Re: intel's Linux compiler w/ Debian
On Sun, 4 Aug 2002, Sean 'Shaleh' Perry wrote:
> >
> > A closer evaluation of the license is needed. Perhaps an installer
> > package in non-free? Perhaps someone could purchase the Intel compiler for
> > Debian and make it available for Debian packages? Perhaps the
> > non-commercial version could be used if Debian was said to be a
> > non-commercial organization? Perhaps, more discussion should be done in
> > debian-legal.
> >
> > Perhaps some Intel's other optimizing tools may be useful too?
> >
> > Oh, I haven't checked if these programs have been discussed before in
> > debian-devel or if a rfp was assigned.
> >
>
> why should we care? What is to gain by allowing users to compile items locally
> with the Intel compiler? The ONLY use I see for this is compatibility testing
> to see that you do not depend on gcc features unknowingly.
>
To see several people on this list say that they use Intel's C++ compiler
already is a good reason. I think the interest in non-free programs is
enough of a reason to make it available in non-free (well, technically
only the interest of a developer is a good reason to have it in non-free).
> Debian exists to promote free software. We exist to promote the right to
> freedom of software. Supporting non-free is something we do reluctantly. As a
> group we are pragmatic and realize that sometimes getting your job done is
> worth more than pride. For a long time it was impossible to function without
> Netscape and items like it. Those days are rapidly dwindling. Supplying hooks
> to proprietary apps (especially ones with a monetary cost) is counter
> productive and goes against what we stand for.
>
Well, there are lots of examples of violations of reluctance. I would like
to point out again that Intel's compilers may not have monetary costs,
that's something that I would like debian-legal to look at. I realize a
cost for commercial use would make it required to be either an installer
package in extra or something in non-free which would make it "not
Debian".
> To support a compiler for one architecture just because it might produce faster
> binaries seems fairly pointless. We do not ship pentium optimized programs
> because it has been demonstrated to be a fairly worthless procedure with more
> costs than benefit. Sure some programs do benefit and if the user desires
> pentium-builder is there to help.
>
I think it's worth supporting as an interesting program. It might produce
faster binaries, it might produce smaller binaries (usually both go hand
in hand, but not always), it can show where special gcc or alike options
have been used and it might compile things that gcc can't (there used to
be a long list of ANSI C++ standard components that did not work with
gcc).
Shipping binaries optimized for Pentiums was a different discussion, and I
agree that it's lessons should be used. If Intel's binaries do not work
better on all applicable arch's then it shouldn't be used to make binary
packages for the archive. I have not tested this, perhaps some developers
would like to though. Intel's compilers being part of the build systems
was only an idea, and I have to tend to agree now that it was likely a bad
idea. I still think that Intel's compilers should be made available.
> As I read this before sending I realize it sounds a little heated, flaming
> even. Please do not take my comments as a comment on you personally. This is
> a technical matter with technical arguments, ego should have nothing to do with
> it.
>
I see that too, but I hope that I have convinced you of the value of the
inclusion of Intel's compilers (or likely just an installer) in the
archives.
I for one would like to see all programs in the archive. It makes for a
good place to manage future free software (limited time on patents and
copyrights) and it makes it easier for people to install and use such
programs (alternatives are good imo). But this paragraph is a better
discussion for debian-qa and perhaps policy. I just wanted to express many
views on why to make Intel's compilers easily available to the Debian
community.
Pro's of including an installer/diff for Intel's compilers:
-Of interest to a few people on this list (likely more off of this list)
-Might produce faster binaries
-Might produce smaller binaries
-Can possibly show compiler specific features (like Linux uses gcc)
-Might compile code that other compilers can't (Intel specific features
perhaps or even standard features nyi in other compilers, see g++ and
ANSI's standards)
Con's:
-Takes up a small amount of space in the archive and BTS
-Creates a small amount of work for a Debian developer
-very likely not available free for commercial use
-not free to make binaries for distribution (a license for an organization
changes this)
I'm of the opinion that the Pro's outweigh the Con's. Developers might
find it useful for finding compiler features. Users may like to play with
optimizing their system. Students may want to compile code that uses
features unavailable in other compilers (I had this problem with gcc and
an ANSI standard a few years ago). Maybe, just maybe Intel's compilers
could be useful for building Debian packages (unlikely, but an interesting
possibility).
Drew Daniels
Reply to: