[gentoo-dev] Amount of useflags enabled by default

[gentoo-dev] Amount of useflags enabled by default

[Thomas Sachau]

[-- Attachment #1: Type: text/plain, Size: 755 bytes --]

Hi,

i would like to start a discussion about reducing the amount of default-enabled USE flags in
profiles, especially in inherited basic profiles.

Is there any policy, when they are added, how the reason for addition is documented and when they
are removed from profiles?

In addition, i see a trend to enabled more more more USE flags (either over profiles or via IUSE
+flag). Whats the reason for forcing a big load of default enabled USE flags on every user including
more dependencies, more compile time, more wasted disk space and more possible vulnerabilities
except some users, who complain about a missing feature and are not able to think and enable a USE
flag for that feature?


-- 
Thomas Sachau

Gentoo Linux Developer


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 316 bytes --]

Re: [gentoo-dev] Amount of useflags enabled by default

[Sebastian Pipping]

Thomas Sachau wrote:
> In addition, i see a trend to enabled more more more USE flags (either over profiles or via IUSE
> +flag).

I'm not sure for how much of the IUSE="+foo" cases this applies but I
can explain one of them:

In  xfce-base/xfce4-session-4.6.1-r1  there is  +fortune  in IUSE
because otherwise previous users of 4.6.1 would be missing the feature
enabled by it.

If you worry that +foo is becoming a trend _without_ strong reason you
could try enforcing documentation for in metadata.xml (through new tags)
maybe.



Sebastian

Re: [gentoo-dev] Amount of useflags enabled by default

[Petteri Räty]

[-- Attachment #1: Type: text/plain, Size: 639 bytes --]

Thomas Sachau wrote:
> 
> In addition, i see a trend to enabled more more more USE flags (either over profiles or via IUSE
> +flag). Whats the reason for forcing a big load of default enabled USE flags on every user including
> more dependencies, more compile time, more wasted disk space and more possible vulnerabilities
> except some users, who complain about a missing feature and are not able to think and enable a USE
> flag for that feature?
> 

One possible reason is that our packages should follow upstream policy
and maybe upstreams usually like to keep things enabled rather than
disabled.

Regards,
Petteri



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]

Re: [gentoo-dev] Amount of useflags enabled by default

[Alistair Bush]

> Hi,
> 
> i would like to start a discussion about reducing the amount of
>  default-enabled USE flags in profiles, especially in inherited basic
>  profiles.

Sounds like a reasonable idea to me, for the base profiles at least.

> In addition, i see a trend to enabled more more more USE flags (either over
>  profiles or via IUSE +flag). Whats the reason for forcing a big load of
>  default enabled USE flags on every user including more dependencies, more
>  compile time, more wasted disk space and more possible vulnerabilities
>  except some users, who complain about a missing feature and are not able
>  to think and enable a USE flag for that feature?
> 

".... who complain about a enabled feature and are not able to think and 
disable a USE flag for that feature?"

What a couple of changes make....

It would be nice if we actually documented why they were enabled.  Does the 
use flag enable significant functionality that would otherwise make the software 
less useful.

I believe we should be trying to find a nice 'middle of the road' balance.   DE 
"related" use flags should be enabled in profiles ( unless of coarse they 
package is already DE related e.g if a kde package has a use flag for kde's 
sound system, this could be enabled at a package level while a package with a 
kde use flag should not default enable it.).

So,  if we were looking at what use flags ppl are enabling/disabling we should 
be seeing similar numbers for the "+'s" and the "-'s".  Not an easy thing to 
do, I admit.   Would be interesting if something like Color Graphing [1] or 
some algorithm could be used to determine whether a use flag should be enabled 
in a profile (including which profile)/package.  Maybe we should have some 
addition metadata for use flags.  Like a category/type/blah/blee. As an example 
we could have a category of DE containing kde/gnome/etc.  How do we know that 
the dirac use flag is codec related without knowing what dirac is?


Alistair


[1] http://en.wikipedia.org/wiki/Graph_coloring

Re: [gentoo-dev] Amount of useflags enabled by default

[Pacho Ramos]

El vie, 23-10-2009 a las 22:54 +0300, Petteri Räty escribió:
> Thomas Sachau wrote:
> > 
> > In addition, i see a trend to enabled more more more USE flags (either over profiles or via IUSE
> > +flag). Whats the reason for forcing a big load of default enabled USE flags on every user including
> > more dependencies, more compile time, more wasted disk space and more possible vulnerabilities
> > except some users, who complain about a missing feature and are not able to think and enable a USE
> > flag for that feature?
> > 
> 
> One possible reason is that our packages should follow upstream policy
> and maybe upstreams usually like to keep things enabled rather than
> disabled.
> 
> Regards,
> Petteri
> 
> 

I don't see any problem in enabling some USE flags by default but,
maybe, would be interesting to have a place where people could consult
why some USEs are enabled and, specially, disabled by default. 

