[gentoo-dev] Default USE changes for fortran and mudflap?

[gentoo-dev] Default USE changes for fortran and mudflap?

[Ryan Hill]

[-- Attachment #1: Type: text/plain, Size: 1037 bytes --]

While I'm adding USE defaults to toolchain.eclass and moving them out of the
profiles, I thought now would be a good time to review a couple default flag
settings.

mudflap:
This is currently enabled by default but I'd like to disable it.  It controls
libmudflap and the -fmudflap flag.  I think the only reason this flag exists is
so we can disable it in crossdev.  It's not required by anything in the tree,
the code is bitrotten and has been removed for GCC 4.9. If you know how to use
-fmudflap, you know how to set a USE flag.  

fortran:
Do we want to keep enabling fortran by default?  The majority of users will
never get the urge to install a fortran package, and the fortran eclass handles
those that do.  I think it should be treated as all the other optional
languages and disabled by default, but I'd like to know if there are other
opinions.


-- 
Ryan Hill                        psn: dirtyepic_sk
   gcc-porting/toolchain/wxwidgets @ gentoo.org

47C3 6D62 4864 0E49 8E9E  7F92 ED38 BD49 957A 8463

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 490 bytes --]

Re: [gentoo-dev] Default USE changes for fortran and mudflap?

[Michał Górny]

[-- Attachment #1: Type: text/plain, Size: 667 bytes --]

Dnia 2014-01-12, o godz. 01:53:47
Ryan Hill <dirtyepic@gentoo.org> napisał(a):

> fortran:
> Do we want to keep enabling fortran by default?  The majority of users will
> never get the urge to install a fortran package, and the fortran eclass handles
> those that do.  I think it should be treated as all the other optional
> languages and disabled by default, but I'd like to know if there are other
> opinions.

Well, I'd say we should work on making 'other languages' buildable
without rebuilding the whole giant gcc stack. Especially that
the stacked build makes it impossible to use distcc at least partially.

-- 
Best regards,
Michał Górny

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 966 bytes --]

Re: [gentoo-dev] Default USE changes for fortran and mudflap?

[Pacho Ramos]

El dom, 12-01-2014 a las 09:24 +0100, Michał Górny escribió:
> Dnia 2014-01-12, o godz. 01:53:47
> Ryan Hill <dirtyepic@gentoo.org> napisał(a):
> 
> > fortran:
> > Do we want to keep enabling fortran by default?  The majority of users will
> > never get the urge to install a fortran package, and the fortran eclass handles
> > those that do.  I think it should be treated as all the other optional
> > languages and disabled by default, but I'd like to know if there are other
> > opinions.
> 
> Well, I'd say we should work on making 'other languages' buildable
> without rebuilding the whole giant gcc stack. Especially that
> the stacked build makes it impossible to use distcc at least partially.
> 

I was also wondering about what we prefer if Michal's suggestion is not
possible:
- Build support for other langs by default -> More time when we need to
emerge gcc or update it but less time as we don't need to rebuild for
that (I remember when I needed to build pdftk and needed to wait for a
gcc rebuild giving gcj support)
- Build support for other langs disabled by default -> Well, does it
save so much time to compensate the rebuilds we would need to do in the
future when we need to emerge some package some time? 

[gentoo-dev] Re: Default USE changes for fortran and mudflap?

[Ryan Hill]

[-- Attachment #1: Type: text/plain, Size: 991 bytes --]

On Sun, 12 Jan 2014 09:40:01 +0100
Pacho Ramos <pacho@gentoo.org> wrote:

> I was also wondering about what we prefer if Michal's suggestion is not
> possible:
> - Build support for other langs by default -> More time when we need to
> emerge gcc or update it but less time as we don't need to rebuild for
> that (I remember when I needed to build pdftk and needed to wait for a
> gcc rebuild giving gcj support)
> - Build support for other langs disabled by default -> Well, does it
> save so much time to compensate the rebuilds we would need to do in the
> future when we need to emerge some package some time? 

