Re: test if there is a gain

[Arnaud Vandyck <avdyk@debian.org>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

(spring cleaning in my mailboxes... ;-))

29 Mar 2005 21:35:23 +0200, 
davidw@dedasys.com (David N. Welton) wrote: 

Hi David,

> "Michael Koch" <konqueror@gmx.de> writes:

Hi Michael,

>> I see no reason to imidiately switch everything to eclipse-ecj and
>> gcc-4.0 when it hits debian. We live from freedom of choice and mono
>> cultures are bad.

Which java compilers do we use?

1° Most packages use jikes;
2° Some packages (very few) in contrib uses non-free javac, but a lot of
   packages in contrib also uses jikes;
3° Less then 10 packages uses gcj

So moving from jikes to ecj is not a problem for me ;-)

I'll discuss of the gcj-4.0 native problem later.

> We've talked about this some before, but I'll put in my two cents here
> for "choice is ok, but sometimes too much choice is bad".  The
> difficult thing is defining "too much choice".  People are often
> worried about choices, especially in cases where a "wrong" choice
> might cost them invested time or money.  And they are even more
> worried when they don't have the tools or knowledge to evaluate that
> choice as well as an expert could.
>
> One of the reasons Ubuntu has been successful, IMO, is that there are
> fewer choices to make.  Experts have made them for you - although if
> you are an expert yourself, of course you can 'fix' any changes that
> you don't care for.

Ubuntu has been successful for a lot of reasons, I don't think fewer
choices was one of the reasons. Also, what is an expert and how can I
trust them? I can give you the name of five or four expert reading this
list and they'll tell you five different very good free JVM ;-)

About the native choice:
- ------------------------

I'm very sorry to read this thread so late. When I saw it I was busy and
was thinking: Whoaw too complicated to read after a day at work! ;-)
Well, that's my fault.

As Michael pointed out, we first need to have benchmarks to know if
there are real advantages to compile things to native. Also, we could
make choice: do we need to build the libs and the apps? only the apps?
Which one? Is it more important to build ant then tomcat? gjdoc then
eclipse? I don't know.

When we (Stefan Gybas and I) met Tom Tromey back in FOSDEM in 2003, we
talked with him after his presentation of gcj and we were really
impressed. We really would like to have this in Debian as soon as
possible. He told us to wait for the compatibility ABI (I think this is
now done with gcj-4.0). Now that it is done, we have to think about
several problems: the first is the gain (as Michael pointed), the second
is that when running a native java application, you can relay on a
unique VM (that's also the choice problem Michael pointed out).

Choosing a single compiler is not really a problem IMHO because we build
the packages from sources and it's very easy to change the compiler and
rebuild the package. Also, we have only one gcc! ;-)

But choosing the VM when there are a lot of different ones is not a good
thing for our users. I said when there are a lot of different ones
because we have a single perl interpreter, a single python
interpreter... but a lot of shells.

If we choose to build java packages to native, they will relay *only* on
gcj. Let's hope it'll never have RC bugs! The important thing is to keep
to ship the bytecode and when we'll have some benchmarks or some
positive feedbacks, we'll try to maybe add cdbs template or create an
'ant task'.

Thanks for your time,

- -- 
  .''`. 
 : :' :rnaud
 `. `'  
   `-    
Java Trap: http://www.gnu.org/philosophy/java-trap.html
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.0 (GNU/Linux)

iD8DBQFCcLAX4vzFZu62tMIRAr3tAJ4sKORVv2Rh7EfDvxwdEpmgHaL3agCfd23A
4CNmnyOdrTAUoPNU6LMBiYs=
=yl4m
-----END PGP SIGNATURE-----