[Debian-astro-maintainers] Bug#852682: [gnudatalanguage] Bug with defining arrays

[Ole Streicher]

Package: gnudatalanguage
Version: 0.9.7-1
Severity: important
Forwarded: https://sourceforge.net/p/gnudatalanguage/mailman/message/35621280/

The following bug report also applies to the Debian package:

From: Alain Coulais <alain.coulais at obspm.fr>
Subject: [gnudatalanguage-devel] Fw: [[gdl-announces]] gdl 0.9.7 delivered; note on C compilers flags
To: <gnudatalanguage-devel at lists.sourceforge.net>, Andy Gallagher <andrew.gallagher at obspm.fr>
Date: Mon, 23 Jan 2017 18:57:02 +0100

This is a very serious issue

Sorry

----- Message Transféré -----

Date: Mon, 23 Jan 2017 14:53:11 +0100
De: Andy Gallagher <andrew.gallagher at obspm.fr>
À: Alain Coulais <alain.coulais at obspm.fr>
Sujet: Re: [[gdl-announces]] gdl 0.9.7 delivered; note on C compilers flags

Hi Alain,

I've already found a bug with defining arrays.

Suppose I want to define a 3 dimensional array:

test = fltarr(nx, ny, nz)
or
test = dblarr(nx,ny,nz)

where nx=1, ny=1, and nz=114.

In IDL and the older versions of GDL, this defined an array:

IDL> help,test

TEST            DOUBLE    = Array[1, 1, 114]

**

TEST            FLOAT     = Array[1, 1, 114]


But in GDL 0.9.7:

GDL> test=dblarr(nx,ny,nz)

GDL> help,test

TEST            DOUBLE    = Array[1, 114]

**

TEST            FLOAT     = Array[1, 114]


The same problem occurs when we try to convert a 3 dimensional double 
array to a float array using FLOAT.

This is causing major problems with our spectrum synthesis code, 
Linfor3D. For the moment, we must run version 0.9.5, as this was the 
last compatible version, however, I would like to see version 0.9.7 run 
Linfor3D, as it is getting difficult to distribute 0.9.5 to all its 
users worldwide. It also would be easier if they could simply download 
the tarball from the CVS database and run that, rather than asking me 
for it.

Best wishes and happy new year
Andy

On 22/01/17 00:05, Alain Coulais wrote:
> Hi GDL users and contributors,
>
> GDL 0.9.7 was just delivered. This version is very close to the CVS
> version and the version (0.9.7v1) mentioned in the previous message on
> this list.
>
> Two very interesting aspects in 0.9.7 : very fast X11 output through
> network, experimental --fussy option ("gdl -h" for details)
>
> This version has been extensively tested (make check)
> on various OS and various arch (Mageia, Debian 7 and 8, 32 & 64,
> Fedora, CentOS 5.11 & 6, Ubuntu 14.04, 15.10, 16.04, OSX 10.9, 10.10
> and 10.11 ...).
>
> Can be compiled with GCC 4.8, 4.9, 5.3, 6.0, ICC 16, Clang 6
> !!! Take care on compilers options, see note below !!!
>
> Please remember you can easily add 3 useful libraries (collection
> of pro files) thanks to Debian packagers : Astron lib, Coyote and
> Mpfit https://packages.debian.org/sid/astro-gdl  (git clone ...)
>
> The minimal_script.sh was updated too
> http://gnudatalanguage.cvs.sourceforge.net/viewvc/gnudatalanguage/gdl/scripts/minimum_script4gdl.sh
>
> We welcome feedbacks, bugs reports, contributions (pro files or
> corrections, c++ codes as long as it is GPL compliant, regressions
> tests (see testsuite/) ... but it is better to post on SF, eg :
> https://sourceforge.net/p/gnudatalanguage/discussion/
>
>
> Since GDL 0.9.7 can be compiled very easily on most "recent" OS,
> we strongly prefer to provide feedbacks on this version
> (and not on 0.9.6 or 0.9.5 ...)
>
>
> with best regards
>
> Alain C. for the GDL team
>
>
> Note on compilation options (C Compilers flags)
>
> One important note : we notice that the **default** compilation
> with ICC or GCC 5 or GDL 6 do not switch on the openmp options
> and the performances are not good on multi-cores
> (thanks to Eigen3 we are faster than IDL on most matrix operations
> since years on multi-cores)
> You can test that with "bench_matrix_multiply.pro" in
> testsuite/benchmark/
>
> Eg: with GCC 4.6
> option -DCMAKE_CXX_FLAGS_RELEASE='*-Ofast -march=native*'
> is way better than :
>
> with GCC 5.3, option
> cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_FLAGS_RELEASE='-Ofast
> -march=native'
> is much better than the default one
>
>
>

-- 
============================================================================
Dr Andy Gallagher
Observatoire de Paris fellow
Laboratoire d'Etudes des Galaxies, Etoiles, Physique et Instrumentation
GEPI Observatoire de Paris, UMR 8111 du CNRS, Univ. Paris-Diderot
5 Place Jules Janssen, 92195 Meudon, France
Tel: +33 (0) 1 45 07 78 61
============================================================================


------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, SlashDot.org! http://sdm.link/slashdot
_______________________________________________
gnudatalanguage-devel mailing list
gnudatalanguage-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/gnudatalanguage-devel