I think for most people the number of times they've upgraded gcc far outweighs
the number of times they've had to rebuild it to install a fortran package.
We should optimize for the common case.


-- 
Ryan Hill                        psn: dirtyepic_sk
   gcc-porting/toolchain/wxwidgets @ gentoo.org

47C3 6D62 4864 0E49 8E9E  7F92 ED38 BD49 957A 8463

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 490 bytes --]

[gentoo-dev] Re: Default USE changes for fortran and mudflap?

[Ryan Hill]

[-- Attachment #1: Type: text/plain, Size: 1653 bytes --]

On Sun, 12 Jan 2014 09:24:20 +0100
Michał Górny <mgorny@gentoo.org> wrote:

> Dnia 2014-01-12, o godz. 01:53:47
> Ryan Hill <dirtyepic@gentoo.org> napisał(a):
> 
> > fortran:
> > Do we want to keep enabling fortran by default?  The majority of users will
> > never get the urge to install a fortran package, and the fortran eclass
> > handles those that do.  I think it should be treated as all the other
> > optional languages and disabled by default, but I'd like to know if there
> > are other opinions.
> 
> Well, I'd say we should work on making 'other languages' buildable
> without rebuilding the whole giant gcc stack. Especially that
> the stacked build makes it impossible to use distcc at least partially.

Bootstrapping makes distcc impossible, and you can't bootstrap these days
without building C and C++.  Even if you're not bootstrapping, the back and
middle ends are shared.  You have to build them to build the front-ends.  Maybe
you could cut out a couple of the target libraries, so you're really not gaining
much.

If the build times really get to you then you could try something like:

GCC_MAKE_TARGET="all" EXTRA_ECONF="--disable-bootstrap" \
  emerge -av1 sys-devel/gcc:4.8

distcc should work for most of the build (the target libs must be built with
the just-built compiler so that can't be distributed).

Anyways, assuming we can't split out fortran, would you be for or against
enabling it by default?  I'm good either way.


-- 
Ryan Hill                        psn: dirtyepic_sk
   gcc-porting/toolchain/wxwidgets @ gentoo.org

47C3 6D62 4864 0E49 8E9E  7F92 ED38 BD49 957A 8463

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 490 bytes --]

Re: [gentoo-dev] Re: Default USE changes for fortran and mudflap?

[Michał Górny]

[-- Attachment #1: Type: text/plain, Size: 2149 bytes --]

Dnia 2014-01-12, o godz. 03:50:53
Ryan Hill <dirtyepic@gentoo.org> napisał(a):

> On Sun, 12 Jan 2014 09:24:20 +0100
> Michał Górny <mgorny@gentoo.org> wrote:
> 
> > Dnia 2014-01-12, o godz. 01:53:47
> > Ryan Hill <dirtyepic@gentoo.org> napisał(a):
> > 
> > > fortran:
> > > Do we want to keep enabling fortran by default?  The majority of users will
> > > never get the urge to install a fortran package, and the fortran eclass
> > > handles those that do.  I think it should be treated as all the other
> > > optional languages and disabled by default, but I'd like to know if there
> > > are other opinions.
> > 
> > Well, I'd say we should work on making 'other languages' buildable
> > without rebuilding the whole giant gcc stack. Especially that
> > the stacked build makes it impossible to use distcc at least partially.
> 
> Bootstrapping makes distcc impossible, and you can't bootstrap these days
> without building C and C++.  Even if you're not bootstrapping, the back and
> middle ends are shared.  You have to build them to build the front-ends.  Maybe
> you could cut out a couple of the target libraries, so you're really not gaining
> much.

Yes but if C & C++ was installed after it is built, distcc/ccache could
be then used to build the Fortran & GCJ & so on. Of course, there's
another matter of matching gcc versions between hosts but that's just
another problem that needs addressing elsewhere.

