[gentoo-dev] Policy for bugs because of use flags

[gentoo-dev] Policy for bugs because of use flags

[Jörg Schaible]

Hi folks,

please let me know, since when do we have a bug policy for "standard
accepted safe use flags"? I don't expect any developer to test a variety of
different use flags, but what if a user reports, that a package is
seriously broken because of a use flag? In former times a critical flag was
just filtered out in the ebuild, but why is such a thing fix now rejected ?

Note, using Gentoo with an own set of flags is one of the benefits for
Gentoo. Otherwise you can stick with prebuild distros and avoid hours of
compiling.

- Jörg


--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Policy for bugs because of use flags

[Ciaran McCreesh]

[-- Attachment #1: Type: text/plain, Size: 1269 bytes --]

On Thu, 11 Nov 2004 23:49:42 +0100 Jörg Schaible <joerg.schaible@gmx.de>
wrote:
| please let me know, since when do we have a bug policy for "standard
| accepted safe use flags"? I don't expect any developer to test a
| variety of different use flags, but what if a user reports, that a
| package is seriously broken because of a use flag? In former times a
| critical flag was just filtered out in the ebuild, but why is such a
| thing fix now rejected ?

Are you talking about USE flags or CFLAGS? If you mean USE flags, it
used to be the case that no combination should cause a failure, but
nowadays certain developers are making things barf in pkg_setup instead
(which IMO sucks, since any failure of this kind should happen *before*
all the dependencies start to be merged).

If you mean CFLAGS, you won't have any problems if you stick with a nice
sane set (on the rare occasions when sensible CFLAGS do cause problems,
they're generally filtered). If you use stupid CFLAGS, expect things to
break. Anyone using -ffast-math globally, for example, deserves
everything they get...

-- 
Ciaran McCreesh : Gentoo Developer (Vim, Fluxbox, Sparc, Mips)
Mail            : ciaranm at gentoo.org
Web             : http://dev.gentoo.org/~ciaranm


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

Re: [gentoo-dev] Policy for bugs because of use flags

[Mike Frysinger]

On Thursday 11 November 2004 05:59 pm, Ciaran McCreesh wrote:
> nowadays certain developers are making things barf in pkg_setup instead
> (which IMO sucks, since any failure of this kind should happen *before*
> all the dependencies start to be merged).

considering Bug 2272 is still not fixed, there isnt much choice

does it suck ? yes
is there another usuable solution atm ? not really
-mike

--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Policy for bugs because of use flags

[Ciaran McCreesh]

[-- Attachment #1: Type: text/plain, Size: 1980 bytes --]

On Thu, 11 Nov 2004 18:07:53 -0500 Mike Frysinger <vapier@gentoo.org>
wrote:
| On Thursday 11 November 2004 05:59 pm, Ciaran McCreesh wrote:
| > nowadays certain developers are making things barf in pkg_setup
| > instead(which IMO sucks, since any failure of this kind should
| > happen *before* all the dependencies start to be merged).
| 
| considering Bug 2272 is still not fixed, there isnt much choice
| 
| does it suck ? yes
| is there another usuable solution atm ? not really

2272 isn't the only kind of problem there... That one will be fixed by
Jason's new dep code if it's ever integrated... What's more annoying is
packages which do something like:

pkg_setup() {
	if use foo && ! use bar ; then
		eerror "You must USE bar if you USE foo"
		die "meh"
	fi
}

The way earlier packages handled it was to either treat bar as being
enabled no matter what if foo was set or to ignore the bar flag entirely
unless foo was set. Similarly, some packages will do things like:

pkg_setup() {
	if use gtk && use qt ; then
		eerror "You cannot USE both qt and gtk"
		die "meh"
	fi
}

The traditional approach was to select whatever worked best for the
package in question when this occured (for example, gvim will select gtk
over motif if both are enabled).

