[gentoo-dev] Testing is not a valid reason to package.mask

[gentoo-dev] Testing is not a valid reason to package.mask

[Jeroen Roovers]

# Gen 2 Developer <someone@gentoo.org> (`date`)
# Masked for testing.
>=rofl-cat/omgpkg-ver


Please people,


   if you want to get something tested, then don't mask it. If you find
that you cannot commit an ebuild because of badly keyworded
dependencies, then drop the relevant keywords and file a bug report
with a KEYWORDREQ.


Kind regards,
     JeR

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Jeroen Roovers]

On Thu, 2 Oct 2008 22:24:35 +0200
Jeroen Roovers <jer@gentoo.org> wrote:

> # Gen 2 Developer <someone@gentoo.org> (`date`)
> # Masked for testing.
> >=rofl-cat/omgpkg-ver
> 
> 
> Please people,
> 
> 
>    if you want to get something tested, then don't mask it. If you
> find that you cannot commit an ebuild because of badly keyworded
> dependencies, then drop the relevant keywords and file a bug report
> with a KEYWORDREQ.

Lest I forget, the exception being that a particular version should
never ever go stable, in which case the masking reason should still be
different. In that case you would still not mark it as "masked for
testing" - what I wanted to clarify is that the mask reason isn't valid
if you want stuff to get tested, as it prevents exactly that from
happening.


Kind regards,
     JeR

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Jeremy Olexa]

On Thu, Oct 2, 2008 at 3:30 PM, Jeroen Roovers <jer@gentoo.org> wrote:
> On Thu, 2 Oct 2008 22:24:35 +0200
> Jeroen Roovers <jer@gentoo.org> wrote:
>
>> # Gen 2 Developer <someone@gentoo.org> (`date`)
>> # Masked for testing.
>> >=rofl-cat/omgpkg-ver
>>
>>
>> Please people,
>>
>>
>>    if you want to get something tested, then don't mask it. If you
>> find that you cannot commit an ebuild because of badly keyworded
>> dependencies, then drop the relevant keywords and file a bug report
>> with a KEYWORDREQ.
>
> Lest I forget, the exception being that a particular version should
> never ever go stable, in which case the masking reason should still be
> different. In that case you would still not mark it as "masked for
> testing" - what I wanted to clarify is that the mask reason isn't valid
> if you want stuff to get tested, as it prevents exactly that from
> happening.

I would argue that overlays are a bigger barrier to testing than being
"masked for testing"

At least they are exposed to the entire Gentoo population if they are
p.masked in the tree. Additionally, there are use cases for p.masking
for testing in the tree, especially if you have users testing it for
you. There shouldn't be a limit to the amount of self-QA that we
provide to "protect" the users, if so deemed necessary.

Just saying...
-Jeremy

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Alec Warner]

On Thu, Oct 2, 2008 at 1:24 PM, Jeroen Roovers <jer@gentoo.org> wrote:
> # Gen 2 Developer <someone@gentoo.org> (`date`)
> # Masked for testing.
>>=rofl-cat/omgpkg-ver
>
>
> Please people,
>
>
>   if you want to get something tested, then don't mask it.If you find
> that you cannot commit an ebuild because of badly keyworded
> dependencies, then drop the relevant keywords and file a bug report
> with a KEYWORDREQ.

So you have made two comments.

1) 'Package.mask not for testing.'

2) 'If you cannot commit an ebuild due to badly keyworded deps; file
bugs to get them fixed.'

I agree with 2 but not with 1.

If pmask is not for testing...what is it for?

-Alec

>
>
> Kind regards,
>     JeR
>
>

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Josh Saddler]

[-- Attachment #1: Type: text/plain, Size: 113 bytes --]

Alec Warner wrote:
> If pmask is not for testing...what is it for?

UT GOTY and nvidia-drivers, of course!


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Jeroen Roovers]

On Thu, 2 Oct 2008 17:56:39 -0700
"Alec Warner" <antarus@gentoo.org> wrote:

> If pmask is not for testing...what is it for?

The name says it all - to prevent people from automatically emerging
stuff, even when ACCEPT_KEYWORDS=~arch is set. First you try for the new
version:

# emerge -va www-client/opera

which doesn't work (it gives you the current version!). Then you try
with a specific version:

# emerge -va =www-client/opera-9.6*

which gives you a good reason to either unmask or not unmask:

!!! All ebuilds that could satisfy "=www-client/opera-9.6*" have been
masked. !!! One of the following masked packages is required to
complete your request:
- www-client/opera-9.60_pre2440 (masked by: package.mask)
/keeps/gentoo/portage/profiles/package.mask:
# Jeroen Roovers <jer@gentoo.org> (26 Aug 2008)
# www-client/opera snapshots are masked. Please read
# http://my.opera.com/desktopteam/blog/

- www-client/opera-9.60_pre2436 (masked by: package.mask)
- [...]