Maybe I wouldn't gain much of Fortran. But if I recall correctly, gcj
has a fair bit of C++ code to build.

But after all, it's all purely theoretical. Unless we're going to patch
gcc build system to make it more friendly.

> Anyways, assuming we can't split out fortran, would you be for or against
> enabling it by default?  I'm good either way.

Considering that it saves you like 5M? I'd rather keep it enabled. 5M
won't save the day, while waiting extra 2 hours for gcc rebuild because
some package needs Fortran is irritating. And I say two hours because,
say, I had to enable USE=gcj earlier and now I have to rebuild that
huge thing.

-- 
Best regards,
Michał Górny

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 966 bytes --]

Re: [gentoo-dev] Default USE changes for fortran and mudflap?

[Andreas K. Huettel]

[-- Attachment #1: Type: Text/Plain, Size: 1117 bytes --]

Am Sonntag, 12. Januar 2014, 08:53:47 schrieb Ryan Hill:
> fortran:
> Do we want to keep enabling fortran by default?  The majority of users will
> never get the urge to install a fortran package, and the fortran eclass
> handles those that do.  I think it should be treated as all the other
> optional languages and disabled by default, but I'd like to know if there
> are other opinions.

The problem with fortran is (was?) a bit that it creeps in in unexpected 
places via scientific libraries (e.g. image processing). Not so far in the 
past it was necessary to build media-gfx/digikam for example...

I don't like the long gcc build times either, but would recommend keeping it 
enabled by default, for a simple reason: We do not have a clean way to depend 
on it (gcc[fortran] dependencies are pointless because of slotting and gcc-
config), and an eclass can only output errors and ask for manual intervention. 
While just about everywhere else we try to automatize as much as possible...

-- 

Andreas K. Huettel
Gentoo Linux developer 
dilfridge@gentoo.org
http://www.akhuettel.de/


[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 966 bytes --]

[gentoo-dev] Re: Default USE changes for fortran and mudflap?

[Ryan Hill]

[-- Attachment #1: Type: text/plain, Size: 2362 bytes --]

On Sun, 12 Jan 2014 11:08:18 +0100
Michał Górny <mgorny@gentoo.org> wrote:

> Dnia 2014-01-12, o godz. 03:50:53
> Ryan Hill <dirtyepic@gentoo.org> napisał(a):

> > Bootstrapping makes distcc impossible, and you can't bootstrap these days
> > without building C and C++.  Even if you're not bootstrapping, the back and
> > middle ends are shared.  You have to build them to build the front-ends.
> > Maybe you could cut out a couple of the target libraries, so you're really
> > not gaining much.
 
> Yes but if C & C++ was installed after it is built, distcc/ccache could
> be then used to build the Fortran & GCJ & so on. Of course, there's
> another matter of matching gcc versions between hosts but that's just
> another problem that needs addressing elsewhere.

Unfortunately not, at least for ccache.  When a new C compiler is installed, it
invalidates the cache that was generated by the previous compiler while building
it because ccache hashes the compiler's mtime.  Furthermore in stage2/3 you
can't use either ccache or distcc.  ccache because the mtime has changed again,
the command line changes between stages (stage1 is built with minimal flags
for speed, stage 2 with normal flags, and stage 3 with -gtoggle to make sure
the addition of debug info doesn't change code generation).  The build system
also bypasses ccache during stage2/3 by directly calling the binary in the
build tree (after all, using cached data between stages would kind of defeat
the purpose of bootstrapping). I'm less familiar with distcc but if the remote
host doesn't have a copy of the just-built compiler I don't think there's much
it can do.

> Considering that it saves you like 5M? I'd rather keep it enabled. 5M
> won't save the day, while waiting extra 2 hours for gcc rebuild because
> some package needs Fortran is irritating. And I say two hours because,
> say, I had to enable USE=gcj earlier and now I have to rebuild that
> huge thing.