Does the old method have drawbacks? Sure. Are they less of a nuisance
than coming back after a day to a failure message half way through a
build of a hundred packages (many of which later turn out to be
unnecessary thanks to USE conflicts) to find a big fat error that could
have been determined before anything was installed at all? *shrug*
maybe.

I seem to recall sort of discussing this one with Stuart (?) and one of
the portage guys a while back... Don't think anyone ever agreed upon
whether a prebuild_check() {} or some scary variable would be better...

-- 
Ciaran McCreesh : Gentoo Developer (Vim, Fluxbox, Sparc, Mips)
Mail            : ciaranm at gentoo.org
Web             : http://dev.gentoo.org/~ciaranm


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

Re: [gentoo-dev] Policy for bugs because of use flags

[Mike Frysinger]

On Thursday 11 November 2004 06:19 pm, Ciaran McCreesh wrote:
> <snip>
> I seem to recall sort of discussing this one with Stuart (?) and one of
> the portage guys a while back... Don't think anyone ever agreed upon
> whether a prebuild_check() {} or some scary variable would be better...

true, this is a different kind of annoyance and one that should be fixed 
properly imho ... have we ever bothered filing a bug about it ? :)
-mike

--
gentoo-dev@gentoo.org mailing list

[gentoo-dev] Re: Policy for bugs because of CFLAGS (was: use flags)

[Jörg Schaible]

Hi Ciaran,

Ciaran McCreesh wrote:

> On Thu, 11 Nov 2004 23:49:42 +0100 Jörg Schaible <joerg.schaible@gmx.de>
> wrote:
> | please let me know, since when do we have a bug policy for "standard
> | accepted safe use flags"? I don't expect any developer to test a
> | variety of different use flags, but what if a user reports, that a
> | package is seriously broken because of a use flag? In former times a
> | critical flag was just filtered out in the ebuild, but why is such a
> | thing fix now rejected ?
> 
[snip]
> 
> If you mean CFLAGS, you won't have any problems if you stick with a nice
> sane set (on the rare occasions when sensible CFLAGS do cause problems,
> they're generally filtered). If you use stupid CFLAGS, expect things to
> break. Anyone using -ffast-math globally, for example, deserves
> everything they get...


Seesh. You're right. I was talking about CFLAGS. Now, is there a policy for
"standard accepted safe CFLAGS" ? I know, it is my problem, if something
really breaks. But I have really a lot of apps running with my (not too
esoteric) settings, and if one single app fails badly (just core dumps) and
I can track it down to a single CFLAG flag, what is the actual policy (if
there is one)? Why is a filter for the flag just recected?

- Jörg



--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Re: Policy for bugs because of CFLAGS (was: use flags)

[Mike Frysinger]

On Thursday 11 November 2004 06:58 pm, Jörg Schaible wrote:
> Seesh. You're right. I was talking about CFLAGS. Now, is there a policy for
> "standard accepted safe CFLAGS" ? I know, it is my problem, if something
> really breaks. But I have really a lot of apps running with my (not too
> esoteric) settings, and if one single app fails badly (just core dumps) and
> I can track it down to a single CFLAG flag, what is the actual policy (if
> there is one)? Why is a filter for the flag just recected?

no policy really

generally if the user is using extended (read: stupid) CFLAGS, developers may 
tell them to fix their CFLAGS

if a package can be found to blow up while using one flag regardless of the 
rest of the CFLAG settings (except for perhaps -O#), then generally we 
flag-filter it

policy ? none
too much policy makes jack a dull boy; dull boys dont get girls; no girls 
means no kids; poor jack dies alone and his seed does not live on
-mike

--
gentoo-dev@gentoo.org mailing list

[gentoo-dev] Re: Policy for bugs because of CFLAGS (was: use flags)

[Jörg Schaible]

Hi Mike,

Mike Frysinger wrote:

[snip]