If it merely says that the masking is for "testing" (and especially if
testing takes many months and apparently takes place in secret) the
whole point is lost on the people who have come so far and still want to
press on - they'll simply ignore your "warning against testing".

There are various valid reasons, but testing means you want to expose
stuff, not hide it. There's simply no way you'd package.mask something,
and at the same time explain you want it tested. Because you're
preventing most ~arch systems from getting automatically widely exposed
to the stuff you're intending to get tested.

Even saying that it would kill puppies would be more valid. Just be
honest and tell people what is going on. Tell them that if they use
Opera snapshots, they shouldn't care about losing mail or experience
frequent crashes while browsing. Anything really, just don't tell them
you're "testing" or you find yourself excluding them from the party
with a really bad excuse.


Kind regards,
     JeR

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Dawid Węgliński]

On Friday 03 of October 2008 04:14:54 Jeroen Roovers wrote:
> On Thu, 2 Oct 2008 17:56:39 -0700
>
> "Alec Warner" <antarus@gentoo.org> wrote:
> > If pmask is not for testing...what is it for?
>
> The name says it all - to prevent people from automatically emerging
> stuff, even when ACCEPT_KEYWORDS=~arch is set. First you try for the new
> version:
>
> # emerge -va www-client/opera
>
> which doesn't work (it gives you the current version!). Then you try
> with a specific version:
>
> # emerge -va =www-client/opera-9.6*
>
> which gives you a good reason to either unmask or not unmask:
>
> !!! All ebuilds that could satisfy "=www-client/opera-9.6*" have been
> masked. !!! One of the following masked packages is required to
> complete your request:
> - www-client/opera-9.60_pre2440 (masked by: package.mask)
> /keeps/gentoo/portage/profiles/package.mask:
> # Jeroen Roovers <jer@gentoo.org> (26 Aug 2008)
> # www-client/opera snapshots are masked. Please read
> # http://my.opera.com/desktopteam/blog/
>
> - www-client/opera-9.60_pre2436 (masked by: package.mask)
> - [...]
>
> If it merely says that the masking is for "testing" (and especially if
> testing takes many months and apparently takes place in secret) the
> whole point is lost on the people who have come so far and still want to
> press on - they'll simply ignore your "warning against testing".

Same way one may see "masked by missing keyword" note and interprete as "not 
for your arch"... So a quick note in p.mask can say it is for testing 
purposes, so user can choose either to install it or not.

>
> There are various valid reasons, but testing means you want to expose
> stuff, not hide it. There's simply no way you'd package.mask something,
> and at the same time explain you want it tested. Because you're
> preventing most ~arch systems from getting automatically widely exposed
> to the stuff you're intending to get tested.

I don't think it's ok. ~arch isn't training ground. It's supposed to work, so 
asking arch teams to keywords packages that are not supposed to work isn't 
good.

>
> Even saying that it would kill puppies would be more valid. Just be
> honest and tell people what is going on. Tell them that if they use
> Opera snapshots, they shouldn't care about losing mail or experience
> frequent crashes while browsing. Anything really, just don't tell them
> you're "testing" or you find yourself excluding them from the party
> with a really bad excuse.

This is the place i agree with you. Anyway i think package still should be 
p.masked with good explanation of why it is masked.

-- 
Cheers,
Dawid Węgliński

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Jeroen Roovers]

On Fri, 3 Oct 2008 04:23:33 +0200
Dawid Węgliński <cla@gentoo.org> wrote:

> I don't think it's ok. ~arch isn't training ground. It's supposed to
> work, so asking arch teams to keywords packages that are not supposed
> to work isn't good.

We have a "testing" branch and a "stable" branch, defined by the
KEYWORDS variable in the ebuilds. Package.masking stuff saying you're
"testing" is at the least uninformative and highly confusing and
unfriendly to would-be testers when in the very same context this
already means something different (namely, it's been too short a
while, wait one or two months for this version to go stable, as the
~arch keywords would suggest).

The same term shouldn't be used to denote two ways of masking ebuilds,
but that's beside the point of providing good reasons to package.mask
ebuilds.

> > Even saying that it would kill puppies would be more valid. Just be
> > honest and tell people what is going on. Tell them that if they use
> > Opera snapshots, they shouldn't care about losing mail or experience
> > frequent crashes while browsing. Anything really, just don't tell
> > them you're "testing" or you find yourself excluding them from the
> > party with a really bad excuse.
> 
> This is the place i agree with you. Anyway i think package still
> should be p.masked with good explanation of why it is masked.

Welcome to the starting point of this thread! ;-)


Kind regards,
     JeR

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Mart Raudsepp]

[-- Attachment #1: Type: text/plain, Size: 867 bytes --]

On Thu, 2008-10-02 at 22:24 +0200, Jeroen Roovers wrote:
> # Gen 2 Developer <someone@gentoo.org> (`date`)
> # Masked for testing.
> >=rofl-cat/omgpkg-ver
> 
> 
> Please people,
> 
> 
>    if you want to get something tested, then don't mask it.

Stuff with high impact better be masked for initial testing, such as new
versions of gcc, glibc and other parts of toolchain and build system,
and other similar things affecting other packages than itself. ~arch is
in my eyes something that updates shouldn't break vastly - a stable
candidate.
Of course when that initial testing is done with helping users, the
reason could be modified to tell things broke and what the tracking bug
is, or unmasked if it works fine with other packages.

-- 
Mart Raudsepp
Gentoo Developer
Mail: leio@gentoo.org
Weblog: http://planet.gentoo.org/developers/leio

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Alec Warner]

On Thu, Oct 2, 2008 at 8:16 PM, Jeroen Roovers <jer@gentoo.org> wrote:
> On Fri, 3 Oct 2008 04:23:33 +0200
> Dawid Węgliński <cla@gentoo.org> wrote:
>
>> I don't think it's ok. ~arch isn't training ground. It's supposed to
>> work, so asking arch teams to keywords packages that are not supposed
>> to work isn't good.
>
> We have a "testing" branch and a "stable" branch, defined by the
> KEYWORDS variable in the ebuilds. Package.masking stuff saying you're
> "testing" is at the least uninformative and highly confusing and
> unfriendly to would-be testers when in the very same context this
> already means something different (namely, it's been too short a
> while, wait one or two months for this version to go stable, as the
> ~arch keywords would suggest).

~arch has always been for testing ebuilds; not packages.  You should
not be using ~arch to test stuff you know doesn't work; that is what
package.mask is for; to prevent users from accidentally installing
broken shit.

>
> The same term shouldn't be used to denote two ways of masking ebuilds,
> but that's beside the point of providing good reasons to package.mask
> ebuilds.
>

I completely agree that useful messages in package.mask are important.

-Alec

>> > Even saying that it would kill puppies would be more valid. Just be
>> > honest and tell people what is going on. Tell them that if they use
>> > Opera snapshots, they shouldn't care about losing mail or experience
>> > frequent crashes while browsing. Anything really, just don't tell
>> > them you're "testing" or you find yourself excluding them from the
>> > party with a really bad excuse.
>>
>> This is the place i agree with you. Anyway i think package still
>> should be p.masked with good explanation of why it is masked.
>
> Welcome to the starting point of this thread! ;-)
>
>
> Kind regards,
>     JeR
>
>

[gentoo-dev] Re: Testing is not a valid reason to package.mask

[Duncan]

Mart Raudsepp <leio@gentoo.org> posted 1223017599.29403.2.camel@localhost,
excerpted below, on  Fri, 03 Oct 2008 10:06:39 +0300:

> Of course when that initial testing is done with helping users, the
> reason could be modified to tell things broke and what the tracking bug
> is, or unmasked if it works fine with other packages.

From previous discussions on this, that's really the point (besides the 
one about not masking it if testing is needed, which toolchain for 
instance pretty much has to do anyway).  If it has a tracking bug, it has 
the necessary info.  If it's just "masked for testing", the necessary 
info isn't there.

This helps me as a user who often does that sort of testing, too.  Masked 
for testing simply isn't that useful.  A tracking bug, where I can see 
how that testing is progressing and what other sorts of stuff I might 
expect to have issues with if I DO test, now THAT's actual practical 
info!  Simply "masked for testing" is little better than no comment at 
all, or than a package revision bump without a changelog entry telling me 
what the big deal was that was worth the revision.  (That's another 
irritating one, but fortunately it doesn't happen so often any more.  
Thanks guys!)

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Thomas Sachau]

[-- Attachment #1: Type: text/plain, Size: 1297 bytes --]

Alec Warner schrieb:
> On Thu, Oct 2, 2008 at 8:16 PM, Jeroen Roovers <jer@gentoo.org> wrote:
>> On Fri, 3 Oct 2008 04:23:33 +0200
>> Dawid Węgliński <cla@gentoo.org> wrote:
>>
>>> I don't think it's ok. ~arch isn't training ground. It's supposed to
>>> work, so asking arch teams to keywords packages that are not supposed
>>> to work isn't good.
>> We have a "testing" branch and a "stable" branch, defined by the
>> KEYWORDS variable in the ebuilds. Package.masking stuff saying you're
>> "testing" is at the least uninformative and highly confusing and
>> unfriendly to would-be testers when in the very same context this
>> already means something different (namely, it's been too short a
>> while, wait one or two months for this version to go stable, as the
>> ~arch keywords would suggest).
> 
> ~arch has always been for testing ebuilds; not packages.  You should
> not be using ~arch to test stuff you know doesn't work; that is what
> package.mask is for; to prevent users from accidentally installing
> broken shit.
> 

Why do you need package.mask here? If you know, it does not work on that arch, dont keyword it. If
you know it does not work anywhere, why would you even think about adding that package?


-- 
Thomas Sachau

Gentoo Linux Developer


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 315 bytes --]

Re: [gentoo-dev] Testing is not a valid reason to package.mask

[Alec Warner]

On Fri, Oct 3, 2008 at 5:06 AM, Thomas Sachau <tommy@gentoo.org> wrote:
> Alec Warner schrieb:
>> On Thu, Oct 2, 2008 at 8:16 PM, Jeroen Roovers <jer@gentoo.org> wrote:
>>> On Fri, 3 Oct 2008 04:23:33 +0200
>>> Dawid Węgliński <cla@gentoo.org> wrote:
>>>
>>>> I don't think it's ok. ~arch isn't training ground. It's supposed to
>>>> work, so asking arch teams to keywords packages that are not supposed
>>>> to work isn't good.
>>> We have a "testing" branch and a "stable" branch, defined by the
>>> KEYWORDS variable in the ebuilds. Package.masking stuff saying you're
>>> "testing" is at the least uninformative and highly confusing and
>>> unfriendly to would-be testers when in the very same context this
>>> already means something different (namely, it's been too short a
>>> while, wait one or two months for this version to go stable, as the
>>> ~arch keywords would suggest).
>>
>> ~arch has always been for testing ebuilds; not packages.  You should
>> not be using ~arch to test stuff you know doesn't work; that is what
>> package.mask is for; to prevent users from accidentally installing
>> broken shit.
>>
>
> Why do you need package.mask here? If you know, it does not work on that arch, dont keyword it. If
> you know it does not work anywhere, why would you even think about adding that package?

Nuances ;)

What does a lack of keyword mean?  It means that no dev has bothered
to test the package on said arch.  It doesn't mean the package does
not work properly on said arch.  Users who run alt arches like sparc
end up ~arch keywording stuff locally all the time; it would be
unfortunate were they to keyword a totally broken package on sparc
just because the dev didn't keyword it.  Users often think this means
'lack of time' not 'does not function'.

What does -arch mean?  It means that the package *will* never work on
said arch (64-bit binaries on x86 for example); it does not mean 'this
package *may* not work'; so keywording broken packages with -arch is
also not quite correct (although arguably you could move from -arch,
to ~arch, to arch and maybe get away with it.)

Package.mask can be used for evaluating packages.  Many developers
would suggest using overlays for these types of packages; however not
everyone has an overlay and not everyone uses overlays so I don't
think there should be a hard and fast rule here.

>
>
> --
> Thomas Sachau
>
> Gentoo Linux Developer
>
>

[gentoo-dev] Re: Testing is not a valid reason to package.mask

[Ryan Hill]

[-- Attachment #1: Type: text/plain, Size: 1397 bytes --]

On Thu, 2 Oct 2008 22:24:35 +0200
Jeroen Roovers <jer@gentoo.org> wrote:

> Please people,
> 
> 
>    if you want to get something tested, then don't mask it.

Um... no?  One thing that package.mask has always been used for is
temporarily masking a package until it can be tested and then unleashed
on the general population.  It's not like we're putting masked stuff in
the tree with the hope that someone will find it and try it out.  You
mask a package, ask the user or whoever to test it, and unmask it when
it's ready.  We don't just throw untested stuff into the tree when we
suspect problems with it. ~arch is not a playground.  Already one of
the major complaints we see against Gentoo time and time again is that
it breaks too often and the maintenance burden is too high.  Why would
we want to exacerbate that?

We don't /want/ ~arch systems to get "automatically widely exposed to
the stuff we're intending to get tested".  That's the whole point of
masking it!  We want it tested by a few people before we expose it to
the unwashed masses.

So, no, I'll continue using package.mask for testing just
as it always has been.  Sorry.


-- 
gcc-porting,                                      by design, by neglect
treecleaner,                              for a fact or just for effect
wxwidgets @ gentoo     EFFD 380E 047A 4B51 D2BD C64F 8AA8 8346 F9A4 0662

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: [gentoo-dev] Re: Testing is not a valid reason to package.mask

[Thomas Anderson]

[-- Attachment #1: Type: text/plain, Size: 394 bytes --]

On Fri, Oct 03, 2008 at 11:44:10PM -0600, Ryan Hill wrote:
> On Thu, 2 Oct 2008 22:24:35 +0200
> Jeroen Roovers <jer@gentoo.org> wrote:
> 
> > Please people,
> > 
> > 
> >    if you want to get something tested, then don't mask it.
> 
> So, no, I'll continue using package.mask for testing just
> as it always has been.  Sorry.

++, especially on unleashing broken stuff to users.

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: [gentoo-dev] Re: Testing is not a valid reason to package.mask

[Michal Kurgan]

[-- Attachment #1: Type: text/plain, Size: 412 bytes --]

On Fri, 3 Oct 2008 23:44:10 -0600
Ryan Hill <dirtyepic@gentoo.org> wrote:

> On Thu, 2 Oct 2008 22:24:35 +0200
>
> So, no, I'll continue using package.mask for testing just
> as it always has been.  Sorry.
> 

As far as i understand, the complaint is not about testing itself, but
about providing more detailed information in the package.mask file.

-- 
Michal Kurgan
http://dev.gentoo.org/~moloh

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: [gentoo-dev] Re: Testing is not a valid reason to package.mask

[Iain Buchanan]

Ryan Hill wrote:
> On Thu, 2 Oct 2008 22:24:35 +0200
> Jeroen Roovers<jer@gentoo.org>  wrote:
>
>> Please people,
>>
>>
>>     if you want to get something tested, then don't mask it.
>
> Um... no?  One thing that package.mask has always been used for is
> temporarily masking a package until it can be tested and then unleashed
> on the general population.

I think there's "testing" and "testing", and we're getting confused 
between the two :)

The testing cycle with packages that you know will badly break 
something, usually involves test, patch, test, patch, etc. During which 
the package is masked for good reason (the reason specified in 
package.mask) and certain users may unmask for whatever reason (helping 
to test, etc).

Then once you're happy to unleash it on ~arch, it still requires some 
amount of testing, but generally isn't "may delete all your data" testing.

>  It's not like we're putting masked stuff in
> the tree with the hope that someone will find it and try it out.  You
> mask a package, ask the user or whoever to test it, and unmask it when
> it's ready.  We don't just throw untested stuff into the tree when we
> suspect problems with it. ~arch is not a playground.  Already one of
> the major complaints we see against Gentoo time and time again is that
> it breaks too often and the maintenance burden is too high.  Why would
> we want to exacerbate that?

But this isn't a complaint against ~arch surely?  The general feeling I 
get from gentoo-user when someone complains about an ~arch "production 
box" or "remote system" that broke, is "well, what did you expect from 
~arch?"

> We don't /want/ ~arch systems to get "automatically widely exposed to
> the stuff we're intending to get tested".

No, not "delete all your data" testing, but yes you do want it exposed 
to "may still be slightly quirky" testing.

>  That's the whole point of
> masking it!  We want it tested by a few people before we expose it to
> the unwashed masses.

I would assume the unwashed masses are arch, not ~arch.  If you're 
installing ~arch:

"~arch keyword means that the application is not tested sufficiently to 
be put in the stable branch" [1]

"We recommend that you only use the stable branch. However, if you don't 
care about stability this much..." [1]

"The testing branch is exactly what it says - Testing. If a package is 
in testing, it means that the developers feel that it is functional but 
has not been thoroughly tested. You could very well be the first to 
discover a bug in the package in which case you could file a bugreport 
to let the developers know about it.
Beware though, you might notice stability issues, imperfect package 
handling (for instance wrong/missing dependencies), too frequent updates 
(resulting in lots of building) or broken packages. If you do not know 
how Gentoo works and how to solve problems, we recommend that you stick 
with the stable and tested branch." [1]

> So, no, I'll continue using package.mask for testing just
> as it always has been.  Sorry.

All IMHO from a user point of view, of course.

[1] Gentoo Linux x86 Handbook http://www.gentoo.org/doc/en/handbook/

cya,
-- 
Iain Buchanan <iaindb at netspace dot net dot au>

fenderberg, n.:
	The large glacial deposits that form on the insides
	of car fenders during snowstorms.
		-- "Sniglets", Rich Hall & Friends