The problem is where to write that information :-/ (into the ebuild,
into metadata file...)

Best regards

Re: [gentoo-dev] Amount of useflags enabled by default

[Thomas Sachau]

[-- Attachment #1: Type: text/plain, Size: 1542 bytes --]

Sebastian Pipping schrieb:
> Thomas Sachau wrote:
>> In addition, i see a trend to enabled more more more USE flags (either over profiles or via IUSE
>> +flag).
> 
> I'm not sure for how much of the IUSE="+foo" cases this applies but I
> can explain one of them:
> 
> In  xfce-base/xfce4-session-4.6.1-r1  there is  +fortune  in IUSE
> because otherwise previous users of 4.6.1 would be missing the feature
> enabled by it.
> 
> If you worry that +foo is becoming a trend _without_ strong reason you
> could try enforcing documentation for in metadata.xml (through new tags)
> maybe.
> 
> 
> 
> Sebastian
> 
> 

This is a nice example for a default enabled USE flag, where i see no strong reason to enable it.

Now i dont know that package nor do i know how many users actually use that feature. If the majority
uses that feature, it seems ok to me to enable it by default (either by a specific xfce profile or
default enabled USE flag). If its on the other hand a minority, which just complains louder, i would
still request the useflag to stay offline by default.

Together with this, i am glad to see the move of KDE/GNOME specific USE flags from normal desktop
profile to KDE/GNOME subprofiles. In most cases, users will either use one of those or none of them.
With this change, it will give a better situation for the majority of our users. And for those, who
want to use both profiles, they can have a look at portage manpage, as betelgeuse suggested.

-- 
Thomas Sachau

Gentoo Linux Developer


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 316 bytes --]

Re: [gentoo-dev] Amount of useflags enabled by default

[Thomas Sachau]

[-- Attachment #1: Type: text/plain, Size: 1554 bytes --]

Petteri Räty schrieb:
> Thomas Sachau wrote:
>> In addition, i see a trend to enabled more more more USE flags (either over profiles or via IUSE
>> +flag). Whats the reason for forcing a big load of default enabled USE flags on every user including
>> more dependencies, more compile time, more wasted disk space and more possible vulnerabilities
>> except some users, who complain about a missing feature and are not able to think and enable a USE
>> flag for that feature?
>>
> 
> One possible reason is that our packages should follow upstream policy
> and maybe upstreams usually like to keep things enabled rather than
> disabled.
> 
> Regards,
> Petteri
> 
> 

With that argument you could request to enable all useflags by default. Its ok in my eyes, if you
follow upstream the way tarballs are created (e.g. qt move to splitted qt packages or the other way
round). Something else would make maintainence part much harder. But i disagree on the part for
"follow upstream policy for default enabled USE flags".
Gentoo is about choice and i would like to have the choice to disable most USE flags by default and
with an easy way, e.g. by choising a profile with less default enabled USE flags. Forcing every user
to disable many or almost all flags independent of his profile would make Gentoo less userfriendly
in general without a good reason. If upstream does not want to support a disabled USE flag, they
should not offer the choice to disable it in the first place.

-- 
Thomas Sachau

Gentoo Linux Developer


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 316 bytes --]

Re: [gentoo-dev] Amount of useflags enabled by default

[Thomas Sachau]

[-- Attachment #1: Type: text/plain, Size: 2472 bytes --]

Alistair Bush schrieb:
>> Hi,
>>
>> i would like to start a discussion about reducing the amount of
>>  default-enabled USE flags in profiles, especially in inherited basic
>>  profiles.
> 
> Sounds like a reasonable idea to me, for the base profiles at least.

Sure. If we have a desktop/kde subprofile, noone should complain, if that profile enables kde
specific USE flags. On the other hand, if i just want a desktop without kde, i dont want to be
forced to either choose a none-desktop profile and enable many USE flags by hand or using the
desktop profile and disabling many USE flags by hand.

> 
>> In addition, i see a trend to enabled more more more USE flags (either over
>>  profiles or via IUSE +flag). Whats the reason for forcing a big load of
>>  default enabled USE flags on every user including more dependencies, more
>>  compile time, more wasted disk space and more possible vulnerabilities
>>  except some users, who complain about a missing feature and are not able
>>  to think and enable a USE flag for that feature?
>>
> 
> ".... who complain about a enabled feature and are not able to think and 
> disable a USE flag for that feature?"
> 
> What a couple of changes make....