> no policy really
> 
> generally if the user is using extended (read: stupid) CFLAGS, developers
> may tell them to fix their CFLAGS
> 
> if a package can be found to blow up while using one flag regardless of
> the rest of the CFLAG settings (except for perhaps -O#), then generally we
> flag-filter it
> 
> policy ? none
> too much policy makes jack a dull boy; dull boys dont get girls; no girls
> means no kids; poor jack dies alone and his seed does not live on
> -mike

Glad to here, but this answer indicates something else:
http://bugs.gentoo.org/show_bug.cgi?id=70627

- Jörg





--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Re: Policy for bugs because of CFLAGS (was: use flags)

[Mike Frysinger]

On Thursday 11 November 2004 07:19 pm, Jörg Schaible wrote:
> http://bugs.gentoo.org/show_bug.cgi?id=70627

fixed in cvs
-mike

--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Re: Policy for bugs because of CFLAGS (was: use flags)

[Paul de Vrieze]

[-- Attachment #1: Type: text/plain, Size: 1012 bytes --]

On Friday 12 November 2004 00:58, Jörg Schaible wrote:
> Seesh. You're right. I was talking about CFLAGS. Now, is there a policy
> for "standard accepted safe CFLAGS" ? I know, it is my problem, if
> something really breaks. But I have really a lot of apps running with
> my (not too esoteric) settings, and if one single app fails badly (just
> core dumps) and I can track it down to a single CFLAG flag, what is the
> actual policy (if there is one)? Why is a filter for the flag just
> recected?

One thing that I use as baseline is that if "info gcc" says that it is not 
safe (such as -ffast-math, -fno-rtti, -fnoexceptions and various others). 
Many of these optimization options are application specific and should be 
set by the upstream build scripts. In other words, filtering away to 
avoid compiler bugs is something I'm willing to do, filtering away stupid 
defaults is not.

Paul

-- 
Paul de Vrieze
Gentoo Developer
Mail: pauldv@gentoo.org
Homepage: http://www.devrieze.net

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

[gentoo-dev] Re: Re: Policy for bugs because of CFLAGS (was: use flags)

[Jörg Schaible]

Paul de Vrieze wrote:

> On Friday 12 November 2004 00:58, Jörg Schaible wrote:
>> Seesh. You're right. I was talking about CFLAGS. Now, is there a policy
>> for "standard accepted safe CFLAGS" ? I know, it is my problem, if
>> something really breaks. But I have really a lot of apps running with
>> my (not too esoteric) settings, and if one single app fails badly (just
>> core dumps) and I can track it down to a single CFLAG flag, what is the
>> actual policy (if there is one)? Why is a filter for the flag just
>> recected?
> 
> One thing that I use as baseline is that if "info gcc" says that it is not
> safe (such as -ffast-math, -fno-rtti, -fnoexceptions and various others).
> Many of these optimization options are application specific and should be
> set by the upstream build scripts. In other words, filtering away to
> avoid compiler bugs is something I'm willing to do, filtering away stupid
> defaults is not.

OK, fine with me. But for the standard ones (turned on by -Os, -O1, -O2)
this should be not the problem.

- Jörg


--
gentoo-dev@gentoo.org mailing list

Re: [gentoo-dev] Policy for bugs because of use flags

[Mike Frysinger]

On Thursday 11 November 2004 05:49 pm, Jörg Schaible wrote:
> please let me know, since when do we have a bug policy for "standard
> accepted safe use flags"? I don't expect any developer to test a variety of
> different use flags, but what if a user reports, that a package is
> seriously broken because of a use flag? In former times a critical flag was
> just filtered out in the ebuild, but why is such a thing fix now rejected ?

why not be a little more specific with a Bug #

but yes, if a package fails to build due to some combination of USE flags, 
generally it's a bug

but not always :P
for example, the games team provides 'noaudio' 'nojoystick' and 'novideo' USE 
flags ... if you enable them, build libsdl, and then some package fails to 
build, we close it as WONTFIX:YOUR_PROBLEM
-mike

--
gentoo-dev@gentoo.org mailing list