can have radical
impacts on your system via the dependency graph.

Stabilizing a virtual before all its dependencies get stabilized will
create hell.

Virtuals can also impose USE flags onto other package via the
dependency graph, and all hosts of similar things.

If you wanted an equivalence, you'd need a virtual without
dependencies, which would be a useless virtual.

I think its better to think of the ebuilds in acct/ as more an obtusely
defined text config-file, for which, the interpretation of that text
file is defined by a shared blob of code, the eclass.

The stability of the mechanics of interpreting those text files is what
is in consideration here, not the text files themselves.

And that mechanism is the eclass.

By comparison, you could easily implement an alternative system, where
nodes in acct/ were not .ebuilds, but, just literal text files instead
of bash, with no provisions for keywords at all, and the interpretation
of said files would be up to the package manager.

Not only shouldn't you try to do keywording in such a situation, you
_couldnt_

That said, I do think making this stipulation may be premature, but its
something that I can't imagine we'll know for sure is wise until we've
been employing this strategy for a good time, and we won't know we'll
need the capacity until the first usecase where its needed arises.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Mikle Kolyada]

[-- Attachment #1.1: Type: text/plain, Size: 1222 bytes --]


On 07.08.2019 18:49, Michał Górny wrote:
> On Wed, 2019-08-07 at 18:20 +0300, Mikle Kolyada wrote:
>> Signed-off-by: Mikle Kolyada <zlogene@gentoo.org>
>> ---
>> As discussed with mgorny at #gentoo-qa
>>  glep-0081.rst | 3 +++
>>  1 file changed, 3 insertions(+)
>>
>> diff --git a/glep-0081.rst b/glep-0081.rst
>> index 082e705..e5a4db2 100644
>> --- a/glep-0081.rst
>> +++ b/glep-0081.rst
>> @@ -107,6 +107,9 @@ a build time dependency (``DEPEND``) must be used.  If the user is
>>  needed at install and/or run time, a run time dependency (``RDEPEND``)
>>  must be used.
>>  
>> +Both ``acct-group`` and ``acct-user`` eclasses have the ``KEYWORDS``
>> +variable predefined, keywords must not be altered locally via ebuilds.
>> +
>>  
>>  Maintaining users/groups
>>  ------------------------
> I've told you I don't believe this belongs in the GLEP.  It's
> implementation detail, and if at all, it should go through tree policy.
>
We did not come to a conclusion about where this should be written down
if you remember, in any way I am concerned about the fact, not about a
place where we write it down, the main intention here is that all the
developers must follow the same rules.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Kent Fredric]

[-- Attachment #1: Type: text/plain, Size: 1019 bytes --]

On Wed, 07 Aug 2019 17:55:21 +0200
Ulrich Mueller <ulm@gentoo.org> wrote:

> Plus, the eclasses explicitly allow KEYWORDS to be overridden by the
> ebuild:
> 
> : ${KEYWORDS:=alpha amd64 arm arm64 hppa ia64 m68k ~mips ppc ppc64 ~riscv s390 sh sparc x86 ~ppc-aix ~x64-cygwin ~amd64-fbsd ~x86-fbsd ~amd64-linux ~x86-linux ~ppc-macos ~x64-macos ~x86-macos ~m68k-mint ~sparc-solaris ~sparc64-solaris ~x64-solaris ~x86-solaris}

I guess the clear limitation that the proposed policy would create, is
that it would be illegal for a package on say, riscv, to be stabilized,
if it depended on any acct-* packages.

Because the eclass would say "riscv ain't stable", and policy would say
"you can't change keywords on acct-*", which would necessitate that all
acct-* be marked stable for riscv via changing the eclass, in order for
the riscv package to be marked stable.

( Yes, this is not a risk now, but it would be an annoying situation
for anyone trying to promote riscv into being a stable arch )




[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Mike Gilbert]

On Wed, Aug 7, 2019 at 12:02 PM Kent Fredric <kentnl@gentoo.org> wrote:
>
> On Wed, 07 Aug 2019 17:55:21 +0200
> Ulrich Mueller <ulm@gentoo.org> wrote:
>
> > Plus, the eclasses explicitly allow KEYWORDS to be overridden by the
> > ebuild:
> >
> > : ${KEYWORDS:=alpha amd64 arm arm64 hppa ia64 m68k ~mips ppc ppc64 ~riscv s390 sh sparc x86 ~ppc-aix ~x64-cygwin ~amd64-fbsd ~x86-fbsd ~amd64-linux ~x86-linux ~ppc-macos ~x64-macos ~x86-macos ~m68k-mint ~sparc-solaris ~sparc64-solaris ~x64-solaris ~x86-solaris}
>
> I guess the clear limitation that the proposed policy would create, is
> that it would be illegal for a package on say, riscv, to be stabilized,
> if it depended on any acct-* packages.
>
> Because the eclass would say "riscv ain't stable", and policy would say
> "you can't change keywords on acct-*", which would necessitate that all
> acct-* be marked stable for riscv via changing the eclass, in order for
> the riscv package to be marked stable.

But that's exactly what we would want to happen: the value in the
eclass should be updated, not individual ebuilds.

Re: [gentoo-dev] [PATCH] glep-0081: prohibit KEYWORDS altering

[Brian Evans]

[-- Attachment #1.1: Type: text/plain, Size: 1395 bytes --]

On 8/7/19 11:20 AM, Mikle Kolyada wrote:
> Signed-off-by: Mikle Kolyada <zlogene@gentoo.org>
> ---
> As discussed with mgorny at #gentoo-qa
>  glep-0081.rst | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/glep-0081.rst b/glep-0081.rst
> index 082e705..e5a4db2 100644
> --- a/glep-0081.rst
> +++ b/glep-0081.rst
> @@ -107,6 +107,9 @@ a build time dependency (``DEPEND``) must be used.  If the user is
>  needed at install and/or run time, a run time dependency (``RDEPEND``)
>  must be used.
>  
> +Both ``acct-group`` and ``acct-user`` eclasses have the ``KEYWORDS``
> +variable predefined, keywords must not be altered locally via ebuilds.
> +
>  
>  Maintaining users/groups
>  ------------------------
> 

Maybe I'm not making myself clear (A bad habit of mine).

I want to keep this out such that maintainers can decide the best
keywords and when it is inappropriate to include a keyword.

For example with my acct-{group,user}/mysql ebuilds, I want to remove
the prefix keywords and only have those which match the virtual and
server implementations.  The rare occasion when a keyword is added is no
more a maintenance task like changing the virtual.  Having other ebuilds
outside the project use this package is not the right thing to do.

The packages in question treat prefix differently so why shouldn't it be
reflected here?

Brian


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 981 bytes --]