Okay, good example.  No[1] one[2] likes[3] building gcj.

[1] http://gcc.gnu.org/ml/gcc/2008-06/msg00425.html
[2] http://gcc.gnu.org/ml/gcc/2010-10/msg00506.html
[3] http://gcc.gnu.org/ml/gcc/2013-11/msg00153.html


-- 
Ryan Hill                        psn: dirtyepic_sk
   gcc-porting/toolchain/wxwidgets @ gentoo.org

47C3 6D62 4864 0E49 8E9E  7F92 ED38 BD49 957A 8463

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 490 bytes --]

Re: [gentoo-dev] Default USE changes for fortran and mudflap?

[Michael Orlitzky]

On 01/12/2014 02:53 AM, Ryan Hill wrote:
> fortran:
> Do we want to keep enabling fortran by default?  The majority of users will
> never get the urge to install a fortran package, and the fortran eclass handles
> those that do.  I think it should be treated as all the other optional
> languages and disabled by default, but I'd like to know if there are other
> opinions.

A lot of "popular" fortran code is old scientific stuff without a build
system, license, upstream etc. that isn't going to be in the tree for
that reason. Some of these have only "install gcc" for build
instructions, so it might cause confusion for a few minutes if "gcc"
can't compile fortran. The only popular modern example I have on hand is
sage[1].

The two easy choices for what-to-do are,

  * Disable everything by default.

  * USE-default everything to match GCC's default --enable-languages[2]
    (c, c++, fortran, java, objc)

Either one is fine. A one-time gcc rebuild for fortran is not a big deal
and at that point is probably the least of your problems.


[1] http://www.sagemath.org/
[2] http://gcc.gnu.org/install/configure.html

Re: [gentoo-dev] Default USE changes for fortran and mudflap?

[Matt Turner]

On Sat, Jan 11, 2014 at 11:53 PM, Ryan Hill <dirtyepic@gentoo.org> wrote:
> While I'm adding USE defaults to toolchain.eclass and moving them out of the
> profiles, I thought now would be a good time to review a couple default flag
> settings.
>
> mudflap:
> This is currently enabled by default but I'd like to disable it.  It controls
> libmudflap and the -fmudflap flag.  I think the only reason this flag exists is
> so we can disable it in crossdev.  It's not required by anything in the tree,
> the code is bitrotten and has been removed for GCC 4.9. If you know how to use
> -fmudflap, you know how to set a USE flag.

Sounds good.

> fortran:
> Do we want to keep enabling fortran by default?  The majority of users will
> never get the urge to install a fortran package, and the fortran eclass handles
> those that do.  I think it should be treated as all the other optional
> languages and disabled by default, but I'd like to know if there are other
> opinions.

I'm for removing fortran by default. Disabling fortran is always one
of the first things I do on a new install. I've disabled it by default
in the mips profile already.

Re: [gentoo-dev] Default USE changes for fortran and mudflap?

[Matthew Summers]

On Sun, Jan 12, 2014 at 1:53 AM, Ryan Hill <dirtyepic@gentoo.org> wrote:
>
> fortran:
> Do we want to keep enabling fortran by default?  The majority of users will
> never get the urge to install a fortran package, and the fortran eclass handles
> those that do.  I think it should be treated as all the other optional
> languages and disabled by default, but I'd like to know if there are other
> opinions.
>

If you do disable fortran by default, please make a news item, if you
don't mind. I know this will effect some percentage of our scientific
user base (including myself).

Thanks!
Matt

Matthew Summers
Gentoo Foundation Inc.
GPG: 111B C438 35FA EDB5 B5D3 736F 45EE 5DC0 0878 9D46

[gentoo-dev] Re: Default USE changes for fortran and mudflap?

