Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.debian.maint.java > #8773 > unrolled thread

Trying to free f2c

Started byAndreas Tille <andreas@an3as.eu>
First post2016-01-29 11:20 +0100
Last post2016-01-29 19:30 +0100
Articles 8 — 3 participants

Back to article view | Back to linux.debian.maint.java


Contents

  Trying to free f2c Andreas Tille <andreas@an3as.eu> - 2016-01-29 11:20 +0100
    Re: Trying to free f2c Emmanuel Bourg <ebourg@apache.org> - 2016-01-29 12:00 +0100
      Re: Trying to free f2c Andreas Tille <andreas@an3as.eu> - 2016-01-29 13:50 +0100
        Re: Trying to free f2c Andreas Tille <andreas@an3as.eu> - 2016-01-29 16:50 +0100
          Re: Trying to free f2c Emmanuel Bourg <ebourg@apache.org> - 2016-01-29 17:00 +0100
            Re: Trying to free f2c Andreas Tille <andreas@an3as.eu> - 2016-01-29 18:30 +0100
              Re: Trying to free f2c "Fernando C. Estrada" <fcestrada@fcestrada.com> - 2016-01-29 18:40 +0100
              Re: Trying to free f2c Andreas Tille <andreas@an3as.eu> - 2016-01-29 19:30 +0100

#8773 — Trying to free f2c

FromAndreas Tille <andreas@an3as.eu>
Date2016-01-29 11:20 +0100
SubjectTrying to free f2c
Message-ID<qWejL-1ff-3@gated-at.bofh.it>
Hi,

since I'm really bored by the non-free f2c package which throws a
certain number of packages into contrib I checked again what we could
do.  It seems the main reason why f2c is non-free is that it includes
goto_trans (non-commercial use, no fee).  The code describes itself
as

JAVAB -- A prototype bytecode parallelization tool.
    This tool can be used to exploit loop parallelism directly at
    *bytecode-level* using the multi-threading mechanism of the JVM.
    In addition, the tool provides some elementary support for the
    automatic *detection* of implicit loop parallelism.

(see CVS[1]).  The code that translates fortran code into java is GPLed
and as far as I can see we need this functionality.  However, I can not
believe that these days Java compilers might need a tool that was
written in 2008 and has not changed like this to fullfill an
optimisation process that might be perfectly usable by a lot of
applications.

So my question is:  Do we possibly rely here on a non-free tool that
is really not needed and we could simply drop it?

Kind regards

     Andreas.

[1] http://f2j.cvs.sourceforge.net/viewvc/f2j/f2j/goto_trans/README?revision=1.1&view=markup

-- 
http://fam-tille.de

[toc] | [next] | [standalone]


#8774

FromEmmanuel Bourg <ebourg@apache.org>
Date2016-01-29 12:00 +0100
Message-ID<qWeWu-1zz-11@gated-at.bofh.it>
In reply to#8773
Le 29/01/2016 11:19, Andreas Tille a écrit :

> So my question is:  Do we possibly rely here on a non-free tool that
> is really not needed and we could simply drop it?

I got a quick look and you might be right, the goto_trans stuff seems
optional. What happens if you don't install javab in the f2j package?

Emmanuel Bourg

[toc] | [prev] | [next] | [standalone]


#8775

FromAndreas Tille <andreas@an3as.eu>
Date2016-01-29 13:50 +0100
Message-ID<qWgEV-33T-1@gated-at.bofh.it>
In reply to#8774
Hi Emmanuel,

On Fri, Jan 29, 2016 at 11:54:16AM +0100, Emmanuel Bourg wrote:
> > So my question is:  Do we possibly rely here on a non-free tool that
> > is really not needed and we could simply drop it?
> 
> I got a quick look and you might be right, the goto_trans stuff seems
> optional. What happens if you don't install javab in the f2j package?

I did a quick test with a local build.  I can easily build
libnetlib-java_0.9.3 fith f2c without the javab executable.  I have no
idea how to do proper performance tests, thought.  I have no idea
whether javab is used in the build process at all - may be I should
do some investigation of access timestamp.

In case your insight says that javab is not needed I'd consider the
following procedure:

   1. I create a tarball without non-free code
      (there is one other remaining file but PrintfFormat.java[1]
       does not sound as if could not be replaced by something else)
   2. Move this to pkg-java Git and create a package from this.

May be somebody can give hints according to performance tests to compare
libnetlib-java build with and without javab.

What do you think?

Kind regards

     Andreas.

[1] http://f2j.cvs.sourceforge.net/viewvc/f2j/f2j/util/org/j_paine/formatter/PrintfFormat.java?view=markup

-- 
http://fam-tille.de

[toc] | [prev] | [next] | [standalone]


#8777

FromAndreas Tille <andreas@an3as.eu>
Date2016-01-29 16:50 +0100
Message-ID<qWjt8-5fy-5@gated-at.bofh.it>
In reply to#8775
Hi again,

On Fri, Jan 29, 2016 at 01:44:14PM +0100, Andreas Tille wrote:
> > 
> > I got a quick look and you might be right, the goto_trans stuff seems
> > optional. What happens if you don't install javab in the f2j package?
> 
> I did a quick test with a local build.  I can easily build
> libnetlib-java_0.9.3 fith f2c without the javab executable.  I have no
> idea how to do proper performance tests, thought.  I have no idea
> whether javab is used in the build process at all - may be I should
> do some investigation of access timestamp.
> 
> In case your insight says that javab is not needed I'd consider the
> following procedure:
> 
>    1. I create a tarball without non-free code
>       (there is one other remaining file but PrintfFormat.java[1]
>        does not sound as if could not be replaced by something else)
>    2. Move this to pkg-java Git and create a package from this.

