using 10.1 with (always custom compiled) latest 2.6.x kernel. for some
reasons i would like to upgrade gcc-* packages to 3.4.3 or 3.4.4. i intend
to compile it from source... using existing 3.3.4, and (perhaps modified)
buildscripts that are in in /testing.
anybody had problems compared to 3.3.4? especially with 3.4.4 and compiling
kernel? do i have to upgrade/recompile anything else? and should it
(positivly, i hope) affect kernel performance? (not the upgrade reason, but
could be nice)

Mad-Eye Moody wrote:

> using 10.1 with (always custom compiled) latest 2.6.x kernel. for some
> reasons i would like to upgrade gcc-* packages to 3.4.3 or 3.4.4. i intend
> to compile it from source... using existing 3.3.4, and (perhaps modified)
> buildscripts that are in in /testing.
> anybody had problems compared to 3.3.4? especially with 3.4.4 and
> compiling kernel? do i have to upgrade/recompile anything else? and should
> it (positivly, i hope) affect kernel performance? (not the upgrade reason,
> but could be nice)

If you build GCC with 'default' configure, it will live in it's own
directory (/usr/local/) - so if you set your $PATH to include that first,
it will be default. In any event of problems you can go back to the old.
You can then also specify in Makefiles what GCC you wish to use by gcc
= /path/to/which/gcc you need.

You should need not recompile anything - just build the new kernel as you
always did.

I am on 3.4.4 here (Slack 10) with no ill effects.

Nick

Nick <nick@ukfsn.org> wrote:
>Mad-Eye Moody wrote:
>
>> using 10.1 with (always custom compiled) latest 2.6.x kernel. for some
>> reasons i would like to upgrade gcc-* packages to 3.4.3 or 3.4.4. i intend
>> to compile it from source... using existing 3.3.4, and (perhaps modified)
>> buildscripts that are in in /testing.
>> anybody had problems compared to 3.3.4? especially with 3.4.4 and
>> compiling kernel? do i have to upgrade/recompile anything else? and should
>> it (positivly, i hope) affect kernel performance? (not the upgrade reason,
>> but could be nice)
>
>If you build GCC with 'default' configure, it will live in it's own
>directory (/usr/local/) - so if you set your $PATH to include that first,
>it will be default. In any event of problems you can go back to the old.
>You can then also specify in Makefiles what GCC you wish to use by gcc
>= /path/to/which/gcc you need.

While that is true, I have commonly used a different method. I
haven't looked at those two versions of GCC, but since this has
been the intended method for some years I would expect it to be
valid with those two also. It is significantly more versatile
because it can be extended to a large number of gcc versions.

For any given version, when compiled all parts except /gcc/ and
/cpp/ will have distinct paths that differ from one version to
another. Hence, on one box here I have all of the following:

/usr/bin/egcs-2.91.66
/usr/bin/gcc-2.95.3
/usr/bin/gcc-3.0
/usr/bin/gcc-3.3
/usr/bin/gcc-3.3.5

Each can be invoked specifically by name, or any one of them can
be symlinked to /usr/bin/gcc. They each use a completely
distinct configuration file, code generators, libraries, etc.

The one difficulty with the above is keeping /cpp/ versions
distinct, and I have made no attempt at doing that (other than
usually, but not always, saving the previous versions), and
instead have just used the latest /cpp/ with all compiler
versions. So far that has never caused a problem for my
particular uses, but for some it might be significant and would
require attention.

>You should need not recompile anything - just build the new kernel as you
>always did.
>
>I am on 3.4.4 here (Slack 10) with no ill effects.
>
>Nick

--
Floyd L. Davidson <http://web.newsguy.com/floyd_davidson>
Ukpeagvik (Barrow, Alaska) floyd@barrow.com

On Sat, 11 Jun 2005 16:44:26 +0200, Mad-Eye Moody wrote:

> using 10.1 with (always custom compiled) latest 2.6.x kernel. for some
> reasons i would like to upgrade gcc-* packages to 3.4.3 or 3.4.4. i intend
> to compile it from source... using existing 3.3.4, and (perhaps modified)
> buildscripts that are in in /testing.
> anybody had problems compared to 3.3.4? especially with 3.4.4 and compiling
> kernel? do i have to upgrade/recompile anything else? and should it
> (positivly, i hope) affect kernel performance? (not the upgrade reason, but
> could be nice)

well, just want to report that experiment went succesful
so, gcc 3.4.4 builds succesful with 3.4.3 script (number changed only)
from /testing. everything i wanted to compile/recompile so far (including
kernel, alsa, rt2500 drivers, ati prop drivers), i did with complete
success.

Mad-Eye Moody <zhillaREMOVEBRA@spymac.com> wrote:
>On Sat, 11 Jun 2005 16:44:26 +0200, Mad-Eye Moody wrote:
>
>> using 10.1 with (always custom compiled) latest 2.6.x kernel. for some
>> reasons i would like to upgrade gcc-* packages to 3.4.3 or 3.4.4. i intend
>> to compile it from source... using existing 3.3.4, and (perhaps modified)
>> buildscripts that are in in /testing.
>> anybody had problems compared to 3.3.4? especially with 3.4.4 and compiling
>> kernel? do i have to upgrade/recompile anything else? and should it
>> (positivly, i hope) affect kernel performance? (not the upgrade reason, but
>> could be nice)
>
>well, just want to report that experiment went succesful
>so, gcc 3.4.4 builds succesful with 3.4.3 script (number changed only)
>from /testing. everything i wanted to compile/recompile so far (including
>kernel, alsa, rt2500 drivers, ati prop drivers), i did with complete
>success.

Now, the question is, does it run faster and/or is it otherwise
smaller?

I suspect that if you recompile everything with gcc 2.95 it
would actually be an improvement. Part of the difference
between gcc 2.x and 3.x is of course the C99 changes to the ISO
C Standard, which improve the language, but probably do not
improve Linux (in particular because the Linux kernel is not
using the newer features anyway). On the other hand, changes
from gcc 3.3 to 3.4 might improve Linux??

Interesting concepts, but *I* am not going to the effort to try
measuring the differences! But I'd sure be interested in the
results if someone ever does.

--
Floyd L. Davidson <http://web.newsguy.com/floyd_davidson>
Ukpeagvik (Barrow, Alaska) floyd@barrow.com