I dont mind, if a flag is really usefull and requested by a big majority of the users. But as Gentoo
is about choice, the minority should be able to easily choose something else, e.g. by a less
heavyweight profile. If a majority of mplayer users want to be able to play audio files, i dont mind
to disable it for myself, if i dont want it. But on the other hand shouldnt a handfull of users be
able to dictate the enabled and disabled USE flags for many other users, which might have a
different interest.

> 
> It would be nice if we actually documented why they were enabled.  Does the 
> use flag enable significant functionality that would otherwise make the software 
> less useful.

Documentation is always usefull. One should also check the additional overhead of the USE flag.

> 
> I believe we should be trying to find a nice 'middle of the road' balance.   DE 
> "related" use flags should be enabled in profiles ( unless of coarse they 
> package is already DE related e.g if a kde package has a use flag for kde's 
> sound system, this could be enabled at a package level while a package with a 
> kde use flag should not default enable it.).

I aggree.


-- 
Thomas Sachau

Gentoo Linux Developer


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 316 bytes --]

Re: [gentoo-dev] Amount of useflags enabled by default

[William Hubbs]

[-- Attachment #1: Type: text/plain, Size: 2558 bytes --]

On Sat, Oct 24, 2009 at 06:24:51PM +0200, Thomas Sachau wrote:
> Petteri R??ty schrieb:
> > Thomas Sachau wrote:
> >> In addition, i see a trend to enabled more more more USE flags (either over profiles or via IUSE
> >> +flag). Whats the reason for forcing a big load of default enabled USE flags on every user including
> >> more dependencies, more compile time, more wasted disk space and more possible vulnerabilities
> >> except some users, who complain about a missing feature and are not able to think and enable a USE
> >> flag for that feature?
> >>
> > 
> > One possible reason is that our packages should follow upstream policy
> > and maybe upstreams usually like to keep things enabled rather than
> > disabled.
> > 
> > Regards,
> > Petteri
> > 
> > 
> 
> With that argument you could request to enable all useflags by default. Its ok in my eyes, if you
> follow upstream the way tarballs are created (e.g. qt move to splitted qt packages or the other way
> round). Something else would make maintainence part much harder. But i disagree on the part for
> "follow upstream policy for default enabled USE flags".
> Gentoo is about choice and i would like to have the choice to disable most USE flags by default and
> with an easy way, e.g. by choising a profile with less default enabled USE flags. Forcing every user
> to disable many or almost all flags independent of his profile would make Gentoo less userfriendly
> in general without a good reason. If upstream does not want to support a disabled USE flag, they
> should not offer the choice to disable it in the first place.
 
 I think there are two issues being put together here.  One is the issue
 of profiles enabling use flags by default, and the other is packages
 enabling use flags by default in IUSE.

At the package level, I do think that we should follow the upstream
policy.  Upstream giving you the option to disable something doesn't
mean that they don't support disabling it, it just means that they are
giving you the choice to disable it.  If it is enabled by default, it
could mean that upstream has found that most of their users prefer to
enable it, so they set it up that way.

To me, the question really is at the profile level since enabling use
flags there has the potential to affect entire systems.  I don't think
flags should be enabled at the profile level unless we are sure that
most users who use that profile will want the flags enabled.

-- 
William Hubbs
gentoo accessibility team lead
williamh@gentoo.org

[-- Attachment #2: Type: application/pgp-signature, Size: 198 bytes --]

Re: [gentoo-dev] Amount of useflags enabled by default

[Ed W]

[-- Attachment #1: Type: text/plain, Size: 2259 bytes --]

Thomas Sachau wrote:
> I dont mind, if a flag is really usefull and requested by a big majority of the users. But as Gentoo
> is about choice, the minority should be able to easily choose something else, e.g. by a less
> heavyweight profile. If a majority of mplayer users want to be able to play audio files, i dont mind
> to disable it for myself, if i dont want it. But on the other hand shouldnt a handfull of users be
> able to dictate the enabled and disabled USE flags for many other users, which might have a
> different interest.
>   

Just as a stake in the ground, but personally I have two modes of 
interest: a) for desktop use I just want a middle of the road profile 
which enables "useful stuff", and b) I have some embedded projects where 
every byte is precious


>> It would be nice if we actually documented why they were enabled.  Does the 
>> use flag enable significant functionality that would otherwise make the software 
>> less useful.
>>     
>
> Documentation is always usefull. One should also check the additional overhead of the USE flag.
>   

I often hear this general kind of commentary.  Just out of interest, 
how/why do you care about the byte count that much? Apart from embedded 
work, or perhaps virtualised servers, I find it surprising to imagine 
that "most people" find the "cost" of minimising installed size (well 
more than the obvious stuff)  to be worth the effort (in general)?

