Re: Getting working version?

[Paul Kienzle]

James Frye wrote:

> On Sat, 28 Dec 2002, John W. Eaton wrote:
>
>  
>
> > I doubt that this is really due to a bug in Octave (the current
> > sources compile cleanly for me with gcc 3.2 on a current Debian
> > system).
> >    
>
> I suppose this depends on one's definition of bug.  As someone trying to 
> compile source code, and finding that it won't compile with the same gcc 
> compiler version that I've been happily using for the last year or so... 
> well, that looks like a bug to me :-)
Yes a bit of grumbling is in order, much like you grumble about going to 
the dentist.  The changes are
painful, but they need to be done to avoid more pain in the future.  In 
particular, a number of things
won't compile on gcc 3.2 unless you bring them in line with the C++ 
standard, or alternatively choose
compiler options which supress the warnings about deprecated features.

The alternative to spending effort to develop new features is to spend 
effort ensuring compatibility in
a changing build environment, plus much ugliness with the preprocessor 
to hide syntax which has changed
since C++ was first proposed.

> > What compiler are you trying to use?  For the CVS version, I think I
> > already told you that you will probably need gcc 3.2.  If your
> > compiler doesn't understand ::isalnum, then I'd guess it is not
> > good enough to compile the current Octave sources.  But gcc 3.2 should
> > work and it is freely available, so you should probably upgrade if you
> > want to build Octave.
> >    
>
> gcc 2.95.  And why should the compiler understand ::isalnum?  I don't :-)
> > From context it would appear to be calling the standard C library function 
> isalnum, but why gunk it up in extra syntax when the plain ordinary 
> function should be available, probably with less overhead?
There is no overhead associated with using ::isalnum.  It is a 
compile-time thing which disambiguates
namespace references.  

::isalnum is very old syntax.  I can't believe that your compiler can't 
handle it.  It simply says use the
global namespace for this function rather than the function defined for 
the class.  I first used it in 1995.

As for gunking up the syntax, you can take it up with Stroustrup.

> > Are you referring to Octave sources, or something else?  I don't think
> > Octave relies on too many arcane langauge features, and I think my
> > development of Octave has been quite conservative with respect to new
> > langauge features.  
> >    
>
> Octave sources.  Look, I've been doing serious programming for a couple of
> decades now, and this is the only instance I can recall where I had to
> upgrade a compiler to get something to compile.  (Not that it's perfect
> even now: with gcc 3.2.1, I get messages about "Dwarf errors".  The make
> builds and installs the executable, which runs at least some simple tests,
> but now barfs on calling eval in the .m files I want to use.)
Then you've lived in a very sheltered world.  Every try running some 
fortran 95 code through a fortran
77 compiler?  How about C99 code through an old C compiler?  How about 
matlab 5 code in matlab 4?
Or for that matter, matlab 4 code in matlab 5?

By choosing to follow standards, Octave can hope one day to be able to 
compile on a variety of
compilers.  If instead it stayed with g++ deprecated syntax and 
libraries then it would forever be
tied to running on older versions of g++.

> > Even now, you can't expect that any serious C++ project
> > will compile with any and all C++ compilers.  Someday, maybe, but for
> > now the compilers still seem to be catching up to the standard, so
> > these kinds of problems are likely to be with us for a while longer.
> >    
>
> Pardon my flaming here, but isn't that a somewhat backwards attitude?  
> Writing code for some mythical handed-down-from-on-high standard, and then
> complaining that existing compilers just aren't good enough to compile it?  
> C++, or any computer language, is a tool, not a frigging religion.  You
> don't get brownie points for the purity of your code.  All us agnostic
> user types out here care about is whether it runs or not :-)
You are missing the point: existing compilers can compile it. You just 
don't happen to have them
installed.

I don't remember the exact details, but IIRC the alternative to moving 
to gcc 3.2 was to reimplement
some code that is in STL so that octave could be extended.  That sounds 
rather like re-inventing the
wheel for the religion of supporting out of date compilers.

I grant that in this case not being able to compile on the largest 
installed base of linux (red hat 7.x)
is a questionable decision (and I did question it myself), but if it is 
a choice between progress on
octave and support for the previous version of red hat, I choose 
progress with minimal grumbling.  
FWIW, my main  development machine is still running octave-2.1.38 
because I don't have space
to build a new compiler.

John, maybe we could indicate the new compiler requirements on the 
download page, with a suggestion
to use 2.1.38 for older compilers?

> > As it is, I think you've already received quite a lot of support for
> > free (I implemented "end" and local functions just recently, in part
> > because of your requests).  
> >    
>
> Though I asked for only a small fraction of it.  Remember that my requests
> were not for you or anyone else to change octave code, just for
> suggestions on how to work around the differences between it and Matlab.
Which I gave you.

> > Would you care to give something back to 
> > the probject (code, or perhaps funding to help pay for the cost of
> > providing the new features you requested or to make binary packages 
> > available), or would you just like to complain about what you're
> > getting for free?
> >    
>
> Well, it hasn't been exactly free on my end.  Say 20 hours of work over
> the last week, plus extra trips into the lab so I could e.g. download some
> 24 MBytes of gcc 3.2.1.  At the very least, I've contributed some
> practical compatability testing with real-world Matlab scripts :-)
I started writing a compatibility guide a while ago.  See 
http://octave.sf.net/compatibility.html.  Feel
free to send updates, and maybe save someone like yourself 20 hours of 
work in the future.

It is a little out of date just now since octave is rapidly changing and 
I'm not actively porting
matlab code.

Paul Kienzle
address@hidden




-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------