[gentoo-dev] emerge test...

[gentoo-dev] emerge test...

[Riyad Kalla]

Does Gentoo have a script that can test all available emerges for
errors? Say have a machine setup, that has fast compiler settings (no
optimizations) that just emerges the entire portage tree all day long...
or do you just wait for people to report problems when emerging
packages?

is there anywya to automate this task? Seems not very likely, but I'm
not sure of the intracacies of portage...

-Riyad


--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] emerge test...

[Brian Magnuson]

This is a nice idea, but I think that the problem becomes intractable when you
take USE flags into account.  To properly test an ebuild you would have to build
it with every permutation of it USE flags.  As an extreme example, mplayer takes
something like 20 possible USE flags.  That's about 1000000 builds of mplayer
that you just signed up for.  Or 32 seperate builds of xfree86, and that's for
every version that current exists in portage, not just the latest one.  You can
see how this gets out of control rather quickly...


* Riyad Kalla <rsk@u.arizona.edu> [2002-12-11 09:45]:
> Does Gentoo have a script that can test all available emerges for
> errors? Say have a machine setup, that has fast compiler settings (no
> optimizations) that just emerges the entire portage tree all day long...
> or do you just wait for people to report problems when emerging
> packages?
> 
> is there anywya to automate this task? Seems not very likely, but I'm
> not sure of the intracacies of portage...
> 
> -Riyad
> 
> 
> --
> gentoo-dev@gentoo.org mailing list
> 

--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] emerge test...

[Thomas T. Veldhouse]

It is relatively safe to make sure that you have tested every use flag, not
every combination of use flags.

Tom Veldhouse

----- Original Message -----
From: "Brian Magnuson" <magnuson@rcn.com>
To: <gentoo-dev@gentoo.org>
Cc: <rsk@u.arizona.edu>
Sent: Wednesday, December 11, 2002 11:28 AM
Subject: Re: [gentoo-dev] emerge test...


> This is a nice idea, but I think that the problem becomes intractable when
you
> take USE flags into account.  To properly test an ebuild you would have to
build
> it with every permutation of it USE flags.  As an extreme example, mplayer
takes
> something like 20 possible USE flags.  That's about 1000000 builds of
mplayer
> that you just signed up for.  Or 32 seperate builds of xfree86, and that's
for
> every version that current exists in portage, not just the latest one.
You can
> see how this gets out of control rather quickly...
>
>
> * Riyad Kalla <rsk@u.arizona.edu> [2002-12-11 09:45]:
> > Does Gentoo have a script that can test all available emerges for
> > errors? Say have a machine setup, that has fast compiler settings (no
> > optimizations) that just emerges the entire portage tree all day long...
> > or do you just wait for people to report problems when emerging
> > packages?
> >
> > is there anywya to automate this task? Seems not very likely, but I'm
> > not sure of the intracacies of portage...
> >
> > -Riyad
> >
> >
> > --
> > gentoo-dev@gentoo.org mailing list
> >
>
> --
> gentoo-dev@gentoo.org mailing list
>
>


--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] emerge test...

[Jon Nall]

[-- Attachment #1: Type: text/plain, Size: 1723 bytes --]


good point, but you can be smart about it and try and test where you get
the most bang for your buck. for example, set all possible USE flags for
a build and test the latest stable release. sure, you're missing some
cases, but you're (hopefully) hitting the common ones. 

in truth, my autobuilder is going to be used to automate building of
packages that aren't already keyworded for a give ARCH. it just happens
that it can be used for what riyad suggests too.

nall.

On Wed, 2002-12-11 at 11:28, Brian Magnuson wrote:
> This is a nice idea, but I think that the problem becomes intractable when you
> take USE flags into account.  To properly test an ebuild you would have to build
> it with every permutation of it USE flags.  As an extreme example, mplayer takes
> something like 20 possible USE flags.  That's about 1000000 builds of mplayer
> that you just signed up for.  Or 32 seperate builds of xfree86, and that's for
> every version that current exists in portage, not just the latest one.  You can
> see how this gets out of control rather quickly...
> 
> 
> * Riyad Kalla <rsk@u.arizona.edu> [2002-12-11 09:45]:
> > Does Gentoo have a script that can test all available emerges for
> > errors? Say have a machine setup, that has fast compiler settings (no
> > optimizations) that just emerges the entire portage tree all day long...
> > or do you just wait for people to report problems when emerging
> > packages?
> > 
> > is there anywya to automate this task? Seems not very likely, but I'm
> > not sure of the intracacies of portage...
> > 
> > -Riyad
> > 
> > 
> > --
> > gentoo-dev@gentoo.org mailing list
> > 
> 
> --
> gentoo-dev@gentoo.org mailing list
> 

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 232 bytes --]

Re: [gentoo-dev] emerge test...

[Brian Magnuson]

* Thomas T. Veldhouse <veldy@veldy.net> [2002-12-11 12:35]:
> It is relatively safe to make sure that you have tested every use flag, not
> every combination of use flags.

Yeah, I suppose you're right about that.  What I was getting at is that a system
like this is inevitably going to leave holes in the testing and that these will
very likely be the very same holes left by the developer of the ebuild if they
did a minimum of testing on their own ebuild.  e.g.  Trying a build with each
one of the use flags.  To do more than this, such as trying permutations of USE
flags or building against multiple versions of dependencies runs into the
exploding problem space issue I mentioned before.

I just think that users will find bugs much faster than a build bot since I
think that a large portion of the Gentoo user population tend to be bleeding
edgers and put that ~arch in their make.conf.  Not that I would discourage
someone from trying to set up such a system.  I just question its usefulness.

-Brian

--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] emerge test...

[Brian Magnuson]

* Jon Nall <nall@gentoo.org> [2002-12-11 12:45]:

> in truth, my autobuilder is going to be used to automate building of
> packages that aren't already keyworded for a give ARCH. it just happens
> that it can be used for what riyad suggests too.
> 

So you intend this script to be used for builds that have been commited but
are still under active development and havn't been declared safe for anyone?
That sounds like a useful tool, but something that would be run individually
by the developer of the ebuild instead of continually looping through all such
builds.  Since the ebuilds themselves are expected to be unstable they will
likely generate lots of build failures and unless your reporting mechanism is
totally passive I think people might get kinda tired of hearing from your
script. :)

--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] emerge test...

[Jon Nall]

[-- Attachment #1: Type: text/plain, Size: 1464 bytes --]


perhaps i wasn't clear enough. say ebuild foo has KEYWORDS="x86", and
all it's dependencies have KEYWORDS="x86 ppc". barring any x86-centric
stuff in the build, there's a fine chance it's going to build and run on
ppc. my script finds these types of cases and tries to build them,
logging whether each was built or not. this is useful for a ppc dev for
keywording purposes. (feel free to substitute sparc or alpha (or the
rare x86) for ppc above).

note: no one gets unwanted mails with this.

nall.


On Wed, 2002-12-11 at 12:11, Brian Magnuson wrote:
> * Jon Nall <nall@gentoo.org> [2002-12-11 12:45]:
> 
> > in truth, my autobuilder is going to be used to automate building of
> > packages that aren't already keyworded for a give ARCH. it just happens
> > that it can be used for what riyad suggests too.
> > 
> 
> So you intend this script to be used for builds that have been commited but
> are still under active development and havn't been declared safe for anyone?
> That sounds like a useful tool, but something that would be run individually
> by the developer of the ebuild instead of continually looping through all such
> builds.  Since the ebuilds themselves are expected to be unstable they will
> likely generate lots of build failures and unless your reporting mechanism is
> totally passive I think people might get kinda tired of hearing from your
> script. :)
> 
> --
> gentoo-dev@gentoo.org mailing list
> 

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 232 bytes --]