[Steven J. Long]

On Sun, Jan 12, 2014 at 01:53:47AM -0600, Ryan Hill wrote:
> While I'm adding USE defaults to toolchain.eclass and moving them out of the
> profiles, I thought now would be a good time to review a couple default flag
> settings.
> 
> mudflap:
> This is currently enabled by default but I'd like to disable it.  It controls
> libmudflap and the -fmudflap flag.  I think the only reason this flag exists is
> so we can disable it in crossdev.  It's not required by anything in the tree,
> the code is bitrotten and has been removed for GCC 4.9. If you know how to use
> -fmudflap, you know how to set a USE flag.  

No-brainer, yeah.

> fortran:
> Do we want to keep enabling fortran by default?  The majority of users will
> never get the urge to install a fortran package, and the fortran eclass handles
> those that do.  I think it should be treated as all the other optional
> languages and disabled by default, but I'd like to know if there are other
> opinions.

Yes, keep it. It's used in the oddest places, and still beats C for numeric
processing. It's not like gcj which is a pig to build, and to which there are
many alternative implementations that may well be preferred, given the state
of Java. IMO it's important to have, and there's no real benefit to keeping
it off, for the general user. Anyone who wants to keep it slim already has it
disabled in package.use.

-- 
#friendly-coders -- We're friendly, but we're not /that/ friendly ;-)

[gentoo-dev] Re: Default USE changes for fortran and mudflap?

[Ryan Hill]

[-- Attachment #1: Type: text/plain, Size: 571 bytes --]

On Sun, 12 Jan 2014 01:53:47 -0600
Ryan Hill <dirtyepic@gentoo.org> wrote:

> While I'm adding USE defaults to toolchain.eclass and moving them out of the
> profiles, I thought now would be a good time to review a couple default flag
> settings.

Okay, we'll be dropping fortran from the profiles and enabling it as a default
USE flag for gcc as requested in bug #372663.  Thanks for the feedback.


-- 
Ryan Hill                        psn: dirtyepic_sk
   gcc-porting/toolchain/wxwidgets @ gentoo.org

47C3 6D62 4864 0E49 8E9E  7F92 ED38 BD49 957A 8463

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 490 bytes --]

Re: [gentoo-dev] Default USE changes for fortran and mudflap?

[Donnie Berkholz]

[-- Attachment #1: Type: text/plain, Size: 690 bytes --]

On 01:53 Sun 12 Jan     , Ryan Hill wrote:
> fortran:
> Do we want to keep enabling fortran by default?  The majority of users 
> will never get the urge to install a fortran package, and the fortran 
> eclass handles those that do.  I think it should be treated as all the 
> other optional languages and disabled by default, but I'd like to know 
> if there are other opinions.

It's essentially only used by the small minority of people installing 
sci-* packages so I'd be fine with that. Reasonable defaults FTW.

-- 
Thanks,
Donnie

Donnie Berkholz
Council Member / Sr. Developer, Gentoo Linux <http://dberkholz.com>
Analyst, RedMonk <http://redmonk.com/dberkholz/>

[-- Attachment #2: Type: application/pgp-signature, Size: 966 bytes --]

Re: [gentoo-dev] Default USE changes for fortran and mudflap?

[Mike Frysinger]

[-- Attachment #1: Type: Text/Plain, Size: 894 bytes --]

On Sunday 12 January 2014 02:53:47 Ryan Hill wrote:
> While I'm adding USE defaults to toolchain.eclass and moving them out of
> the profiles, I thought now would be a good time to review a couple
> default flag settings.
> 
> mudflap:
> This is currently enabled by default but I'd like to disable it.  It
> controls libmudflap and the -fmudflap flag.  I think the only reason this
> flag exists is so we can disable it in crossdev.  It's not required by
> anything in the tree, the code is bitrotten and has been removed for GCC
> 4.9. If you know how to use -fmudflap, you know how to set a USE flag.

it was the hot new checker when it was first added, but it never really caught 
on and asan is the way to go now.  i occurred to me a few weeks ago to turn it 
off by default after watching another cross-compile go by with it on, but looks 
like you got it covered.
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]