I implemented the plan in pkg-java git.  The only problem to build the
package is the now missing PrintfFormat.java[1] which now leads to

/usr/bin/javac -source 1.2 -target 1.2 -d obj org/j_paine/formatter/*.java
warning: [options] bootstrap class path not set in conjunction with -source 1.2
org/j_paine/formatter/Formatter.java:556: error: cannot find symbol
      s = new PrintfFormat(fmtstr).sprintf(o);
              ^
  symbol:   class PrintfFormat
  location: class FormatI
org/j_paine/formatter/Formatter.java:748: error: cannot find symbol
      s = new PrintfFormat(fmtstr).sprintf(o);
              ^
  symbol:   class PrintfFormat
  location: class FormatF
org/j_paine/formatter/Formatter.java:839: error: cannot find symbol
      s = new PrintfFormat(fmtstr).sprintf(o);
              ^
  symbol:   class PrintfFormat
  location: class FormatE
3 errors
1 warning
Makefile:11: recipe for target 'f2jutil.jar' failed


I wonder whether some Java programmer could provide some simple
replacement.  If this would be provided I'd upload to experimental first
and let follow the whole chain of dependencies as well to experimental.
The final consumer of all the mess will be the beast-mcmc package and
this one has performance tests that we could run to get a final decision.

Kind regards

         Andreas.

> [1] http://f2j.cvs.sourceforge.net/viewvc/f2j/f2j/util/org/j_paine/formatter/PrintfFormat.java?view=markup
[2] https://anonscm.debian.org/pkg-java/f2j.git


-- 
http://fam-tille.de

[toc] | [prev] | [next] | [standalone]


#8778

FromEmmanuel Bourg <ebourg@apache.org>
Date2016-01-29 17:00 +0100
Message-ID<qWjCN-5kt-3@gated-at.bofh.it>
In reply to#8777
Le 29/01/2016 16:44, Andreas Tille a écrit :

> I wonder whether some Java programmer could provide some simple
> replacement.

Sun did that for us 11 years ago in Java 5 ;) Try this replacement:

    s = String.format(fmtstr, o);


Emmanuel Bourg

[toc] | [prev] | [next] | [standalone]


#8780

FromAndreas Tille <andreas@an3as.eu>
Date2016-01-29 18:30 +0100
Message-ID<qWl1V-6td-35@gated-at.bofh.it>
In reply to#8778
On Fri, Jan 29, 2016 at 04:58:48PM +0100, Emmanuel Bourg wrote:
> Le 29/01/2016 16:44, Andreas Tille a écrit :
> 
> > I wonder whether some Java programmer could provide some simple
> > replacement.
> 
> Sun did that for us 11 years ago in Java 5 ;) Try this replacement:
> 
>     s = String.format(fmtstr, o);

What do I need to import to access this?

Sorry for the beginner question but Google points me to

   package com.javacodegeeks.core.string;

which seems wrong ...

       Andreas. 

-- 
http://fam-tille.de

[toc] | [prev] | [next] | [standalone]


#8781

From"Fernando C. Estrada" <fcestrada@fcestrada.com>
Date2016-01-29 18:40 +0100
Message-ID<qWlbB-6xj-23@gated-at.bofh.it>
In reply to#8780
Hi Andreas,

El Vie, 29 de Ene de 16, a las 11:20 AM, Andreas Tille escribió:
> On Fri, Jan 29, 2016 at 04:58:48PM +0100, Emmanuel Bourg wrote:
> > Le 29/01/2016 16:44, Andreas Tille a écrit :
> > 
> > > I wonder whether some Java programmer could provide some simple
> > > replacement.
> > 
> > Sun did that for us 11 years ago in Java 5 ;) Try this replacement:
> > 
> >     s = String.format(fmtstr, o);
> 
> What do I need to import to access this?

I guess nothing at all, it's currently included [0] in java.lang package
by default.

Regards,

[0]
https://docs.oracle.com/javase/1.5.0/docs/api/java/lang/String.html#format%28java.lang.String,%20java.lang.Object...%29
-- 
Fernando C. Estrada

[toc] | [prev] | [next] | [standalone]


#8782

FromAndreas Tille <andreas@an3as.eu>
Date2016-01-29 19:30 +0100
Message-ID<qWlXY-79P-13@gated-at.bofh.it>
In reply to#8780
On Fri, Jan 29, 2016 at 06:20:27PM +0100, Andreas Tille wrote:
> > 
> > Sun did that for us 11 years ago in Java 5 ;) Try this replacement:
> > 
> >     s = String.format(fmtstr, o);
> 
> What do I need to import to access this?

Forget this question - I've found

--- a/util/Makefile
+++ b/util/Makefile
@@ -5,7 +5,7 @@ JAR=jar

 UTIL_JAR=f2jutil.jar

-VER_TARGET=1.2
+VER_TARGET=1.5

 $(UTIL_JAR):
        mkdir -p $(OUTDIR)

as solution.

Sorry for the noise

      Andreas.

-- 
http://fam-tille.de

[toc] | [prev] | [standalone]


Back to top | Article view | linux.debian.maint.java


csiph-web