Bugzilla Bug 183393

Intel fortran cannot be installed.  The dev-lang/ifc-9.1.40 is
fetch-restricted, but at Intels homepage I cannot find ifc-9.1.40,
only version 9.1.43 or 10.something.  Maybe a version bump is needed.

Also, it would be VERY helpful if the ebuild (and the icc ebuild)
printed some kind of pointer to where the user can find the binary.
Most other fetch-restricted packages do that.

Best regards

Jakob


Reproducible: Always

Enough; this has been totally unmaintained for ages, it's not supported as a
compiler in Gentoo and generally makes baby jesus cry. Remove these two ebuilds
please.

(In reply to comment #1)
> Enough; this has been totally unmaintained for ages, it's not supported as a
> compiler in Gentoo and generally makes baby jesus cry. Remove these two ebuilds
> please.
> 

You are probably right, I am not even sure the ebuilds help, it may even be
easier to install the Intel installation files manually :-)

Perhaps the icc useflag should be removed at the same time?

Some of the science guys may protest, however, since performance may
dictate using an Intel compiler.  I personally need it badly, but
assume that it can be installed without an ebuild.

/Jakob

Or you could keep it in an ebuild overlay where you maintain the ebuild
yourself.

(In reply to comment #3)
> Or you could keep it in an ebuild overlay where you maintain the ebuild
> yourself.

Yeah, this would be much better place in some kind of overlay where interested
users can maintain this. Apparently there's no interest in this proprietary
package among Gentoo developers. 

(In reply to comment #0)
> Intel fortran cannot be installed.  The dev-lang/ifc-9.1.40 is
> fetch-restricted, but at Intels homepage I cannot find ifc-9.1.40,
> only version 9.1.43 or 10.something. 

hmm, i guess if you look a bit closer, you'll find it; at least i was able to
grab it from there a few weeks ago (theres a drop-down menu somewhere on intels
web-pages where you can specify the version you want).

(In reply to comment #5)

> hmm, i guess if you look a bit closer, you'll find it; at least i was able to
> grab it from there a few weeks ago (theres a drop-down menu somewhere on intels
> web-pages where you can specify the version you want).
> 

I did find a drop-down menu, but it only allowed me the choice between version
9.1 and version 10.  When I chose 9.1 i got 9.1.43, not 9.1.40.  So I just made
a new ebuild with the new version number, and installed it

> When I chose 9.1 i got 9.1.43, not 9.1.40.

please look more closely; at least when i choose 9.1 i get a second drop-down
menu where i can select 036 - 051...

however, this is off topic anyway...

(In reply to comment #7)

> please look more closely; at least when i choose 9.1 i get a second drop-down
> menu where i can select 036 - 051...

I am probably elsewhere on their many web pages.

> 
> however, this is off topic anyway...
> 

Not really, that was my original topic when I reported this bug: The
right file is hard/impossible to find, and there is no help from the
message the ebuild prints.

Somebody edited the title so it looks like I requested the ebuilds
removed.  I did not.

As someone using ifc for scientific (HEP) applications, I want to throw my
opinion on this into the ring: I haven't benchmarked F77 code, but if you want
to generate fast code from F90/95 source, I don't know of an (free) alternative
to ifc. After switching on a whole host of optimizations, gfortran generates
code that still is slower than fully optimized ifc code by a factor of at least
3, and g95 (which isn't in portage) is much worse. Also, there still are
glitches in both which prevent some F90 code from being compiled.

Installing ifc without an ebuild is awkward as Intel depends on RPM and
therefore, the automated installer script is not an option; you have to
manually extract the rpm contents and set them up in the correct place, pretty
much what the ebuild does.

So, I would very much prefer to keep at least ifc in portage, not so much
because someone might want to compile the few ebuilds that need a F77 compiler,
but as an important tool for those who have to develop fast F90 code. But of
course, I'm not the one who has to maintain it, so if finding a maintainer for
ifc is the only show-stopper, I would possibly volunteer :)

Gentleman,

please do not remove the icc from the portage tree. I've been using icc to
compile my whole Pentium-M Gentoo system and together with the tools from the
gentoo-wiki, it makes a great tool. 

I can't judge on the speed in an objective manner, but the Intel compiler has
some advantages over gcc especially when it comes to parallelization. On IA64
the difference between using gcc and icc will be significant (due to EPOC;
there I have numbers).

There is a document "Porting and Optimizing LHC Software on Itanium 2 Platform"
written in 2004 by "Michał Kapałka", who studied gcc and icc for some
scientific applications in their local HPC. At the time of his writing
apparently gcc 3.5 was the most current and he was indicating that it would not
be that easy to make the gcc fully utilizing all aspects of the EPOC nature of
Itanium2. Similar statements I heared from people working on the MIPS platform.

If I would have completed my Gentoo quiz I would even volunteer to maintain the
ebuild. ;)

Dirk.

I am taking icc and ifc maintainership and bumped them to 10.0.026 a month ago.
I sent a list to gentoo-dev a while back for possible problems, and contacted
upstream about licensing. 
icc and ifc are used in many scientific projects. So I won't remove them for
now. Closing as INVALID. Re-open if you really see an issue of having it in the
tree. 

Sébastien