What kind of size of install do you run?  Sub 200MB? Sub 50MB?  How much 
"bloat" are you seeing by fiddling with changing your profile from defaults?

Personally I recently figured out how to create my own local profiles, 
and this allows me to control the main USE flags to my liking.  
Personally I find that minimising the number of interpreted languages 
installed (perl etc) and optimising locale size is by far the dominant 
factor in the size of the install.  Thereafter controlling whether 
mplayer also plays x264 files seems largely second order (at least on my 
install)?

If you are worried about security issues in dependencies then do also 
look at hardened (esp. with the gcc-4.4 hardened overlay) and perhaps 
grsecurity - this can very effectively mitigate the effects of many 
security holes.

Good luck

Ed W

[-- Attachment #2: Type: text/html, Size: 2826 bytes --]

[gentoo-dev] Re: Amount of useflags enabled by default

[Duncan]

Ed W posted on Fri, 06 Nov 2009 23:40:30 +0000 as excerpted:

> I often hear this general kind of commentary.  Just out of interest,
> how/why do you care about the byte count that much? Apart from embedded
> work, or perhaps virtualised servers, I find it surprising to imagine
> that "most people" find the "cost" of minimising installed size (well
> more than the obvious stuff)  to be worth the effort (in general)?

> If you are worried about security issues in dependencies then do also
> look at hardened (esp. with the gcc-4.4 hardened overlay) and perhaps
> grsecurity - this can very effectively mitigate the effects of many
> security holes.

What a lot of (at least non-dev) folks don't realize is that particularly 
on a distribution such as Gentoo, in addition to the size bloat, and 
security considerations, both of which you brought up, there's the simple 
or general ongoing maintenance consideration.  Of course that's not quite 
so big of an issue on embedded, where presumably you install it and let 
it be for long periods of time (perhaps for the life of the unit), but 
for general "computer" use, at /least/ desktop, the ongoing updates and 
maintenance costs **FAR** outweigh any size consideration in the usual 
case, and really, except to the extent that updates and security are tied 
together, they outweigh the security aspect of the additional features as 
well.

It was actually a couple years into my Gentoo experience that the effect 
of "bloat" in the form of optional dependencies (USE flags on Gentoo) 
began to dawn on me, and I've only appreciated it more, since, 
with the effect /particularly/ emphasized to me while I had both kde3 and 
kde4 installed (luckily without Gnome to worry about as well).  That's a 
/huge/ number of additional packages to worry about keeping updated, for 
the revdep-rebuild I run after every update to check and maybe flag for 
rebuild, etc.

To a rather lessor but more frequent extent, updating codecs and/or 
imagemagick invariably triggers a revdep-rebuild on transcode, mplayer, 
xine-libs, and k3b -- and that's with --as-needed in my LDFLAGS, without 
it things would be MUCH MUCH worse.

So if I'm not using a codec, or if I /might/ use it say once every year 
or two, it's DEFINITELY better to have that USE flag off and not have to 
deal with that codec triggering revdep-rebuilds every time it updates, 
and in the event that I DO run across somethign that needs it, just turn 
it on --single-shot, compile what I need with it, do what I want, then 
turn it off and emerge -N and revdep-rebuild to put everything back to 
not using it again.

Of course with the big DEs the effect is far bigger.  It's to the point 
where when looking for an app for some new purpose, if it's dependent on 
the desktop I don't run (GNOME, for me, but KDE for others), that's an 
almost insurmountable barrier to overcome to have me even try it, because 
the cost of continual maintenance of even the basics of an entire DE are 
simply way too high to be worth it for a single app or even two or three, 
unless they happen to be primary functionality for what I'm doing, of 
course, in which case may I'll switch DEs.  I long since settled on KDE 
apps for most of my primary functionality, and the cost of doing 
otherwise is high enough that's unlikely to change unless KDE or my own 
needs change enough that I dump KDE entirely.  (Actually, with kde4, a 
lot of folks have found just that, that KDE changed out from under them 
and no longer meets their needs, so they're switching away from it.  I 
came close, but had enough other reasons to stick with it that I found or 
in some cases scripted my own solutions to the missing or b0rk3n kde4 
functionality, so ended up sticking with it... but at enormous personal  
time and resources investment to do so... enough that comparably, paying 
a grand for MS software would be a reasonable tradeoff... if there 
weren't bigger issues I was worried about.)

But you didn't even mention the cost of continuing maintenance factor for 
all that "bloat", and on a Gentoo system, at least desktop, that's really 
the big one.

BTW, if you could, please turn off the HTML.  Some people find it 
troublesome to deal with.  You (or your client) do include plain text, 
which helps, but do you /really/ need the HTML, at the cost of making 
life harder for some readers?

-- 
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