The grand function can fail on 64 bits machines. Run the script below on a 64

Reported by Michael BAUDIN

Originally assigned to Michael BAUDIN

-- Bug description --


The grand function can fail on 64 bits machines.

Run the script below on a 64 bits Linux/Ubuntu machine with Scilab 5.3.0-beta-5:
* mt, clcg2, clcg4: work
* fsultra, kiss, urand: fail

# C  [libscirandlib.so+0xbde9]  snorm_+0x179


-- Scilab error message --


michael@michael-desktop:~/Bureau/scilab-5.3.0-beta-5/bin$ #
# A fatal error has been detected by the Java Runtime Environment:
#
#  SIGSEGV (0xb) at pc=0x00007f54f41b6de9, pid=4782, tid=140003652069136
#
# JRE version: 6.0_18-b07
# Java VM: Java HotSpot(TM) 64-Bit Server VM (16.0-b13 mixed mode linux-amd64 )
# Problematic frame:
# C  [libscirandlib.so+0xbde9]  snorm_+0x179
#
# An error report file with more information is saved as:
# /home/michael/Bureau/scilab-5.3.0-beta-5/bin/hs_err_pid4782.log
#
# If you would like to submit a bug report, please visit:
#   http://java.sun.com/webapps/bugreport/crash.jsp
# The crash happened outside the Java Virtual Machine in native code.
# See problematic frame for where to report the bug.
#
Aborted


-- How to reproduce the bug --


grand ( "setgen" , "kiss" ); // FAILS
r = grand(100,100,"nor",0,1);
m = mean(r);
//
grand ( "setgen" , "urand" ); // FAILS
r = grand(100,100,"nor",0,1);
m = mean(r);
//
grand ( "setgen" , "fsultra" ); // FAILS
r = grand(100,100,"nor",0,1);
m = mean(r);
//
grand ( "setgen" , "mt" ); // OK
r = grand(100,100,"nor",0,1);
m = mean(r);
//
grand ( "setgen" , "clcg2" ); // OK
r = grand(100,100,"nor",0,1);
m = mean(r);
//
grand ( "setgen" , "clcg4" ); // OK
r = grand(100,100,"nor",0,1);
m = mean(r);

added 5.0 through 5.3.x Blocker Bugzilla Linux_x64 Numerical Math PC labels

By Michael BAUDIN

The same script runs on 32 bits Windows.

By Michael BAUDIN

It may be caused by the use of "long int" and "unsigned long int" in the sources code: sizeof(unsigned long int) = 4 bytes on 32 bits machines, 8 bytes on 64 bits machines sizeof(long int) = 4 bytes on 32 bits machines, 8 bytes on 64 bits machines

By Allan CORNET

5.2.2 win64: ___________________________________________
scilab-5.2.2

             Consortium Scilab (DIGITEO)
           Copyright (c) 1989-2010 (INRIA)
           Copyright (c) 1989-2007 (ENPC)
    ___________________________________________        

Initialisation : Chargement de l'environnement de travail

Start Apifun Load macros Load help Load demos

-->grand ( "setgen" , "kiss" ); // FAILS

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "urand" ); // FAILS

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "fsultra" ); // FAILS

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "mt" ); // OK

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "clcg2" ); // OK

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "clcg4" ); // OK

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->

-->m m =

0.0042817 

5.3 beta5 win64 windows

    ___________________________________________        
                scilab-5.3.0-beta-5

             Consortium Scilab (DIGITEO)
           Copyright (c) 1989-2010 (INRIA)
           Copyright (c) 1989-2007 (ENPC)
    ___________________________________________        

Startup execution: loading initial environment

-->win64 ans =

T

-->grand ( "setgen" , "kiss" ); // FAILS

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "urand" ); // FAILS

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "fsultra" ); // FAILS

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "mt" ); // OK

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "clcg2" ); // OK

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->//

-->grand ( "setgen" , "clcg4" ); // OK

-->r = grand(100,100,"nor",0,1);

-->m = mean(r);

-->

-->m m =

0.0042817

By Michael BAUDIN

That is wrong, in the sense that not all 64 bits machines are associated with a long int made of 8 bytes. Here are some results for a Windows 64 bits machine:

short int : 2 bytes int : 4 bytes long int : 4 bytes unsigned int : 4 bytes unsigned long int : 4 bytes long long int : 8 bytes unsigned long long int : 8 bytes

(Thanks to Allan.)

(In reply to #8560 (comment 1126160201))

It may be caused by the use of "long int" and "unsigned long int" in the sources code: sizeof(unsigned long int) = 4 bytes on 32 bits machines, 8 bytes on 64 bits machines sizeof(long int) = 4 bytes on 32 bits machines, 8 bytes on 64 bits machines

By Michael BAUDIN

Ok, got it.

Linux 32 bits (gcc) - long int = 4 bytes Linux 64 bits (gcc) - long int = 8 bytes Windows 32 bits (Visual) - long int = 4 bytes Windows 64 bits (Visual) - long int = 4 bytes

The remaining trouble is : why does the original code uses "long int". I guess that this is not necessary.

By Michael BAUDIN

The unit tests do not check the output of the uniform random number generators.

There is no example in the help page of grand. The grand function has too many completely separated features: producing the examples would create an extremely lengthy help page, which is already too long.

By Michael BAUDIN

The existing unit tests must also be redone:

• they contain interactive graphical comparisons...
• they do not check all distribution functions,
• they do not check all uniform random number generators.

By Michael BAUDIN

Commit : https://gitlab.com/scilab/legacy_codereview/-/wikis/changes/2741

By Allan CORNET

merged @​​ 1e1d49ea

Allan

closed

By Michael BAUDIN

Thank you Allan.

By Francois Granade

(originally entered in version "5.3.0 beta")